   |
|||
|
|||
|
:
News
: Archives
|
|
|
|
|
|
|||
|
|
|||
|
|
|||
|
April 4, 2006 For the past four months we have been hearing a lot about Intel's 65 nm process and the products being released on it. Unfortunately, I think that there is a lot of misunderstanding about where Intel exactly is at, as well as how their roadmap is shaping up. As tech enthusiasts we tend to be quite optimistic about how a certain product family or process node is going, because we simply want to see these new products and lines come forward as soon as possible so we can see the results and buy these new products. This optimism often leads us into making fallacious claims about the healthiness of a certain design or process node. Intel has a great marketing group, and with two Intel Developer Forums a year to spread the good word of Intel, we are often inundated with very positive information about Intel's products and production lines. This little editorial is not meant to reflect poorly on Intel's products and processes, but rather to help ground people in the fact that Intel is not perfect and they have to work hard to get these products and nodes out. Let us take a look at previous node changes at Intel. When Intel went to 130 nm, their first major part on that node was the Tualatin. This was a shrunk Pentium !!! and it was targeted at the server space. Very few of these parts were taken to the desktop, as the fairly new Pentium 4 was dominating that space. Tualatin was able to scale to 1.4 GHz fairly easily on this process node, as compared to the maximum 1 GHz that the Pentium !!! design was able to achieve at 180 nm. Intel used this well known design to really tweak their 130 nm process, and we were treated to the outstanding Northwood core several months after the release of the Tualatin. Intel was not under much pressure from AMD at this time, so they could roll these products and this new node out at their convenience. 130 nm turned out to be an excellent node for Intel and their products based on the Northwood core. Intel also had a lot of time to improve their 130 nm process, both in terms of yields and speed bins, as well as put out a few new steppings of the Northwood core. It wasn't until AMD put out the Athlon 64 at 130 nm that Intel was under any pressure. When the Athlon 64 hit, Intel saw that they had their work cut out for them. After the release of AMD's Athlon 64, Intel was setting the stage for the Prescott and their 90 nm process. This time around Intel did not feel they had the luxury of porting an older design to 90 nm rules, much as they did with Tualatin and 130 nm. They went straight to a new, large part on 90 nm, and it was not nearly as successful as the Tualatin/Northwood combination. Prescott turned out to be big, power hungry, and very hot. While it was able to run at the aggressive speeds Intel put forth, it did not do so happily. The first review chips that were received usually ran so hot that they would often discolor the PCB of the motherboard, or if those boards were set upon Styrofoam or plastic for testing, it would often melt those materials. The other problem was that Prescott did not show any significant performance advantages over the previous Northwood core. Unlike the jump from the 180 nm Willamette Pentium 4 core to the outstanding 130 nm Northwood core, Prescott ran hotter, pulled more power, and was not any faster than the previous Northwood. Many around the industry were scratching their heads and wondering why Intel did not port the Northwood design to 90 nm, thereby giving Intel a much more solid product to introduce the new 90 nm node at. Intel persevered and did improve both their 90 nm process as well as the Prescott design. It still was a hot and power hungry design, but Intel was able to improve those aspects through time. Unfortunately for Intel, the AMD Athlon 64 turned out to be their worst nightmare. While AMD was a good 8 to 12 months behind Intel in terms of process node changes, their products continued to compete well. The advances that Intel was able to apply to their 90 node and Prescott design were often undermined by AMD pushing the clockspeed on their products, so Intel had to push the clockspeed of the Prescott design further and further, thereby negating any work done to decrease the leakage, power draw, and heat production of the design at the previously available clockspeeds. Still, Intel's 90 nm node and Prescott design were flexible enough for them to put two Prescotts on one die and sell it as a dual core chip. Heat and power were still pretty atrocious, but it still worked with good results. Intel took a page from its own book when going to 65 nm. The first shipping products were based on the Prescott design ported to 65 nm. Instead of going with a new, large part on a brand new process, Intel took a more conservative route. This is starting to show some very good results for Intel, as they can more simply optimize the new process by using a well known product on the new design rules for 65 nm. We also can see that Intel has improved its process over the past six months to a great degree. The first indication that Intel made major improvements to their 65 nm process was the release of the Extreme Edition 965 processor. Here was a processor based on the same Pressler core as the previous EE 955 processor which ran at 3.4 GHz, but was able to run at 3.73 GHz yet still consume less power and produce less heat than the initial 955. All of this despite the 373 MHz clockspeed advantage the 965 has over the 955. We can see that from the introduction of Intel's 65 nm process that there was still a lot of work to do to make it a viable and economic process. Something that many do not seem to understand is that Intel is still ramping up 65 nm production, and the majority of chips still hitting the streets from Intel are 90 nm. Most of the 65 nm production has been going to the new Core Solo/Duo chips, as well as the dual core 9x0 series. Intel only expects about 20% of its shipments to be 65 nm by this summer, and they will continue to ramp so that it reaches 80% of shipments by late Spring of 2007. If we look at the numbers, this means that throughout this year, the vast majority of chips shipped by Intel will still be 90 nm Prescott based products. The Cedar Mill/Pressler products will slowly displace those products, as well as the new Conroe family of parts. While Intel will be introducing the Conroe family of products this summer, we really can't expect this chip to become the majority of shipments out of Intel until Winter of 2007 (January, February, and March are considered Winter). Yes, there will be product available shortly after launch, but it will be a while before Conroe based shipments will overcome the Prescott/Cedar Mills/Pressler based products. By looking at the recent 65 nm product releases, we can see that Intel is now comfortable with the health of the process, and they will start to really ramp it from here on out. This ramp will still take time, and a major architectural shift will slow things down as well. Unless Intel moves far faster than they have intimated at their recent IDF's, we will not really see a large proportion of products being Conroe based until at least December of this year. Intel has been very bullish on how their 45 nm process is going, and many around the industry have taken this optimism and run with it. There are those that have proclaimed that Intel will have 45 nm products early in 2007, but that is pushing things (by a great degree). When Intel is only expecting about 80% of their product shipments to be based on 65 nm by Spring of 2007, there is no way that they will have 45 nm ready for manufacturing. Intel may hit 2007 for initial 45 nm parts, but it will be very late in 2007. I feel that the introduction of 45 nm parts will likely mirror what we have seen this past year with the transition from 90 nm to 65 nm. I have a feeling that in December of 2007 we will see Intel introduce in small numbers 45 nm Conroe based products to the market, and we will see Intel tweak the 45 nm process for the next 6 to 8 months. After that they will aggressively ramp 45 nm and whatever new cores they will have developed. So, it will be Summer of 2008 before we see large amounts of 45 nm parts from Intel. This is a far cry from those posting that Intel will have 45 nm parts in Summer of 2007 and will simply bury AMD because they will only then have the majority of their Athlon 64 parts at 65 nm at that time. Intel is a very aggressive company when it comes to process technology, and they are the leader in the industry when it comes to developing and migrating to a new process node. Intel is not a superhero though, and they can't jump nodes in a single bound. What they can do is stay a good 8 to 12 months ahead of the competition when it comes to introducing parts based on a next generation node though. But by the time AMD is fully ramped at 65 nm, AMD will have a good year before they have to sweat out Intel pushing 45 nm parts out the door in volume. As it is, AMD still has a very large window of opportunity to continue to take marketshare away from Intel. With AMD's very forward looking architecture (memory controller, HyperTransport, glueless multiprocessing, etc.), they still have a very solid infrastructure and compelling performance to continue competing with Intel even though AMD does not have the process advantages that its larger competitor does. Intel is doing some great things with process tech, but they still have to follow the laws of physics and there are still many hurdles out there to cross before they have production ready 45 nm parts on the market. There is a discussion about this that can be found here. Speaking of AMD, they made an announcement today that they have begun revenue shipments from Fab 36, which is AMD's 300 mm wafer fab based on 90 nm. At Fab 36 they have also been running 65 nm development test chips, and AMD claims to still be on track to release 65 nm products by the end of this year, and have the majority of their product lines switched to 65 nm by the end of Spring 2007. AMD's marketshare gains may slow once Conroe and its family is released, but they still have the opportunity to continue their drive to gain more customers throughout this next year. March 30, 2006 Memory, AMD, and other Stuff - Josh MS over at Lost Circuits has an excellent interview with Jeff Lewis of ISi, which is the company selling IP for Z-RAM (a much more dense and possibly efficient memory type for use in many applications). AMD made a big splash about possibly using this technology to increase the size of their L2 caches by up to 5X without taking up more die space. The interview gets pretty complex, so it is not for the weak of heart. Otherwise, some really good information about why this works, and why the CPU guys using SOI are really gung-ho about it. If AMD can get this design into play soon, we can expect to see L2 caches of up to 4 MB per core using the same die space as the standard 1 MB S-RAM. Some time back I noticed that AMD was attending quite a few job fairs in the Colorado/Wyoming area, and the reason is that it looks like AMD is pulling the trigger on putting a new design center in Ft. Collins, CO. Ft. Collins is only about 64 miles away from me, and it appears as though some high profile Intel Itanium engineers are heading over to AMD, so AMD is building a design center around this group of individuals. Knowing this area, Ft. Collins is a nice place to live and work. It is not too far from Denver, yet still far enough way that Ft. Collins still has its own personality and lifestyle (and a pleasant one at that). Originally AMD was quite mum on the situation, as they were not sure that it was going to actually happen. Now with all of these leaks and announcements, it seems a sure thing that AMD will have a pretty hefty design center in the Rocky Mountain region. No word on what these folks will be working on, but needless to say we won't see their first product for a couple of years. Charlie from the Inq is saying that AMD will have a 65 nm product announcement next week. I am not in the loop, so I have no idea what is going on. This shouldn't be a product release, as AMD's 65 nm process is not anywhere close to producing production quality silicon (still very much a work in progress). I would expect the first 65 nm parts to hit around December at the earliest, but apparently AMD does have some news they feel they need to share. There are hints that it actually deals with AMD using Chartered Semi as a second source for 65 nm parts. It appears as though supply of 7900 GT and GTX parts are pretty tight right now, and these products are in high demand. NVIDIA and their partners are shipping these as soon as they can, but then again we are still only three weeks out of the initial launch. As you may or may not know, it is standard practice for NVIDIA to create the first batch of boards for their partners, and the partners buy them complete from NVIDIA. Once the first few batches are out of the way, this has left enough time for the partners to implement their own iterations of the boards (each manufacturer has a different mix of components that they specify for their product, such as brands of capacitors and other SMT components, to meet price goals). For this first batch, the partners usually just slap a sticker on the board and package it in their own box. Afterwards companies like BFG will specify their own board design with a custom cooler, but for better or worse the high end cooler by NVIDIA is damn good at what it does, so most companies are not willing to spend their own money working out a unique design when the stock one does so well. Apparently a new, large batch of 7900 parts have hit the US, and we should see them listed online very shortly. While Newegg may be out of stock at the moment, there are still plenty of other online dealers who have cards in stock. ATI announced their 2Q 2006 earnings before the market opened today, and while they beat analyst estimates, it was not as strong of a quarter as last year at this time. Still, beating the analysts is good, and the stock price has taken a sharp rise off of the news. Things are not all grim at ATI, and they are certainly moving product, but gross margins have taken a beating due to ATI shipping a lot of integrated chipsets with margins apparently around 5%. They also attribute this to a drop in discrete desktop chip sales, and considering ATI's last few quarters with the R520 at the helm, it is very understandable. Especially in the face of NVIDIA's GeForce 6 and 7 series of cards across the board. Things should be improving for ATI, and they expect more sales this next quarter. Their latest R580, RV515, and RV530 all are doing brisk business, but there are some manufacturing issues with the RV530 and R580. These issues are not technical in nature, but rather have to do with the current competitive landscape- more on this situation later. Still, it was a strong quarter for ATI considering the market pressures brought against them. March 23, 2006 News Around the Industry - Josh As you have all read by now, Dell has acquired Alienware. Dell has been trying to break into the enthusiast market with their XPS series of desktops and laptops, and while the laptops have done very well the desktops have not seen the movement they were expecting. By buying Alienware, Dell now has a solid foothold in that lucrative market. What is interesting is that the Alienware website is already converted to the Dell layout. The next question this brings up is if Dell will continue to support AMD processors with the Alienware brand. In the near future it will be a sure thing, but once Conroe comes out I seriously wonder if that AMD support will erode over time. I guess the deciding factor will be AMD's ability to compete with Conroe with their upcoming and unannounced parts. It will also be interesting to see how Dell will incorporate the support system that Alienware already has in place. Will it stay as such, or be engulfed by Dell's current system? I am betting the bean counters are already preparing the pink slips for these folks, and transferring the numbers to the current support system located in both North America and India. Dell will have to work hard to keep Alienware a viable and independent name in the industry, and time will tell if this will be a good acquisition, or if Alienware's marketshare will fade by being associated with the giant Dell. PC Perspective is at the GDC this week, and they had the chance to quiz ATI about their relationship with Havok and the upcoming Havok FX engine. It appears as though the latest X1x00 series of GPU's from ATI will in fact support acceleration of Havok FX, much like NVIDIA's latest GPU's were announced to do earlier this week. ATI of course said that their solution would be faster because they have 48 pixel shader pipelines to throw at the situation vs. NVIDIA's 24 pixel shader pipelines. We won't really know how well each will perform until at least August, but until then the gamesmanship will continue between the two graphics giants. Still, it is good that Havok FX will also support ATI GPU's. The kind folks at PC Stats have put up an Asus X1900XTX review. Asus packs their cards with a generous bundle, and I have rarely heard any issues with Asus quality as of late. If you are looking to purchase this nice little number, go take a look at what they have to say. Blue Gears is still alive and kicking, and they have put up a page concerning their upcoming b-Oxygen 8788 sound card. This is based on the C-Media Oxygen 8788 chip, which hosts a load of Dolby Digital and DTS encoding features. This is a big step up from their previous HDA/Auzentech based product, the X-Mystique. It still isn't apparent why BlueGears and HDA/Auzentech ended their relationship, but the buzz basically says that HDA wanted a direct presence in the North American market and made things difficult for BlueGears once that foothold was achieved. BlueGears did a good job in getting their name out there, and the X-Mystique sold very well as a direct competitor to Creative's products. So, I for one will be very interested in how this sound card will perform and sell. The board they have shown has both a optical input and output, as well as analog mic and line in, and 7.1 channel analog outputs. March 22, 2006 Wednesday Happenings - Josh Today NVIDIA is releasing two new products on the GPU scene. The first is the 7600 GS which is aimed at the sub $149 market. This is based on the same GPU as the 7600 GT, but this one is clocked to 400 MHz core and utilizes 400 MHz DDR-2 (800 MHz effective). It apparently beats out the X1600 XT, but is priced more around the X1600 Pro. What is really nice about this product is that with a core clock of 400 MHz, and the use of TSMC's 90 nm Low-K process, the G73 chip this is based on does not get terribly warm. Most AIB's will probably use a passive cooler with this product, and by doing so they are aiming this product squarely at the HTPC crowd. The 7600 GS has full PureVideo support, and the recent driver updates now support features like inverse telecine in high definition content. This will probably drop the prices on the current 6600 DDR-2 products by a few more dollars, and it does make for a compelling product at that price range. The second product released today from NVIDIA is the 7300 LE. This is not exactly a new product, as it has been an OEM offering for some time. The 7300 LE uses the same GPU as the recently reviewed 7300 GS, but it is clocked down to 450 MHz vs. the 550 MHz of the GS. It also features slower DDR-2 memory, but AIB's will get to decide how much it will get. This product is aimed at the sub $79 level, and still represents a significant performance increase over integrated graphics. This will probably totally displace the 6200 TC series in the next few months. These two budget oriented products from NVIDIA have helped to flesh out the lower end GeForce 7 series of cards, and they certainly make for some solid gaming and video playback experiences at each price point. NVIDIA has been very aggressive as of late in competing with ATI's products across the board, and so far it appears as if they have done well. Needless to say, the choice of products across every price point is quite good from both manufacturers, and the competition is always exciting to watch. Ageia Officially Launches PhysX PPU Ageia and BFG Tech have announced today that they will begin offering the PhysX add-in PPU in select Alienware and Falcon Northwest computers within the next week, and a general retail release by the end of May. This will be the first dedicated Physics Processing Unit, and will come with 128 MB of dedicated GDDR-3 memory. While support for the card is currently non-existant, there are more than a handful of publishers, developers, and game engines which will utilize the PhysX PPU in the near future. While Havok may have more support at this time, Ageia is working hard to create and support a highly advanced physics framework to base games around. So far the games and applications which have been previewed really show the power and effectiveness of the PPU. More and varied physics effects inside games make for a more immersive experience. By the end of this summer there should be a handful of games supporting the standalone PPU, but the big push will be when Unreal Engine 3 titles will begin shipping. Epic has shown very strong support of the PPU from Ageia, and titles using this engine will benefit from hardware accelerated physics. Hot Hardware has a very nice preview of the technology, as well as coverage from the launch event that is happening at the GDC today. They cover pretty much all aspects of the technology, as well as mention the support for the PhysX engine across several platforms, notably PowerPC (X-Box 360), Cell (PS3), the PPU, and finally software support for Intel and AMD processors. One thing to note is that while there are plans to support PCI-Express, the first generation of products will all be PCI based. Intel Launches New 965 Extreme Edition CPU Hot Hardware and The Tech Report are two of the first sites to preview this new chip from Intel. Apparently Intel has been actively tweaking their 65 nm process and have finally been able to get the power consumption and heat production down on their latest Pentium 4 based parts. While the 965 is clocked at 3.73 GHz, it actually consumes less power than its older sibling, the 955 EE. This is a pretty healthy accomplishment for Intel, as it does make the latest Pentium 4 parts a lot more competitive with AMD in terms of performance and power. Its amazing what can happen in three months, and it makes you wonder if Intel really was correct in pushing its 65 nm releases. The amount of improvement we have now seen from Intel's 65 nm process is pretty phenomenal, as the overall design for these processors have not undergone any significant change. March 21, 2006 NVIDIA Announces GPU Physics - Josh Today's GPU's are massively parallel processors in their own right, though with more of a bent towards pixel processing. With the advent of highly programmable pixel and vertex shaders, it is now possible to utilize the power available in GPU's in other applications. Previously we have heard quite a bit about the GP-GPU movement (General Purpose GPU), but few realworld applications have ever been shown to work with modern day GPU's. Working with physics engine designer Havok, NVIDIA is announcing the upcoming Havok FX engine, which will be able to leverage the power of NVIDIA GPUs in physics simulations and applications. Havok's game engine has been used in hundreds of games so far, with notables such as Half Life 2, Painkiller, Age of Empires III, and many more. The new Havok FX engine is expected to be available by the end of summer, and there are already quite a few clients lined up to use it. NVIDIA says that Havok FX will be able to be accelerated by GeForce 6 and 7 series of cards, but the recommended minimum will be a 7600. This means that even a single video card can help accelerate physics simulations. Where the real power shows up is with SLI systems with the latest NVIDIA GPU's. Users will have the option of utilizing one of the graphics cards in a SLI setup to primarily do physics, while the other is dedicated to rendering. In a custom built demo, NVIDIA has said they have seen a 10X performance increase in using a dedicated GPU for physics processing, but because Havok FX is not officially released yet, there is no 3rd party verification. Still, I doubt that NVIDIA is wildly exaggerating, as a programmable floating point processor is a logical solution to certain types of physics interaction. It appears as though the GPU portion will primarily focus on interaction between objects, while the CPU will handle other aspects of physics. While the GPU will not accelerate 100% of all physics processes, it will offload the highly parallel stuff which it can address best. In more CPU bound titles this will prove beneficial, but if the title is more fillrate or shader bound, then the GPU will be able to provide little to no acceleration of physics. Physics and graphics calculations can be handled by the same GPU at once without appreciable overhead for any kind of state change in the GPU. The Havok FX API allows the GPU to store physics data in its local memory, and also allows the GPU to do the majority of its job without having to send much data back to the CPU. Currently this application is only planned to work with GPU's and not a standalone physics card (like the AGEIA Physics Processor Unit- PPU). In Quad SLI situations, the performance enhancements could be quite dramatic. While NVIDIA has not detailed how exactly such a situation would be handled, it is logical that the easiest solution would be for two of the cards to render graphics while the other two handle the physics. Depending on how flexible Quad SLI is, we could see other combinations. I currently have that question out to NVIDIA, and once I receive the answer I will update here. Since Havok and AGEIA are competitors, neither will allow their physics engines to work with the others' products. This means that there are no plans at the moment for NVIDIA to support physics acceleration on GPUs with games using the Novodex or PhysX physics engines. This also means that Havok FX will not run on the AGEIA PhysX PPU. There do appear to be plenty of games using one or the other physics engine, so it appears that high end machines will be able to utilize both SLI physics acceleration and the standalone PhysX PPU card. This is a nice little freebie from NVIDIA and Havok for people who already own an NVIDIA GeForce 7 series card. It is a big boost for those already owning SLI systems, as they essentially have a PPU already present in their system. There is no reason why Havok's engine cannot be ported for use by ATI SM 3.0 hardware for acceleration, but there is no announcement as of yet for that type of support. Due to possible anti-competitive laws, we may in fact see the Havok FX engine support ATI's X1x00 GPU's in the near future, but the exact details and nature of NVIDIA and Havoks' relationship is not known at this time. Abit Rises Again It seems that USI has finished in its acquisition of the Abit brand. This means that the Abit name will live on, and the current lineup of Abit products will continue to be produced. USI is a large company with many diverse assets, and this acquisition will mean that more and varied products will be released under the widely recognized Abit name. Abit will still be active in the production of motherboards, but they will expand into new areas such as personal audio. I will be very curious to see what the upcoming generations of Abit products bring to the table, as it is a strong brand with some good engineering talent. Within the next 4 months we will see what new products are brought to the table, and if the legendary engineering talent of Abit has survived intact. March 14, 2006 AMD, Intel, and Undermining Current Quarter Sales - Josh It seems that Intel's little Conroe demonstration caused a lot of waves in the computer industry. It isn't so much that it appears significantly faster than the AMD Athlon 64 X2 (that seems more of a concern for enthusiasts than most OEM's), but rather that Intel is letting the cat out of the bag so prematurely that it could have a dramatic effect on computer sales for the next two quarters. Not only has Intel sabotaged their own Pentium 4 sales, but they have also negatively impacted the companies selling the final product in full blown computer setups. Showing off the Conroe may have put off anywhere from 10% to 20% of potential buyers for the current selling season (with Q1 and Q2 as typically slow as they are), any drop like this is going to have big repercussions on the bottom line of a lot of companies. Considering that Intel thinks the Conroe and its brethren will be seeing the market in force about six months from now, that could be a lot of potential buyers delaying their purchase of new computer systems. The truth be told, it probably has a more detrimental effect on Intel than it does on AMD. AMD continues to sell out their entire production coming from Fab 30, and Fab 36 is apparently in the final stages of validation and AMD expects to start seeing end product ramping up from here on out before the end of April. Though Fab 36 will be running 90 nm parts, it will be on 300 mm wafers. So, AMD has the ability to produce many more Athlon 64's here by June than they ever have before. Still, Intel has made the Pentium 4 look very unappealing to many buyers, and while we knew that the Conroe would be a big step above the current Intel lineup, we didn't realize that it is looking so good. IT administrators, enthusiasts, and savvy PC shoppers are all considering holding back their budgets to buy up the new Conroe based systems when they become available. It could seriously be as much as 30% of the buying public/corporations that were planning an upgrade for this quarter will wait a few more months to delay that purchase for Conroe. So everyone producing some kind of computer component or another are not really happy with Intel right now. What Intel did would be akin to NVIDIA just releasing the 7900 GTX card, but then coming out and giving a couple of sites an fully working G80 (next gen NVIDIA DX10 part) and allowing them to publish benchmarks and say "this product is coming out within 6 months". How many users thinking of dropping $500 to $600 on a new video card would rethink their position and tell themselves, "You know, I can wait a few more months for that card"? While it will not be everyone, a significant portion of buyers would in fact take that position. Intel should know that its actions have really only hurt themselves and their partners, and the only detrimental effect it has had on AMD is their share price. Speaking of share price and AMD, several recent downgrades have had a very negative effect on AMD's stocks. The underlying cause for this is that most analysts see an upcoming price war between AMD and Intel stemming from Intel's very solid looking Conroe. While I think the Rev. F chips from AMD will be overall faster than the current products, they will no longer have the performance advantage over the competing Intel parts. This will lead to price erosion for AMD across the board. Needless to say, the playing field just became more level with Intel having a very competitive product. The next two quarters for AMD though will still be phenomenal, as they are selling all of their production of CPU's, plus are introducing some new notebook and Opteron models throughout this next quarter. AMD will still gain marketshare in these two areas, but it looks as though the big gains in the desktop market is going to come to a rather abrupt stop once Conroe hits the streets. Still, AMD is not resting on their laurels, and they have been hard at work on future products. Lately they have been very quiet about what is coming up. Come June with the release of the AM-2 products and Computex, AMD will supposedly open up a lot more and let out hints about their upcoming products (like K8L, which is rumored to have some really beefed up SSE and FPU units, plus some other goodies). They will also spill some of the beans on their 65 nm plans, and what kind of enhancements we will be seeing in those products. The pressure is on AMD right now, but they have survived worse situations than what we are seeing now. NVIDIA Post Release Roundup It has been several days now since NVIDIA released their 7900 and 7600 parts, and for the most part the response has been very positive. While there were many that were disappointed by the lack of new features in these new parts, the increase in performance was well noted. The 7900 GTX is about as fast as the X1900 XT and XTX products, winning some and losing some depending on the application. Where the real surprise showed up was with the 7600 GT. Nobody expected a 128 bit memory part to beat up on the older 6800 GS/GT parts which were 256 bit. The 7600 brings a lot to the table for not a whole lot of scratch. ATI responded with the bigger and much more complex X1800 GTO, which is a cut down R520 chip with a 256 bit memory bus. The 7600 GT matched fairly well against this giant, but typically comes in about $50 cheaper. The one thing that NVIDIA and its partners have really stressed with this launch is that margins on finished products are key. The shrink of the G7x series of chips to 90 nm, as well as their redesign to pare off transistors, has lead to chips that are significantly cheaper than what ATI can offer, yet still provide a level of performance equal to that of the competition. For the first time in the graphics industry, a company has taken their products and actually kept performance yet dropped transistor counts dramatically. The 7900 GTX is a fast card, but dropping the introductory price to $499 is a great step in the right direction for consumers. As we all remember, the 7800 GTX retailed for $599 at introduction, and the X1900 XTX came in at $649. Oh, and let's not forget the 7800 GTX 512 which MSRP'ed at $649 yet consistently sold at $749 and above. I think this is the right move for the graphics industry, as they will see a lot more movement of products at a $499 price point then they will at a $649 and above. This is not to say that there will no longer be enthusiast only cards available at these higher price points, but the introductory price to a high end card has now gotten back to a sweeter spot than what we have been seeing previously. Die size and performance per millimeter squared is the mantra at NVIDIA these days, but like all things this could change. It is good to have that in the back of the mind though when designing a large GPU. The upcoming G80 series of parts will mark NVIDIA's entry into the DX10 market, and we have absolutely no idea what size these will come in. The increased complexity of a unified part may cause NVIDIA to step away from the "performance per mm" paradigm, but that is still around 5 to 7 months away before we see the fruits of that particular labor. Still, NVIDIA has produced a series of parts that run fast, keep fairly cool, and are a lot easier to produce than the competing products at their respective price ranges. ATI is not standing still as well. The company has a lot of experience with unified shader cores (Xbox 360), and their R6x0 series of products will build upon that expertise. ATI knows that they have probably overbuilt the R5x0 series of chips, and the transistor usage was a bit out of control. They are still nice parts, but the overall cost of each board built and shipped is probably 20% to 30% greater than that of NVIDIA's current lineup. And the majority of that cost comes from the raw die prices. So, this coming Fall we will see some interesting products from both manufacturers, and we will see who has the upper hand when it comes to the DX10 generation of parts. The competition we are seeing here is great, and it really is a wonderful time to be an enthusiast with so many outstanding choices in the graphics field. March 9, 2006 NVIDIA Releases the 7900 GTX, 7900 GT, and 7600 GT - Josh NVIDIA has finally fleshed out its entire lineup of cards with 90 nm parts. The first discrete 90 nm part from NVIDIA was the 7300 GS launched last month, and today NVIDIA is completing the lineup with three brand new cards. The 7900 GT and GTX both share the same GPU, but the GTX is clocked at 650 MHz core with GDDR-3 memory running at 800 MHz (1600 MHz effective). The GTX features the same dual slot cooler as the previous 7800 GTX 512. The 7900 GT runs at 450 MHz core with 256 MB of GDDR-3 memory running at 660 MHz (1320 MHz effective). Unofficially this is the G71 core, which features 16 ROPS, 24 pixel shader pipelines, 24 texture units, and 8 vertex shaders. Functionally this chip is identical to the older G70 which powers the 7800 GTX series of cards. The G71 is made on TSMC's 90 nm Low-K process using 300 mm wafers. While the G71 is functionally the same as the G70, NVIDIA did a lot of engineering work to increase the speed yet decrease the transistor count and die size. Instead of having 302 million transistors as the G70 does, the G71 has cut that number down to 278 million transistors. This savings of 20 million+ transistors allows NVIDIA to have a die size close to 1/2 that of ATI's R580. Performance per clock is identical to the G70 though, so the increase in clock speed allows the 7900 GTX to outperform the ATI X1900 XTX in a large majority of applications. While NVIDIA did not expand upon the featureset of the GeForce 7 series, the die size and transistor count savings allows NVIDIA to lower the introductory price of these products. The GeForce 7900 GTX features 512 MB of GDDR-3, and has a MSRP of $499 for the base product. Many manufacturers will introduce specialty SKU's that may cost more, but the starting price for the basic card is quite a bit lower than previous high end cards. Again, the savings that NVIDIA makes by decreasing the die size and transistor counts on their chips allows them to place more chips per wafer than the competition. Good dies per wafer are also increased because the dies are individually smaller. The ability to re-use PCB's and cooling devices from the previous generation of cards also lowers costs. The 7900 GTX has similar heat and power properties to the older 7800 GTX 512. The GeForce 7900 GT is slightly faster than the older 7800 GTX, but its introductory price is a stunning $299. This replaces the 7800 GT, and there is the option of seeing a 512 MB version from AIB's in the months to come for a slightly higher price. This card also features power and heat characteristics that are lower than the standard 7800 GTX, and the heatsink and fan on the 7900 GT is the same as on the 7600 GT. AIB's will customize this SKU, but with the basic price being $299, this will find a very open market for itself. The final introduction is the very small, yet powerful GeForce 7600 GT. This product replaces the 6800 series of cards priced from $179 to $249. The 7600 GT is a very apt performer, and it features an outstanding price/performance ratio. I reviewed the card here. While this release was not what many were thinking it would be (32 pixel pipelines, improved ROPS, improved texture filtering), NVIDIA has changed the rules of the game by focusing on speed and doing more with fewer transistors. As we see transistor counts skyrocket in other high end parts, NVIDIA has taken a different road and focused on efficiency of performance and production. Power consumption and heat production have been kept consistent from the previous generation to this one, and at the lower end of the spectrum, heat and power are not issues at all. This is a hard launch, and there is product available from a wide assortment of online and brick and mortar retailers. Products from the 7900 GTX down to the 7300 GS will be available, and it appears as though there is a lot of stock already out there. While ATI impressed us all with their X1900 series of cards, NVIDIA is doing the same but in a totally different manner.
NVIDIA passed these pictures on so we could see that these major e-tailers all have stock as of yesterday. Apparently the TigerDirect.com guy rarely wears a tie, so this is a big occasion. eVGA GeForce 7600 GT CO Superclocked Review NVIDIA and eVGA were kind enough to provide me with a sample to review for this introduction. I believe that last introductory video card I reviewed here was a GeForce 4 Ti 4600... which was quite some time ago! eVGA has produced one of the fastest "stock" 7600 GT cards on the market, and its performance is quite surprising! This is no simple replacement for the 6600 series of cards, in fact it is faster than the 6800 GS and 6800 GT. I was quite amazed by how well this product performed. Here is a quote from the review:
You can read the entire review (not preview) right here. March 8, 2006 As I had mentioned earlier today, Anand was able to spend an hour with two machines set up by Intel to see how the Conroe at 2.66 GHz matches up to an overclocked AMD FX-60 running at 2.8 GHz. In the majority of tests shown, the Conroe simply walks away from the AMD part. The Conroe in fact was upwards of 40% faster than the already fast FX-60. This of course brings up many, many questions about the viability of AMD in the very near future, as well as what exactly Intel is up to. The first thing we need to do is sit down and realize that Intel set up these two machines. While Anand couldn't find anything wrong with them, there is the possibility of some tricks going on to make the Conroe look like a huge favorite. For example, Apple is well known for using tricks and specific tests that made its previous products look like the second coming as compared to the products in the PC market. Compiler optimizations, low level driver settings, and other little tricks could work together to give the Conroe an unfair head start. I am not saying that Intel is out and out cheating here, but the guys setting up these machines do not want their products to look bad, and I can see them "optimizing" the Conroe machine to give better results. So, until a 3rd party can get their hands on a Conroe and run some independent tests, we should take these results with a pretty big grain of salt. There could be a lot going on that we don't know about, and until we can get some 3rd party verification, the Conroe should just be considered an exciting but unreleased product. Something that did strike me as very interesting is that Anand mentioned that this would be a part that is six months from being released. I find this quite odd as I had thought Intel was pushing for an early July release date, which is approximately four months away. There is something strange about the timeline Anand is pointing to, as it just doesn't jive with what most of us have heard. A 2.66 GHz part should be released during the introduction in July, but if it is going to be coming out in a September time frame, what exactly is going to be released in July? Currently Intel is hemorrhaging marketshare all over the place, and if a 2.66 GHz Conroe in its current state is slapping around a FX-60 clocked at 2.8 GHz, and doing it with lower power and lower heat production, why is this chip not ready for prime time? Why is Intel not releasing this product ASAP, especially in the face of rumors of Dell going to AMD products, as well as AMD's marketshare gains in multiple areas? Intel certainly has the fab capacity to ramp this design very quickly, and their 65 nm process is very robust. So what is the holdup? Unless Intel is stockpiling these products for a massive release, in a business sense it makes no sense to sit on such a product if it really is this ready and is beating the competition so badly. I do believe that the Conroe will be a very fast part, and it is refreshing to see Intel finally step away from the failed Netburst architecture and embrace something new. They look to have solved their power and heat issues that plagued the Netburst processors with Conroe and its brethren, and OEM's around the world are overjoyed about it. Intel is a big company, and they have a lot of very talented and smart engineers there. There was little doubt in anyone's mind that they could produce a good, fast processor. It seems that they have done so, but we still don't know how well it will truly fare against AMD and their upcoming Rev. F offerings. So, let us assume that the results of Anand's tests are truly valid. This means that the current Athlon 64 parts will not be able to compete with Intel in any way, shape, or form. AMD is apparently not terribly worried about this, as DDR is starting to dry up and they are transitioning to the new Rev. F DDR-2 based processors. Current tests and reports have all shown that the DDR-2 based Athlon 64's are not any faster than current Athlon 64's, and in fact the new ones can be a bit pokier than the old. There are many out there panicking that these Rev. F chips will just not pass muster and AMD will again be a has-been in the CPU industry. Personally I believe the proclamations of AMD's death is premature. AMD has a history of "leaking" out parts to motherboard manufacturers for validation that are not fully functional or show the true potential of the part. I distinctly remember when the original Athlon was two months from release and Firing Squad got a hold of one of these chips and benchmarked it. The Athlon sample they had was a 550 MHz part and it was actually beat up really badly by a 500 MHz Pentium 3, not to mention it fell behind the K6-III 450 overclocked to 550 MHz. When the real Athlon 550's hit the market, they were without a doubt the fastest things out there. Bar none. AMD has done this several times since then, and some of the first K8's sent to motherboard guys were partially disabled as well. I believe that the Rev. F's will show up and be a lot beefier than what we have seen so far. There are those out there quite vehement that Rev. F will surprise people with its performance. I am a bit more conservative here in that I think Rev. F. will be faster than the current Athlon 64's, but if Conroe performs in the real world like it did at IDF, then these new AMD chips will be taking a back seat. Something else to consider is that AMD now has a lot more fab space open for production since Fab 36 went online, and they have the ability to massively increase the L2 cache on the Athlon 64 even on their 90 nm process. Bloating out the die size is an option for AMD at this point, and an Athlon 64 could certainly show an improvement in performance with 2 MB of L2. Apparently Fab 36 is producing 90 nm parts at a rapid pace as we speak, and it will be producing even more come this summer. If AMD wants to shore up any performance deficiency, they could certainly turn to more L2 while not negatively affecting their ability to provide parts to customers. We will see our first real look at the Rev. F Athlon 64's this June at Computex, and then a month after that Intel is planning on releasing some of its next generation parts. Conroe will definitely be a contender for the performance crown, and it is a huge leap forward for Intel from the current Pentium 4 based chips. My little editorial here was not meant to give the impression that Intel is lying and Conroe will be a failure. Far from it, I believe Conroe will be an excellent part, one which we have been expecting from Intel for years since they abandoned the Pentium !!! architecture. My only thought here is caution. Both AMD and Intel have great engineers working for them, and in the end I believe that both will produce very good parts that will compete well with each other. One might get a jump over the other at one time or another, but unlike the last few years, the offerings from each will be much more competitive with the other. Some Quick Links - Josh Since I have been rather remiss on linking other articles as of late, I thought I would catch up a tad. Lost Circuits takes a look at the Sapphire X1600 Pro AGP. This could very well be ATI's final AGP card, and MS does a very thorough job in seeing what it can do. While it is a solid offering, it doesn't seem to be able to touch the older 6800 series, as well as NVIDIA's new 7800 GS AGP. Still a very interesting read though. Hot Hardware reviews the ATI All-In-Wonder 2006, which is based on the X1300 core. While it doesn't have the pixel pushing power of the X1900 version, it is a complete video package for those who need it. Finally we have the article that is stirring up the hornet's next. Anandtech got a chance to benchmark a 2.66 GHz Conroe vs. a similarly spec'd out AMD FX-60 clocked at 2.8 GHz. The Conroe smacks the AMD part all over the place in this one. We must keep this in perspective though, as Intel set up these machines themselves, and we don't know how they did it. I will be commenting on this a bit more later. March 7, 2006 eVGA GeForce 7300 GS Review - Josh Recently I had the chance to review the eVGA GeForce 7300 GS, and I was quite surprised by how competent the entire package is. While it is not an enthusiast type product by any stretch of the imagination, it is a very solid little videocard that can run through a wide range of 3D applications with gusto. Here is a quote:
You can read the entire review here. March 6, 2006 A Busy Week and Interesting Things - Josh If all things go according to plan, this is going to be a very busy week here. First off I will have a X1300 review by the end of today, during the middle of the week I will follow this up with a 7300 GS review, and at the end of the week I will have a nice surprise that I can't talk about at this time. Testing is complete on both the X1300 and 7300 GS, and now I have the sometimes odious task of putting these results into a readable review! The past few weeks I have been increasingly focused on outside projects, so I haven't had time to even update the site. During that time we have seen ATI release a very impressive chipset for the AMD market (Xpress 3200) which has the northbridge supporting 2 x 16X PEG slots, as well as other lanes to handle other applications (I believe the chip has 40 PCI-E lanes total on the northbridge, and several more on the southbridge when using the ULi M1575). Most of the usual suspects have reviews of the Asus board supporting this combination (A8R32-MVP), and there are very few that have come away unimpressed. ATI has shot a big volley across NVIDIA's bow with this chipset, and as a consumer I am quite happy by what ATI has done. Now all ATI has to do is get the SB600 southbridge out and working correctly, then they will have few detractors about the chipset quality of ATI parts. Still, the ability to use the ULi southbridge gives this solution a lot of muscle, as the M1575 is a robust part and feature rich. Things have been quiet on the CPU front since the release of the FX-60 and EE 955 parts from AMD and Intel respectively. AMD is busy readying the underlying support architecture for the AM-2 socket products to be introduced this June. Apparently the NVIDIA nForce 5 series of chips are complete and appeared to be ready for the initial March release of these parts, but it appears as though NVIDIA now has a few more months to polish these products. I believe the Xpress 3200 also supports the upcoming AM-2 socket architecture, so there should be some very nice initial motherboard products out in time for the release. Currently I am doing all of my testing with a AMD Athlon 64 X2 3800+ clocked at 2.4 GHz (a very conservative overclock by most standards for this chip), and I couldn't be happier with the snappiness and performance. I was so happy by the sample that AMD sent, that I bought one for my work machine and have never looked back since. I cannot say enough good things about this processor. It is very fast, it is very stable, it runs very cool (even when overclocked), and it is just an awesome piece of technology. AMD has certainly been hitting on all cylinders for the past two years, and even at stock speeds the X2 3800+ is a solid performer in both single and multi-threaded applications. While the price is still a bit high to be considered a budget part, its price/performance is not in question. It really does smack around the other Intel parts in its price range, and it is certainly a lot cooler running! So, hold on tight for what should be a crazy week around here.
If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2006 PenStar Systems, LLC. |
|||
