   |
|||
|
|||
|
:
News
: Archives
|
|
|
|
|
|
|||
|
|
|||
|
|
|||
|
October 29, 2007 The State of 3D: Q4 2007 Instead of releasing a 8800 GT preview, I decided to go back to my editorial roots and take a look at not only the 8800 GT, but the upcoming competition. This is not merely a G92 vs. RV670 article, but a good analysis of the thinking behind the companies and their products, and how they relate to other technology trends as well as distribution. The next few months will bring a whole slew of products that some have not foreseen, and the reactions from both companies could very well inflame the marketplace with outrageously competitive prices and marketing vitriol. Needless to say, it is going to be a whole lot of fun from my perspective.
You can read the entire article here. October 22, 2007 AMD's Quaterly Results and Analysis AMD's latest quarterly report was seemingly the very definition of "bittersweet". The company improved its revenues by a significant margin from the previous quarter with $1.63 billion in sales. It unfortunately posted a $396 million loss. So while AMD was able to report record sales, they still have some significant hurdles to overcome to break even by the end of this year. There are several very positive pieces of news from the previous quarter. The most obvious is that demand for CPUs and GPUs is quite strong, and looks to stay that way through the next quarter. The second piece of good news is that AMD was able to raise margins to a respectable 41%. That is up quite a bit from the previous quarter's 33%. The final piece is that the graphics division currently has three major chips that they are selling for a profit (R600, RV615, and RV630). AMD is slowly converting their CPU mix to 65 nm, which is a much more efficient process to produce CPUs on. The 65 nm chips are obviously smaller and run cooler. So while AMD is offering dual core chips at rock bottom prices, their manufacturing ability allows them greater efficiency when considering overall die size and the number of good chips per wafer. The majority of 90 nm starts look to be focused on Opterons and higher speed binned Athlon 64's (such as the X2 5600, 6000, and 6400) all of which are 1 MB L2 cache chips. So the combination of increased sales with CPU prices holding fairly steady, and the mix of products skewing towards the 65 nm node, we can see why AMD has reported much better margins than before. On the graphics side we saw the introduction of the high end HD 2900 XT (R600), the HD 2600 products (RV630), and the HD 2400 cards (RV615). The R600 is a 80 nm chip fabbed on TSMC's 80 HS (High Speed) process. It is an expensive chip, but the margins on it appear solid due to the $350+ price of the overall product. The other RV6xx chips are 65 nm products, and the die size is very small on them both. AMD had a bit more leeway with pricing on those products, and the new chips all combined to increase overall sales of the graphics division by around 29%. Demand for the 65 nm products appeared to be very good, as they provided similar performance to the current NVIDIA parts but with a lower power draw and less heat. One of the bigger downsides to the results was the continuing cost of integrating AMD and the former ATI. Quite a few of the top people at ATI have left, and it looks as if some of the middle management is also jumping ship. From what I hear, the cultures are not all that dissimilar, but the graphics division is trying very hard to be as independent as possible from the AMD corporate entity. This has caused a lack of communication where needed, and most likely a lot of political tension between the groups. AMD wants to be one big happy family, but apparently the ATI group certainly enjoyed their existence as a separate entity that answers only to themselves... not some guys in Texas. Time will heal most wounds here, and as ATI had engulfed several other graphics chips makers, they too will eventually be fully integrated and things will work out. Something else of great interest that has popped up recently is the "leakage" of next generation AMD chipsets by motherboard manufacturers. Apparently AMD has been done with the RD790 and RD780 chipsets for some time now, and have been shipping them to the motherboard guys. Originally AMD wanted to release these chips in the middle of November, but the motherboard guys were not terribly keen on sitting on these chipsets and waiting for AMD to officially release them. With margins as tight as they are, it makes no sense for the motherboard guys to be sitting on chipsets that they have already bought, when these products could be adding to the bottom line for the current quarter. AMD has no interest in pissing off the motherboard guys, because of the simple logic of "no motherboards, no CPU sales". So AMD is not saying a word as we see the first batches of 790 based motherboards hitting the market. So far MSI and Gigabyte have released their products in certain areas around the world, and the situation will only get worse/better the closer we get to the middle of November. On the positive side, there is a lot of information leaking out that the upcoming RV670 chip, which will fill the $199 to $299 area, will be a very impressive part. AMD appears to have fixed a lot of the issues with the R600 chip, and the RV670 looks to fulfill the promise of the architecture. One of the areas that is rumored to have major work is the speed of the shader based anti-aliasing (color resolve was done in shaders under most circumstances). This was the biggest complaint by far with the architecture, and if in fact they have fixed it, then the RV670 will be a cheaper yet faster successor to the much maligned HD 2900 XT. They have also added double precision floating point into the mix for GPGPU. By all accounts, production of R600 chips came to an end several months ago, and supplies of HD 2900 XTs are drying up fairly quickly. Barcelona has yet to make a big splash, and it appears the reason for that is production is still gearing up. Once the CPU is designed and the major bugs squashed, the designers sign off on the chip and send it to the fab guys. Once there, even more work is done. The fab guys work on the process mix to optimize both bins and yields, as well as do a lot of metal work to overcome speed issues with the design. This means if they find a critical path which inhibits higher speeds, then they do metal layer work to alleviate the situation. This is a very simplified explanation, as they also adjust the masks as well. Once the fab engineers get a chance to work on the design and cure these problems, then they ramp up production. Right now it appears as though the engineers had their time with the chips, and they are starting to ramp production. The first chips off the line were all lower clocked Opterons, but this November we are supposed to see the first Phenom desktop chips hit the market. These will need to be faster than what we have seen so far. As the months roll by, we will see gradual improvements in power consumption and speed bins, and by January/February we will likely start seeing some significantly faster chips from AMD. But until that time, the stream of chips look to be something a little bit more than a trickle. It appears as though Intel has eased up on the price war, and this is good news for AMD. The bad news is that Intel is going to release their first 45 nm parts shortly, and ramp production up for Q1 of next year. AMD is at least 6 months behind Intel with 45 nm production, and my guess is more like 10+ months behind. While AMD is saying that the first 45 nm products will be at the beginning of the second half of next year, I think that a more accurate timeframe for mass production is likely Q1 2009. This is not the end of the world for AMD, but it certainly keeps them from more adequately competing with Intel when it comes to overall performance and margins. Q4 should get AMD a lot closer to break-even, but I think that they will still be around $100 million in the red. There is far too big of a hurdle in front of them for this quarter, and the product mix they will be putting forward is unlikely to get them over the hump. Phenom, new chipsets, and RV670 will not be enough, but AMD will continue to improve on efficiency and they will likely take less of a hit from further ATI acquisition charges due to most of the top brass and redundant employees at ATI being gone. The current quarter will be an interesting one to watch though, get ready for a lot of AMD action this coming November. October 17, 2007 Biostar TForce TA690G Review Biostar was kind enough to ship me their TForce TA690G motherboard based on the AMD 690G integrated chipset. Having had a good experience with the previous Biostar motherboard I reviewed, I was quite excited to see what this product had in store. Needless to say, I came away very impressed by Biostar's budget integrated offering. This board can certainly find a good home in nearly any class of machine, even the budget enthusiast! Here is a quote:
You can read the entire review here. Intel Reports Outstanding Results It looks as if the transition to 65 nm and plenty of belt-tightening have lead to a tremendous quarter by Intel. The chip giant reported revenues of $10.1 billion, and a net income of $1.86 billion. Intel sites very strong microprocessor sales for the 3rd quarter, which has given many in the industry hope that other PC companies will also reflect good numbers from increased demand. Intel has also worked hard to implement 65 nm production across the board, which allows for greater production rates of high end processors. Intel has also cut its workforce by around 12% from last year, and they expect to save a lot of money from that restructuring alone. AMD is set to report their earnings after the close of the market on Thursday, and though Intel has regained lost marketshare, many are expecting AMD to show stronger sales due to higher demand for CPUs. Intel certainly is in the drivers seat, especially considering how slow AMD has been at releasing their native quad core Barcelona products. Also consider how much of AMD's production is still 90 nm. Intel will be releasing their first 45 nm products in the very near future, which will spell more trouble for AMD when it comes to margins and competing products. October 15, 2007 45nm Quad Core Extreme Issues While this is by no means officially confirmed, it is looking like people who received the upcoming 45 nm quad core Extreme processors from Intel are having issues running them on many current motherboards. I know of two issues people are having with 680i boards with the latest P35 bios, as well as some other motherboards that should be compatible. In one case only two cores can be run at full speed, and when the other two cores are enabled in the BIOS the system becomes unresponsive quickly. The issue so far looks to stem from the fast heat up of the quad cores. One would expect that the 45 nm Penryn based cores would run cooler and pull less power than the current QX6850. So far, that might not be the case. It could also just be BIOS incompatibility with the new cores. We will find out more at the end of this month when the NDA's run out. I do find this release rather odd though, as Intel is not distributing X38 motherboards with their quad core samples. One would think that they would have a distinct interest in making sure that the reviewers had the necessary platform to complete the review, and have good results. As it is, the hodgepodge of motherboards that are likely to be used for this review may cause some issues due to immature BIOS support or incompatibilities. Time will tell, but I am certain that Intel and their motherboard partners will iron things out. October 10, 2007 As most are aware right now, we are coming up on some new graphics cards releases in the next month and a half. NVIDIA will be leading the way with their 8800 GT on October 29 supposedly. This is based on the G92 chip, which is a new 65 nm chip likely composed of 128 stream processors (G80 has 128 while G84 features 32). We also know that it is a 256 bit memory bus chip, so it does not feature the widened 320 bit and 384 bit bus of the G80. Initial estimates show the top end 8800 GT card will have 512 MB of memory, run at 600 MHz core, and run around $249 to $299. The 128 stream processors have also been slightly optimized, as well as running faster than 1.5 GHz. This chip will also be the basis for the new Tesla cards which will support double precision FP64 in the stream processors, but double precision will not be enabled in the consumer level cards. The G92 is not a dumb shrink, as it adds some other goodies to the mix. It goes with the G84/86 texture address/filtering setup rather than the older G80. It also features PureVideo 2. The chip pulls less power than its larger and older brethren, but initial plans for this part have likely changed due to AMD’s upcoming RV670 based boards. From all indications it seems like NVIDIA was going to place this between the 8600 GTS and the 8800 GTS, and likely eliminate the 320 MB version of the 8800 GTS. Cooling was originally not an issue, as NVIDIA did not look to clock it above 550 MHz. Now that word has leaked out on how potent the RV670 is, I think NVIDIA is working overtime to get the performance of these parts up. So, this means that the original cooling meant for this card may not be entirely adequate, as NVIDIA has put out a call to OEMs to send in their systems that will be featuring this card to check for thermal properties and make sure that the 8800 GT will not overheat. This all makes me honestly wonder if the initial clocks of the 8800 GT will not be significantly higher than 600 MHz. We also have to wonder if NVIDIA will rework the cooling and make a much beefier solution. While the 8800 GT will represent the upper midrange of the market, the 8800 GTX and Ultra will still continue to address the high end. These will still be the fastest NVIDIA cards until the beginning of next year. This all leads up to AMD’s RV670. What was initially thought to be just a midrange chip is looking to turn out to be much, much more. This 55 nm new chip from AMD contains all of the primary functional units of the older R600. This means 320 stream processors, 1024 bit ringbus, and what looks to be the same amount of RBEs and texture/filtering units. A big change is that the chip will “only” feature a 256 bit bus, like its competition from NVIDIA. Memory bandwidth may not really be a problem though, as faster GDDR-4 is already available. It may not be the 512 bit bus that the R600 had, but that in all honesty seemed like overkill because the chip was never able to really utilize that bandwidth. It seems that a lot of work has been done on the RBEs as well. The Achilles’ heel of the R600 was its poor anti-aliasing performance due to its reliance on shader based color resolve. I do not yet know the details, but it appears that AMD has fixed this problem and we can expect much better AA performance. AMD may still be doing shader based color resolve, but apparently it is now highly optimized. When running without AA the HD 2900 XT was often able to compete with the 8800 GTX/Ultra in many applications, but once AA was turned on performance typically was brought down to the 8800 GTS level. The chip will run around 800 MHz with fast GDDR-4 at the high end. We can expect this card to be around $300, which is cheaper than the current HD 2900 XT. It will also thoroughly spank the HD 2900 XT in overall performance as well as features. It will have the complete UVD unit, and implement SM 4.1 functionality. And it does it all with single slot cooling and a power draw of around 120 to 130 watts depending on the configuration. Some of the initial cards from other AIBs will likely features dual slot cooling, plus overclocked configurations. I also wouldn’t be surprised to see an 8 pin PCI-E version that will really clock up there. This looks to be the next high end for AMD for a while, until at least the R670 comes into play. From all indications the R670 is actually two RV670s on one PCB. Expect that card to come out early Q1 2008, right around the time NVIDIA will release its 8800 GX2 type product for the high end. AMD took a chance on releasing a product on TSMC’s 55 nm process, and it looks like this gamble really paid off. NVIDIA is sticking with the more mature 65 nm process, but in this case their conservative bet did not look to pay off. 55 nm is about 90% the size of the 65 nm products, but for some reason TSMC added a few other features to the process and are expressing that designs will pull 10% to 20% less power according to Daily Tech. This gives AMD a little bit of an advantage in both size and power over competing NVIDIA designs. AMD stumbled with the R600, and the RV630/610 parts certainly did not impress many people either, but it appears as though AMD has righted the ship and the RV670 is going to be a pretty impressive product at its price point. Launch of the RV670 appears to be on the same day as the launch of the AMD Phenom and RD790 chipset (and its derivatives). So reviewers will have their hand’s full getting the Phenom, motherboard, and accompanying graphics cards benchmarked and written up! This will be a huge day for AMD, and they are hoping that overall it will be thermonuclear explosion of information and performance figures. Something that is quite telling about this launch is that AMD and its partners stopped sampling HD 2900 XT cards some time ago. A PR person from one of the partners was overheard saying, “Why would I want to sample anymore HD 2900s when the RV670 is so much faster and almost here?” The rest of this year is going to be very, very interesting. October 1, 2007 Things are finally becoming quite clear in the upcoming graphics updates. The next several months will see a whole slew of products from both NVIDIA and AMD, and we will start seeing them beginning in November. Before we hit the graphics though, one chipset tidbit has come to my attention. AMD's chipset refresh for the Phenom processors has been pushed back by about a month, so this next November we will see the 790X, 780X, 780G, and 740G. These will span the high end to the low end. The big question is if they will launch with the SB600 or SB700 southbridge. SB600 has been out for a while, and the delay of the chipset release might coincide with the availability of the SB700. This is not set in stone though, and historically speaking southbridges have been the more problematic part to design and produce. NVIDIA will be releasing their nForce 6 series of AMD chipsets in November as well. These both will coincide with the introductory release of the Phenom processor from AMD. Currently AMD is shipping their new HD 2900 Pro, which is a fully functional R600 chip clocked at 600 MHz. From my understanding the yields on the R600 chip have been good, but the power draw and speed bins for these parts have been the limiting factor. It is not surprising that AMD has put out the lower clocked HD 2900 Pro. They have also just released the HD 2900 GT which has a SIMD unit disabled, so it features 240 stream processors vs. the standard 320. Due to the design of the R600 chip, we can assume that one RBE (render back end) unit has also been disabled, as well as a texture unit. So overall it looks like a 12/12/240 part. Also from my understanding is that these parts will not stick around long. From all indications AMD has stopped ordering wafers based on the R600, and is instead concentrating on the RV670 from here on out. The RV670 is essentially a full R600 chip, but produced on TSMC's 55 nm process. It also goes from a 512 bit memory bus to 256 bit. Most likely these chips will be bundled with DDR3 for the midrange parts, and GDDR4 for the upper speed units. We can assume that the upper end boards will clock and perform at the current HD 2900 XT level, but pull a lot less power and produce less heat. Leaked pictures have already hit the web showing single slot cooling and with one 6 pin PCI-E power connection. The high end version will likely be two RV670 chips on one board. There has been much speculation if AMD will put in UVD in this chip, and I still do not have a definitive answer. One would think it would be a good idea to include, as the low end SKUs with this chip will hit the $150 mark. The current 65 nm chips from AMD will also get a refresh at the beginning of next year, migrating to 55 nm as well as supporting PCI-E 2.0. From all indications there will be no major architectural changes, so it is quite likely that we will not see an increase in anti-aliasing performance as compared to the current batch of chips. While none of AMD's chips can be considered "slow", they certainly take a big hit when AA is enabled due to the MSAA resolve being done through shader hardware. This may have saved on transistor budgets, but is much slower as compared to the MSAA hardware resolve based in the ROPs/RBEs. Hopefully AMD took the time to work on the architecture and improve its per clock performance, but I would consider this unlikely due to the timeline of these chips. On the NVIDIA side we will see the midrange G92 refresh this November. This looks to be a shrunk G80 chip on 65 nm, but with some small architectural improvements as well as the inclusion of PureVideo 2. The first products will start to replace the 8800 GTS 320, and potentially the 8800 GTS 640. This chip looks to mirror the entire featureset of the G80 along with the G84 improvements (128 stream units, 32 texture address, 32 texture filter, and 24 ROPS). The chip features a 256 bit memory bus though, which is down from the 320/384 bits of the 8800 series. These new cards will utilize 256 MB, 512 MB, and 1 GB memory densities depending on the price and market focus. Clockspeeds for the G92 based chips look to reach around 800 MHz for the core, and the shader clocks could reach upwards of 2 GHz. For a "midrange" chip, this looks to be a screamer. NVIDIA will also refresh their current midrange and budget parts with 65 nm variants, but those will likely be released around the New Year. The enthusiast class card will also be released right after the New Year, and it will be comprised of 2 x G92 chips in a card very reminiscent of the GeForce 7950 GX2. This high end unit will not last as long as the 8800 GTX/Ultra has, and looks to be replaced by another high end single chip solution within six months of its release. That chip's name hasn't been leaked so far, but we can expect it to be a large monolithic chip made on TSMC's 55 nm process. Another possibility though is that this "chip" will in fact be two chips on one substrate/package. That certainly is not outside the realm of possibilities, as that is the direction that AMD is pursuing with its R700 class of graphics chips. So, between November and February we will see a whole slew of chips and cards hit the 3D market. It will be a very interesting time of course, and we will see 3D performance again get a shot in the arm at all price points. September 26, 2007 AMD's Better By Design Program While the Core 2 series of products have been taking all the limelight as of late, AMD is still chugging along and trying to address its core market with competitive and inexpensive products. The basis for one of their larger pushes is the Athlon 64 X2 processor and AMD 690G chipset. I was able to take a look at two full systems featuring AMD and Intel products, and put them head to head and see who comes out on top. I cover many topics in this review, so it is a very good overview of what is offered for the integrated market. Here is a quote:
You can read the entire review here. September 25, 2007 NVIDIA Releases GeForce 7 IGPs for Intel Intel is one of the graphics leaders in the field mainly because their integrated graphics are sold in nearly 90% of all integrated motherboards for the Pentium 4 and Core 2 series. AMD had previously released the Radeon Xpress 1250 chipsets for Intel, but due to the focus that AMD has on its own processors, these chipsets have not been widely picked up by motherboard manufacturers. Now NVIDIA has finally released their Intel compatible GeForce 7150, 7100, and 7050 mGPU products. These little numbers bring useable SM 3.0 support to the Intel integrated platform (as the G3x and 9x5G series of Extreme Graphics have large compatibility and image quality problems). The upper versions support HDMI output, DVI, and multi-display support. The low end 7050 only features the standard DB-15 output. The top end products are all bundled with the 630i southbridge which features Gig-E networking, RAID 0,1,0+1,5 support, 4 SATA, and 2 IDE. The 610i carries over most of these features with the exception of standard 10/100 ethernet and no RAID 0+1 and 5 support. The top end 7150 product sports a clockspeed of 600 MHz+, but how much more than 600 MHz is not known at this time. This is a significant upgrade from the AMD platform which had these GPUs running in the 425 MHz range. The next level down is the lower clocked 7150 running at 600 MHz, while the budget oriented 7100 and 7050 are at 500 MHz. All of the chips except the 7050 support 1333 MHz FSB Core 2s, and should be able to support the correctly clocked 45 nm Penryn based processors. The 7150 models support DDR-2 800 speeds while the 7100 and 7050 support DDR-2 667. I have confirmed that this product is made on TSMC's "80G" 80 nm process. We must also consider that the graphics portion has direct access to the memory controller on the Northbridge, rather than going through the HyperTransport tunnel to the AMD processor which has the IMC. This could lead to some significant improvements in performance for this part, and should be able to outrun anything Intel has as well as be very competitive with the AMD/ATI Xpress 1250 offering. NVIDIA has a good chance to take a significant portion of integrated marketshare from Intel, and OEMs are certainly excited about this development. It is probably pretty amazing how many support phone calls that these companies get because the integrated Intel Extreme Graphics parts do not run very many games, and of those they do run they may not run them well (graphical glitches, unsupported features, poor performance, etc.). With AMD/ATI essentially abandoning the Intel integrated market, NVIDIA should pick up quite a few customers for their integrated parts. Again, the prospect of ending a lot of support calls concerning casual gamers who just want their titles to work, expect to see Dell, HP, and others start introducing these parts in short order. One thing that is missing from the lineup is support for PureVideo. Currently none of the Intel based IGPs support PureVideo. NVIDIA recommends getting an inexpensive 8500 based card if PV support is needed, or utilize their 7050 PV AMD based chipset. I am not entirely sure why this was left out, as AMD provides Avivo support for their Xpress 1250 chipset, and Intel has their ClearVideo feature. AMD Releases the HD 2900 Pro The massive price difference between the HD 2600 XT parts and the HD 2900 XT has caused a great amount of concern for AMD because of the large gap left in their lineup. NVIDIA has filled this gap rather nicely with their 8600 GTS and 8800 GTS 320 products, and have cleaned up rather nicely. AMD has finally released a $250 level part that should even the playing field a bit. The HD 2900 Pro is a full R600 graphics chip featuring 320 stream processors, but at a lower clockspeed. We will see two version in retail, one with 512 MB and one with 1 GB of memory. The GPUs are clocked at 600 MHz while the GDDR-3 memory is clocked at 800 MHz. Unlike the HD 2900 XT 1 GB with GDDR-4 memory, the 1 GB version of the Pro will feature slower GDDR-3. The 512 MB versions will sell for $249 while the 1 GB is at a fairly reasonable $299. Samples are limited right now, but we should see a growing number of these cards hit retail in the next few weeks. Apparently these cards overclock quite handily and can reach HD 2900 XT speeds with a bit of luck. Initial results show this card to be fairly competitive with the 8800 GTS 320 MB card, which is a good thing considering their respective price points. So far it appears that early January will be the time when AMD releases a high end new product based on the RV670 chip(s). Until that time, AMD has to hang its hat on the HD 2900 series to give it a presence in the midrange and high-end market. Further Thoughts on Triple Core Something that had escaped my thinking, but was brought to my attention by Damon at AMD, was that the triple core isn’t just about utilizing parts with defective cores, but also maximizing the speed bins of these parts. For instance, if a quad core works perfectly fine, but one core runs at 3 GHz, two of them run at 2.8 GHz, while the final core runs at only 2.0 GHz. Would it be an effective choice to simply brand it as a 2.0 GHz quad core and sell it for a lower price, or would they rather sell it as a 2.8 GHz triple core processor? Apparently the Fabs in Dresden are able to test each individual core on a Barcelona chip and see what its maximum clock speed is, along with power consumption and other factors. This allows the engineers there to not only help tune the process, but more adequately sort out the chips and fit them into meaningful SKUs. Any CPU has a lot of redundant circuitry in it, so it is not all that uncommon for there to be small defects on a die that can be fixed. But when dealing with maximum clock speed, few things can be done. So, this is another wrinkle in the story of why AMD is introducing a triple core chip into the marketplace. It will find plenty of homes in many systems due to its price vs. performance, as well as the ability to run as fast as a dual core, but offer a compelling performance increase in applications that can address more than two cores. All the while pulling less power and producing less heat at the same clockspeed as a fully working quad core. There are a lot of positives in there, and AMD is really maximizing their limited production capabilities.September 20, 2007 AMD’s Triple Core – Josh AMD has a big problem. They are small. Intel is large. To try to overcome that massive disadvantage, AMD has to not only execute on their plans, but often take their planning and thinking at right angles from what conventional wisdom dictates. AMD often has to do a lot more serious long term planning, and get that planning right because they simply do not have the resources to change directions suddenly (much like what Intel did with the Pentium 4 to Core 2 switch). AMD’s triple core does not seem to be a product of long term planning though, but rather from need that has arisen due to manufacturing. AMD’s quad core Barcelona is a large chip at 280+ mm square. The larger the die size, the more likely that die will have some kind of defect which can ruin the entire chip. So, AMD has a large problem on their hands. With money as tight as it is, how do they extract money from defective die? They are planning on doing the same thing the GPU guys have done for years. Disable a part of the chip and offer it at a lower price than a fully functioning unit. There are several other advantages beyond selling chips that would otherwise be defective. With a portion of the die disabled, then power consumption will obviously go down as a whole. AMD does have a lot more power saving features in Barcelona, but when one of the cores is totally disabled, it is not pulling power. Also, that extra “dead” space on the die will not be producing heat, but it keeps the same surface area as the fully functional quad core. This means that the chip will pull less power, but have the same surface area to dissipate the heat it produces. Most applications are single threaded, more are coming out that can handle at least two threads. There are in reality only a handful of programs at this time that actually utilize more than two cores on a processor. Most of those are more professional oriented titles. We are quickly coming to a point where we shall see a lot more applications take advantage of two or more CPU cores. By releasing a triple core product, AMD is essentially creating a new market for themselves. Here is a product that will perform identically to a similarly clocked quad core in single and dual thread applications, but cost significantly less. That same product when compared to a standard dual core chip in the same price range will handily outperform it in applications which can utilize more than two cores. One of the last arguments out there is that because of the triple core design of the Xbox 360, there are already game ports out there that can adequately utilize a triple core architecture. This is probably the weakest reason behind this idea, but it still is a valid point. Then again, most PC ports from consoles are heavily recompiled (PowerPC to x86), so some of those potential advantages could be swept away depending on the folks porting the title. The success or failure of AMD’s X3 product will lie squarely in the hands of consumers. They will decide if the idea of a triple core x86 processor is good or bad, and make the decision with their dollars. I personally think it is an interesting idea, and one that has proven to be quite good with the GPU guys. It really makes sense that AMD follow their lead and harvest as much money as possible out of partially defective chips. IDF is going on right now and what Intel is showing is very impressive. AMD's Barcelona does not look to overpower the current Core 2 chips, much less Penryn, but hopefully they will still compete well. We will certainly have a better idea how things will stack up until AMD gets to 45 nm next summer (hopefully) once November swings around.
If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2007 PenStar Systems, LLC. |
|||
