   |
|||
|
|||
|
:
News
: Archives
|
|
|
|
|
|
|||
|
|
|||
|
|
|||
|
April 27, 2006 Scraps around the Internet - Josh When NVIDIA first announced their Quad SLI product some months ago, the cards were amazingly long, and they probably would not fit in every case around. Well, it seems after a couple of PCB spins later, NVIDIA has actually decreased the size of the upcoming GeForce 7950 GX2 so that it is actually slightly shorter than the current 7900 GTX. HKEPC has the goods on this product, and apparently we are supposed to see this by late May. NVIDIA is continuing to really push the multi-GPU arena with products, but due to the PCI-Express divider chip on the board (which supplies each chip with a full 16X PCI-E connection as well as communicating with the motherboard at 16X PCI-E) the card will appear as one chip, and not as SLI. This could lead to some interesting performance claims, as well as a few nice features (rumor is single card 32X AA). The board uses the G71 chips, but each is clocked at 500 MHz vs. the 7900 GTX's 650 MHz for obvious heat and power reasons. The GDDR-3 memory is also clocked at 625 MHz (1250 MHz effective), but between the two boards that is a lot of bandwidth. No price has been quoted for this part, but I would imagine that it should be approaching $1,000. NVIDIA and its partners have been quietly releasing Quad SLI enabled systems in the market since early this month, but they have been fairly rare and expensive for the average user. Speaking of video cards, Biostar decided to throw their hat into the ring with their own products. Previously they had only concentrated on motherboards, but now they are starting to release graphics cards based on NVIDIA chips. They have a complete lineup, and we can expect them to customize their products to get some retail and OEM attention. I am still not sure what markets they will be trying to break into, but we can bet the initial target is Asia, and possibly moving onto North America. They have a full range of PCI-E based products, as well as a few AGP products for upgrade. Their heaviest AGP hitter is a 512 MB enabled GeForce 6800 XT running at 325 MHz core and 350 MHz (700 MHz effective) memory on a 256 bit bus. It is great to see a new competitor in this arena, and hopefully we will see something fun and interesting from them in the near future. Blue Gears is starting to make some noise again, and they are starting off simple and building up. You can take a look at their site and see that they are currently offering their b-Cool 80 mm cooling fans, which have the integrated LED readout. I have one of these now, and I must admit that it is a pretty nifty little product. I will have a full review of it shortly. Blue Gears first became famous with the X-Mystique sound card, which was the first standalone PCI Dolby Digital Live Encoder on the market. Due to some external pressures and internal changes, Blue Gears broke away from being a distributor of HDA soundcards (now known as Auzentech, which continues to carry the X-Mystique and X-Plosion sound cards). You can take a look at Blue Gear's site and see what other products they are getting into. It certainly seems like a company with a lot of energy and some good ideas, and I know of one product in particular that could really help make this company a household name (though I can't tell you what it is as of yet). The AMD and Intel race is still going strong, and it is only going to get worse starting next month. Both companies will be releasing products, and Intel is having a record ramp with their Conroe family of parts. We still are not sure about the exact timeline for Conroe, as reports have varied wildly (from general availability in July to only limited availability through November). Needless to say, Intel is betting the farm on this product, and from what we have seen so far it is a good bet. Conroe will run cool, pull far less power than the products it replaces, and certainly has industry leading performance. AMD will have AM-2 this summer to help push things along, but they will be at a disadvantage to Conroe until they can release their 65 nm parts. Still, the server space will be well supported by AMD, especially once the socket F based Opertons hit the market. Intel will try to make a resurgence here with the Woodcrest series of chips. The competition is great for the consumer, and now we will finally see Intel have parts out that are worthy competitors to the Athlon 64 that has dominated the performance charts since its introduction in 2003. April 19, 2006 GeForce 7900 Go and Other News - Josh Yesterday NVIDIA launched their latest mobile chips, and Dell already has them integrated into their new XPS M1710. Toshiba also offers the Satellite P105 notebook based on these graphics chips as well. The Go 7900 GTX and Go 7900 GS are both based on the G71 chip that powers the latest 7900 series of desktop cards. The Go 7900 GTX is clocked at 500 MHz and consumes around 45 watts and has the full complement of pixel and vertex shaders that its desktop brother sports. The Go 7900 GS is slightly different as it is clocked to 375 MHz and features 20 pixel shader pipelines and 7 vertex shader units. This lower clockspeed and disabled units gives it a power consumption of around 20 watts. Memory clocks for these products should be 600 MHz and 500 MHz respectively for the two parts. It also appears that the GeForce Go 7900 GTX will be the first laptop with 512 MB of graphics memory. Though NVIDIA cut down the transistor count of the G71 from the G70, it still includes all of the aggressive clock gating technology that previous NVIDIA mobile chips have shown. NVIDIA promised last year that they would be very aggressive in pursuing the mobile market this year, and so far that has been the case. The previous top end chip was the GeForce Go 7800 GTX, which was clocked very close to the desktop variant (400 MHz vs. 430 MHz). ATI once had a very significant technology lead when it came to mobile parts, but we are starting to see that swing the other way. While ATI has released a mobile version of the X1800, this has seemingly not been a popular chip due to heat and power. The X1600 mobile is doing quite a bit better, as it still has a good mix of performance, features, and power consumption. Where ATI has a huge market presence is with its Xpress 200 series of integrated graphics. The X1300 and X1400 are doing well also, and it seems that the X1400 is a favorite for the thin and light market. NVIDIA has already released their other G7x 90 nm chips for notebooks, with the Go 7600/7400/7300. Market data has not been released yet, but we can see that the two big guys have products that are very competitive with each other. Last year at this time NVIDIA really didn't have a whole slew of products that they have now, and their marketshare was similarly as scarce in the notebook market. With these new series of 90 nm chips NVIDIA now has a pretty full stable of products to offer manufacturers in the mobile arena. Notebook refresh cycles are quite a bit longer than that found on the desktop, and as such it seems that NVIDIA hit the Spring refresh with their entire lineup of chips quite nicely. We will start to see a much broader selection of parts than we have in the previous few years when it comes to graphics, and that is certainly good for the consumer. It appears also that TSMC is having some issues with their new 80 nm process. Unlike the previous 110 nm half step, 80 nm comprises three separate lines: high speed, low power, and a standard process. 110 nm was an optical shrink, but transistor performance was similar to that of 130 nm FSG. Because the jump to 65 nm is looking to be tough for the 3rd party Fabs, TSMC is trying to offer a wider variety of parts based on the smaller geometries of 80 nm. It seems that in doing so they have added complexity to the process, and now it appears as though it is delayed. This looks like it is causing some major problems at ATI as they have had to shuffle around their designs to compensate for the lack of a viable 80 nm process. In my latest State of 3D I said I was pretty sure that NVIDIA would not port any of their current designs to 80 nm, and the first 80 nm part we would see from them would be the G8x series of chips. ATI was planning on releasing a whole slew of current designs on 80 nm to take advantage of the size and transistor performance tweaks that the new process would offer. Now it looks as if the earliest we will se a 80 nm part could be in late Fall of this year. ATI certainly hasn't had a good year when it comes to designs and luck, but hopefully they will get back on track once TSMC's 80 nm process family is working as it should. AMD Pushes Up AM-2 Launch It seems the latest scuttlebutt around the industry is that AMD has officially pushed up the AM-2 launch to May 23rd vs. the initially proposed June 6. Apparently Intel will be paper launching their new Conroe based products at that time, though initial availability looks to be as early as late July, or as late as the end of November. Intel does appear to be ramping this product quite aggressively though, and they estimate that the Conroe based parts will constitute around 70% of their shipments by the end of Q1 2007. According to Anand (who has a pretty late model AM-2 processor) we can expect to see about a 5% increase in performance with these latest DDR-2 enabled parts over their previous DDR-1 counterparts on a per clock basis. This isn't a massive increase by any means, but it will allow AMD to utilize all of the new DDR-2 parts that are coming out. These chips will also feature the extra security features as well as virtualization. These chips also look to pull slightly less power than previous versions, and we will probably see a 2.8 GHz dual core FX chip at launch. With the extra line space that Fab 36 brings to the market, apparently AMD has been very successful in producing these new parts in good quantities. Good enough that they are planning on shipping large quantities to OEM's by May 15. Initially it appeared as though AMD would release these products in early March, but due to the lack of a competing solution from Intel as well as issues with the new DDR-2 memory controller, AMD pushed that back to June. Now that the dust has settled, and Intel looks to launch their new parts in late May, AMD is not holding back. All of the chipset guys had working parts by the beginning of March, and so the extra two months have given the motherboard guys time to design good products and stock up on parts for release. Both ATI and NVIDIA have new parts just waiting in the wings, and we can bet that VIA will also have a supporting part as well. There should be some very nice products ready and available at launch, and I bet that end users will also be able to purchase AM-2 chips from retailers like Newegg, Monarch, Tiger Direct, and others. AMD is going to have its work cut out for it through the rest of the year. While Intel will still be ramping its Conroe products, mindshare will probably be swinging away from AMD. It will still take Intel a while to transition over to the new architecture, but the overwhelming buzz is going to be about the power and performance of the Conroe family of parts. AMD is working hard on their next gen 65 nm parts, and these will hopefully put them on an even footing with the Conroe parts. We can expect to see the first iterations of these products by late December, but if AMD is being as conservative talking about their product releases as they have been with their finances, we could see these parts earlier than expected. AMD is still working diligently on their 65 nm process, and all that they have said about it is that all major milestones have been reached on schedule and things are looking good. We of course cannot expect parts coming off the line today to be production quality, but AMD is well on the way to getting out 65 nm by the end of this year. Once this happens then we will again see a bit more parity between AMD and Intel at the top end in terms of performance and power consumption, but the days of AMD having an overwhelmingly faster part in the mix may be over. April 13, 2006 AMD Reports Earnings - Josh Yesterday after the bell, AMD reported a $178 million profit on $1.33 billion in sales. This is a huge increase from a year ago when it reported revenues of $1.23 billion (which included $446 million from the now spun off Spansion). So basically processor sales rose in one year from $784 million to $1.33 billion, which is about a 70% increase in sales. Q1 is typically a slow quarter for the industry, as most buying occurs during the Holiday season. This obviously was not the case for AMD. Their previous quarter showed sales of $1.3 billion, so even with the typically slow first quarter, AMD was able to actually increase sales and double the profit quarter on quarter. AMD says to expect this next quarter to be flat to down, this is not a surprise as Q2 is often even slower than Q1. At least with Q1, some increased sales are still generated from the closing of the Holiday season, and with the backorder of chips that AMD has it is not surprising that sales actually increased. Q2 is going to be a very interesting time in the market, as the balance between shipping AMD and Intel products won't really be known until July. While AMD may be saying sales will be flat, we won't know the extent of the market conditions until Intel announces its own results. So, during that time we may see AMD's sales slip, but if Intel's sales slips as well, we know that AMD continues to erode Intel's base of sales. I think Q2 is probably going to be the last quarter of large gains against Intel, as the market may very well swing back Intel's way with the release of the Conroe series of products. Desktop and notebooks should get a big boost from this new architecture from Intel, but it will take a lot more work to regain ground lost to AMD in the server world. This is an area where AMD's forward looking infrastructure really has a large advantage over Intel's products. While it has not been officially confirmed that Intel's answer to HyperTransport (CSI) is cancelled, it does look like it is pushed back even further than before. We may not see CSI enabled server chips until 2008/2009. This does give AMD a big leg up in servers, especially with the new Socket F products are going to be released this year. While it is not officially confirmed, it appears as though AMD will be widening the HyperTransport links between the CPU's to better scale to 8 processors (current HT implementations have been shown to be bogged down in 8P systems just with cache coherency requests, not to mention the NUMA activity). With Fab 36 now shipping products for revenue, AMD will see its supply issues ease by a great deal. Add to that the Chartered Semi deal that will start to produce parts in the second half of this year, AMD could pick up some new customers and product lines because they can now more adequately fill demand. Q3 will be a very important quarter for AMD, as it will start to compete with the Conroe based products as well as start the transition to 65 nm at both Fab 36 and Fab 30. AMD expects revenue shipments of 65 nm parts by late 2006. AM2 parts from AMD do not look to be competitive at the high end with Intel's Conroe, but it will still be fast and have a very solid foundation in terms of supporting architecture. Since AM2 parts are still based on the current HyperTransport specification, supporting motherboards and chipsets can be designed and shipped in a short amount of time if needed. Most motherboard manufacturers already have AM2 supporting chipsets, as these had originally been scheduled for a March release. Products from NVIDIA and ATI are already in their final form, and there should be an excellent selection of AM2 supporting motherboards from their partners. AMD could still gain marketshare from Intel, but much of that will depend on how AM2 is accepted by consumers as well as how aggressive Intel will be with the Conroe release. AMD has gained a large amount of ground on Intel, but they do not appear to have a competing part to Intel's Conroe until their own 65 nm parts start to ship. Again, that will not be until the end of the year at the earliest, so between July and December AMD will have to be more aggressive with their prices to continue to eat marketshare. The only way that AMD will continue to gain marketshare without eroding their prices will be if Intel has a hard time actually releasing Conroe based products. So far it appears as though producing Conroe may not be as easy for Intel as some have thought, but until Intel starts shipping parts in volume we will not know the extent of this rumor. April 11, 2006 Comments and Links - Josh I have received some interesting feedback from the State of 3D article, and of course it is usually balanced between "good article" and "you don't know what you are talking about". Such are the joys of the internet! One very interesting topic that it did bring up was the possibility of extra redundancy in today's GPU's. While it is well known that there is quite a bit of redundancy in there already, and we see the GPU guys often take partially defective parts and disable portions of it and sell as a lower SKU, how is ATI being able to get so many good R580 chips out without apparent defects? The R580 is a very large part, and according to current yield models should only have a couple dozen good chips per wafer. But this does not jive with reality apparently. Dave Orton doesn't have a garden path in his backyard paved with defective R580 parts, and so far we haven't seen one of the lower end SKU's yet utilizing these partially defective dies (though it is rumored that by the end of this month we will see a X1900 GT or GTO, depending on what they decide to call it). Still, R580 has been available in good quantities since January, and if yields for that size of chip fit the industry standard, then we should see a really large portion of production being partially or totally defective chips. That does not appear to be the case. This then brings up a second point. Since both ATI and NVIDIA utilize redundancy in their larger designs, was much of that redundancy taken out with the G71 chip from NVIDIA? Could this explain further where some of these transistors were taken from? Was NVIDIA right on the cusp of the yield equation that would say either they needed the redundancy or could achieve similar yields without redundancy because the die size was so much smaller than the G70 and R580? I can see how in this situation because the G71 was so much smaller, they could make it even smaller by removing much of the redundancy and actually achieving as good of yields without the extra transistor count? Again, the article was a lot of speculation because actual yields and low level design decisions are considered trade secrets, and journalists like me are not let in on them. That is probably good, because I would want to let all of you in on these secrets. Hot Hardware takes a look at the Sapphire Blizzard X1900 XTX, and they like what they see. The Blizzard uses a small, enclosed liquid cooler instead of the stock blower. While temps are down slightly from stock, it does produce quite a bit less noise. Considering the variable noise nature of the typical ATI cooler, this is a big positive. Anand gets his hands on a preproduction AMD Athlon 64 AM-2 processor and puts it through its paces. While it is slightly faster than a standard DDR-1 based Athlon 64, it may have a hard time competing against Intel's Conroe based parts. Of course while this still may not be reflective of actual performance of AM-2 parts coming this June, it probably is very close to the final product. AMD expects a lot of these processors to be available for the June launch, as they are supplying OEM's with final product in mid-May. We still are not entirely certain of Intel's release schedule though, but from hints dropped we should see good availability of Conroe parts by October. We shall see who shakes out which product first and can supply them in good quantities. April 7, 2006 The State of 3D: Economics of 3D - Josh While ATI and NVIDIA are duking it out across a wide range of products at different price points, I thought it would be interesting to see how some of the production aspects of these products match up. NVIDIA is now pushing the "performance per mm squared" metric, and while that may be all well and good, there are other factors in design that need to be considered. While NVIDIA certainly has products that are cheaper to produce at every level, we need to take a look at how these production costs match up as well as how the differing design philosophies of each company may affect the outcome of the next generation of parts. Here is a direct quote:
You can read the entire article here.
If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2006 PenStar Systems, LLC. |
|||
