November 14, 2005
Creating the 512 MB 7800 GTX - Josh
Being a writer for the industry you tend to bump into a lot of people who work at many of the companies you write about. Not everyone you meet is in PR or marketing, so oftentimes we hear stories of what goes on behind the scenes. Not only funny annecdotes, but also products that had been considered as well as ideas that had been chucked out the door. The tale of the 512 MB 7800 GTX started late last year, and only now has it come to fruition. Here is a quick quote:
To read more about the twists and turns that this product underwent through it development, take a look here.
More 512 MB 7800 GTX Previews and News
Today is 512 MB 7800 GTX day, and there are more than a handful of previews out there. The 512 MB 7800 GTX comes with a base clock of 550 MHz core and 850 MHz GDDR-3 (1700 MHz effective). This is one fast video card with a pricetag to match. NVIDIA again is offering instant availability with this product, but so far the lowest price I have seen is $699. MSRP is set at $649, but supply and demand is rearing its ugly head here and retailers are grabbing what profit they can! Here are a few previews so far:
BFG Tech would also like to announce that they have delivered the 512 MB 7800 GTX at 580 MHz and 875 MHz GDDR-3 (1750 MHz effective) bundled with Call of Duty 2. They are also bundling this title with their current 7800 GTX and 7800 GT line. Expect to see another fun little surprise with BFG 7800 GT and GTX cards in the coming weeks as well!
November 8, 2005
As you may well know, we are in the middle of some big process changes. As time goes on, it gets much harder and more expensive to adopt the next node. Not news there. Intel is leading the pack with 65 nm production about ready to ramp up. We are expecting to see the first 65 nm parts from Intel near the beginning of next year. I would say that parts would be fairly available by February. Intel isn't changing their product designs to any great degree, so we will still be seeing Prescott and Pentium M type processors, except being 65 nm instead of 90 nm. This means smaller, cooler chips for Intel (and considering Prescott, this is a wonderful thing). Conroe and its derivatives will be hitting around early Fall of next year, and they should be very solid products that will give AMD a real run for its money.
AMD doesn't have the treasure chest of R&D funds that Intel does, so they have to go about things a bit differently. They have a longstanding partnership with IBM and process technology, and the two are working together to get 65 nm going as fast as possible. AMD opened Fab 36 last month, but the initial production will all be 90 nm with a transition to 65 nm around late Fall of next year. The fab is still based on 300 mm wafers using the 90 nm node, so this will give AMD a pretty hefty boost in processor availability (potentially doubling production). It will still be a while before we see AMD introduce 65 nm products, but they are confident that their upcoming parts will be more than a match for what Intel has coming anytime soon. Socket M2 based processors will receive a new core to handle DDR-2 (the memory controller on the current Athlon 64 processors cannot support DDR-2). I would expect this new core to have some really nice overall improvements in per clock performance as well as the ability to scale a bit higher in terms of clockspeed. AMD appears to be going to dual core for the FX processors, with the FX-60 being a 2.6 GHz dual core product. AMD will continue to support single core products on the M2 socket, but they do not appear to be going above 2.4 GHz. When AMD hits 65 nm we can expect a slightly new core, but it probably won't be anything radical. AMD is pretty conservative by staying with well known designs ported to a new process, and once the process is solid then they introduce new cores to take better advantage of that new node. Needless to say, by the time Intel introduces their Conroe based parts, AMD should have 65 nm Athlon 64's close to being ready.
3rd party foundries are typically well behind the curve when it comes to process node changes, and only now are TSMC and UMC starting to put out significant amounts of 90 nm products. TSMC has several 300 mm wafer fabs putting out 90 nm products, but for the past nine months or so the majority of parts coming off of the lines have been fairly simple as compared to the monster GPU's of ATI and NVIDIA. TSMC usually has a higher performing product than most of the other foundries, and by all accounts their 90 nm process is quite robust. There now appear to be plans to do a smaller jump at TSMC to 80 nm before they go to 65 nm. This is much in line with TSMC's intermediate processes (220 nm, 150 nm, and 110 nm). This shrink though is not going to be performance oriented, but is aimed at providing smaller dies with the same transistor performance as the 90 nm Low-K process. From rumors around the industry TSMC is going full bore on producing 90 nm products for both NVIDIA and ATI, but the majority of these products will not be done until after the holiday season. I do not expect TSMC to release 80 nm products for at least a year, and they will not hit 65 nm until about two years from now.
ATI is having a heck of a time as of late, and their issues with the R520 and RV530 chips have thrown a real wrench into their current situation. ATI didn't want to spend a lot of money on 90 nm orders if they were not confident that they had fixed their problems with their designs, and so far only good quantities of RV515 chips have been delivered. The rest of ATI's lineup are in severe allocation due to a couple of bad quarters and the cost cutting that was most likely implemented. Now that the fix is confirmed to work, ATI probably made large orders of the RV530, and only middling orders for the R520. Good quantities of R520 do not appear to be likely available throughout the holiday, and I highly doubt that it will be a product with a lot of supply anyway. Many are suggesting that ATI is spending as little money on R520 as possible, yet still providing product to keep NVIDIA in check. The R580 appears to be the product that will really make ATI competitive with NVIDIA again by Spring of next year. Until then, it appears as though ATI is biding their time and cinching their belts tight.
NVIDIA is on a real roll with their recent introduction of new parts. The 6600 DDR-2 looks to be the new budget champion, and many of these products will hit the $99 to $125 mark, and will probably be quite successful for the Holiday Season. DDR-2 is in good supply now, and spot market prices on these chips are lower than DDR-1 of the same density. We will see quite a few DDR-2 enabled cards by the end of this year, and most of them will be paired with 6600 chips. The recent 6800 GS is a cheaper alternative to the 6800 GT, and will take over the $199 to $249 spot for the next few months. This will stay the midrange champ for NVIDIA until the 90 nm 7600 series will be released most likely in Spring of 2006. To replace the aging 6200 and 6600 chips, it appears as though NVIDIA will release their 7200 product around February of 2006. NVIDIA's current 7800 series are taking the top end by storm, and the G70 design on 110 nm has been a complete success. Most G70's are able to clock to 500 MHz (and slightly above) with standard cooling, and this means that chips coming off the line could probably achieve more with some tweaking. Yields and bins for this chip on 110 nm are very mature, as we can see by the rapidly declining price of the 7800 series from NVIDIA since their June and August introduction.
NVIDIA already has products released utilizing TSMC's 90 nm process, but these are currently limited to the GeForce 6100 and 6150 series of chips for the motherboard market. In February we will expect to see the 7200's, and in March the 7600 launch. Depending on ATI's R580, NVIDIA will position its 90 nm high end product accordingly, so they will attempt to cut ATI off at the pass again. Rumor has it that NVIDIA already has high end 90 nm silicon back and running, and the RSX chip used for the PS3 is in the lab as well. NVIDIA will milk the current G70 for as long as possible, and it will continue to be the high end for NVIDIA until Spring of 2006.
If I were to speculate on what the 90 nm high end product from NVIDIA will be, I would guess that it would encompass 32 pixel shader pipelines, 10 vertex shaders, and continue to have 16 ROPS. These will be improved ROPS though, with better performance over the current design as well as having the ability to do AA and HDR at the same time. There has also been some side talk about improving the texture filtering of this next generation product (which is SORELY needed). We can also expect some real improvements in SM 3.0 performance, as it will be introduced a full two years after NVIDIA's first SM 3.0 enabled product. Perhaps we will see NVIDIA quietly whisper "SM 3.0 really done right"? A year from now we will see the first 80 nm parts, and these will probably span from the budget to the high end. After that, the crystal ball gets really cloudy...
November 7, 2005
NVIDIA Releases 6800 GS for Holidays - Josh
NVIDIA has done a little freshening up of their upper-midrange with today's announcement of the 6800 GS. As you may or may not be aware, NVIDIA originally created the NV41 as their first native PCI-E solution. This chip was used in the first GeForce 6800's for PCI-E, and was produced on IBM's 130 nm process. These chips didn't clock terribly high, but they were solid at the time. NVIDIA then decided to port the design over to TSMC's 110 nm process to save a few dollars, shrink the die size, and get away from IBM's troublesome process. This in turn became the NV42. The 6800 GS is based on the NV42 chip, and having been in production for almost a year it is a very mature product. The NV42 features 8 ROPS, 12 pixel shaders, and 5 vertex shaders.
So what differentiates the 6800 GS from a regular 6800? Well, the 6800 GS runs at 425 MHz core while the regular 6800 usually is clocked around 325 MHz. NVIDIA is also specifying GDDR-3 running at 1 GHz instead of regular DDR running at 700 MHz. It still has a 256 bit bus, so bandwidth isn't an issue. The 6800 GS overall has about the same performance as the older 6800 GT, but there should be some room to overclock this product as well. NVIDIA and its partners are aiming at a $249 price point, which is the same area where ATI is aiming their X1600 XT. In pretty much every review I have seen, the 6800 GS simply destroys the X1600 XT (except in 3D Mark 2005, which is Vertex Shader limited most of the time).
This may not be a long lived product, as early next year many are expecting the 7x00 series to be introduced into the midrange and budget sectors. Still, it looks to be a good product by today's standards, and will certainly find quite a few new homes this holiday season for folks not wanting to spend over $320 for a 7800 GT and above. Buying these in SLI would be somewhat silly though, as a single 7800 GTX offers about the same amount of performance for less than the price of two 6800 GS.
NVIDIA is promising instant availability with these cards from a number of manufacturers. If NVIDIA holds to their promise then these cards should be readily available by the middle of this week.
Here are a few reviews:
So far nobody has tried overclocking the reference board, which is something of a shame. I would have been very interested to see if these cards could hit 500 MHz core (which would really add some value) as well as how the latest GDDR-3 that it is sporting will clock. I wouldn't doubt if a few manufacturers offered models clocked at 450 MHz core and 1050 MHz memory, but time will tell. Still, it is a solid offering for the price point, and most likely we will see products hitting $225 pretty shortly. A good buy if a user is looking to build or upgrade a PCI-E system for not a whole lot of scratch.
November 1, 2005
November a Hot Month - Josh
Though we here in Wyoming don't believe the above statement, it is probably going to be true for the tech industry. The month leading up to the Holiday Season (aka Christmas) usually sees some shakeups and new products introduced. I know of several interesting products coming up, and that is probably just the tip of the iceberg. We can also expect to see the X1800 XT's finally hitting the market, and word is that ATI will have another surprise in store for us (though I have not a clue what it is). The CPU market will stay pretty calm with Intel releasing a few Xeon type chips, but their 65 nm parts won't be showing up until at least the first of the year. AMD looks to introduce their new sockets and DDR-2 support at the first of the year as well, and due to having to redesign the memory controller I am better an entire new revision of the Athlon 64 is in order, and it might be pretty special as well.
So far not too many comments on the GeForce 6600 DDR-2 product, but I must admit that I did enjoy working with that card immensely. I think it will be a great value product and if it actually shows up at the $99 price point, it is going to be a VERY attractive offering for a large group of people. Monetarily it is an easy buy, and what it brings the user over an older video card or integrated video is pretty spectacular. 6200, X300, and X600 users should really take notice. Especially if they are hoping to play some of the latest games at decent rates and resolutions.
I am honestly dreading having to buy new DDR-2 memory for next year, as I have been really spoiled by the high performance DDR that I have essentially been able to hang onto for the past three years or so. I did make the jump to 2 x 1 GB Patriot PC3200 DIMMS for testing (and of course for my main machine), but I can see that in January I am going to have to buckle down and choose some good DIMMS for future testing. Such is life. Still, DDR-2 has really come a long ways, and I think by the beginning of next year we will see some interesting new products (and hopefully AMD will officially support DDR-2 800 speeds).
Here are a few interesting reviews and articles around the web:
The Tech Report takes a look at how important memory timings are to overall system performance. I have always been a fan of low latency DIMMS, but not everyone sees the world the same as I do (probably a good thing).
Chris Tom at AMDZone takes a look at CrossFire performance on Half-Life 2's Lost Coast level. He notices a few nasty things going on in the works. Bring on X1800 CrossFire I say!
October 31, 2005
XFX GeForce 6600 DDR-2 Review - Josh
Last week I received a surprise package from XFX containing the yet to be released GeForce 6600 DDR-2. This new SKU features 256 MB of DDR-2 memory which is aimed to be the new budget performer in the NVIDIA lineup. With the decline of DDR production, many memory manufacturers are switching to DDR-2. XFX is one of the first to bring this budget performer to the market, and it is a surprising product. Here is a quote:
You can read the entire review here.
October 28, 2005
Quake 4 Timedemo Issues - Josh
In doing some testing yesterday I noticed a nasty little flaw with Quake 4. First I created my own demo, one that spanned from inside to outside environments, then I did timedemos of that demo. After watching it a few times I began to notice quite a few issues with the rendering that was occurring.
The first big thing I noticed is that many of the animated textures on screens, guns, etc. were not being rendered at all. These textures were rendered as black spaces. The second thing I noticed is that explosions were not being rendered at all. When using the grenade launcher it looked like I as just shooting cans of pop at the Strogg and they were dying when it hit them. The biggest problem was that going from outside to inside, the lighting effects were also not properly rendered (and lighting is a big part of the Doom 3 engine).
I have tested this out on both ATI and NVIDIA cards with the latest drivers for each, and the results are the same. Demo playback in Quake 4 appears to be currently broken and does not render the scene correctly. This means that as a timedemo, Quake 4 is pretty useless. The only real way to get good results from Quake 4 is to run Fraps and go through an area where the play is pretty consistent from run to run.
October 27, 2005
NVIDIA Releases GeForce 6600 DDR-2 - Josh
In a move aimed to redefine the sub-$150 market, NVIDIA has released a new member of the 6600 family. The 6600 DDR-2 is a budget oriented product that should fit very nicely between the regular 6600 and the 6600 GT. Offering 256 MB of DDR-2 should give it a small advantage over the more highly priced GT version in games that do utilize more than 128 MB of memory. The reference core is set at 350 MHz while the memory runs at 400 MHz (800 MHz effective). This gives it about 12.8 GB/sec of bandwidth, as compared to a regular 6600 which provides around 8.8 GB/sec. Just like other 6600's, it utilizes 8 pixel shader pipelines, 4 ROPS, 3 vertex shaders, and a 128 bit memory interface. Many manufacturers are choosing to clock this product around 400 MHz, so these should give a pretty hefty bang for the buck. NVIDIA had a big winner on their hands when they released the 6600 series last year, and they continue to market and sell these in large quantities.
The move to DDR-2 is not an unexpected one. With the desktop growth of DDR-2 getting bigger, and with the introduction of the Athlon 64 supporting DDR-2 coming early next year, many memory manufacturers are shifting production from regular high speed DDR that is used in graphics to DDR-2. Apparently it is getting harder and harder to get a hold of the regular DDR for graphics that we have essentially taken for granted. GDDR-3 is staying at the high end of the market, but to address the smaller quantities of regular DDR now available, we are going to start seeing the switch to DDR-2 for the low end and midrange markets for graphics.
The 6600 DDR-2 looks to be a solid performer, and most likely they can overclock to near GT levels giving a solid gaming experience to any user buying one. The price point these are aimed at will make it a popular product with system integrators as well as home builders and budget enthusiasts. Two of these go very well together in SLI, and due to the extra memory it might be a better prospect getting two of these as compared to the GT version. A user clocking these at a fairly modest 450 MHz can expect some solid results.
As a member of the 6600 series of cards, full support of HDR, PureVideo, and SM 3.0 are a given. Older titles can easily be played at 1600x1200 with AA/AF enabled, while many of the newer titles will have to be played around 1024x768 (such as Quake 4, Battlefield 2, F.E.A.R, etc.). While not a hardcore enthusiast product, it will appeal to many users due to its solid performance and low price. These should be hitting the market very shortly, as I have a full retail XFX version sitting right here. This product is also aimed to create more competition for ATI's upcoming X1300 and X1600 products, both of which have yet to be seen as well on the mass market.
X1800 Availability and Competition for the High End
ATI has done fairly well getting the X1800 XL out to the public, it only took about a week after launch for the initial retailers to have some stock. Prices were pretty high at that time, but we have seen some SKU's go down dramatically in price so that OEM versions of the card are retailing just slightly above that of the competing 7800 GT. We still have not seen or heard tell when exactly we can expect the X1800 XT's to hit in force, but many expect to see it by November 15. From all the rumors swirling about, it looks like NVIDIA will try to trump ATI with a 512 MB version of the 7800 series. We can expect instant availability again, but the kicker is that we will probably see it before ATI can get good quantities of the X1800 XT out the door. Not only that, but I think we will see something quite special from these cards.
October 19, 2005
Industry News - Josh
NVIDIA announced that they have secured an exclusive deal in bundling Call of Duty 2. So far BFG, XFX, and EVGA have bundled CoD2 with their 7800 GT series of cards from quite a few major etailers. This should be a pretty good title (I enjoyed the demo), but who has time to play games anymore? Obviously someone does (but not me).
Intel posted record results as well this quarter with $9.96 billion in revenues and a whopping $2 billion in net income. It just gives you a bit of perspective with AMD posting $1.5 billion in revenues with $70 million in net income. AMD has a big hill to climb to get even 30% of the market from Intel, especially considering the resources Intel has at its disposal. Still, Intel has a wider product base than AMD does, as they supply a lot of wireless, network, chipset, and flash products.
Many are saying that DDR-2 will gain a lot of acceptance in the first half of 2006 due to Intel's i945 series of chipsets, as well as AMD's jump to DDR-2 support. I feel that this is most likely true, as DDR-2 is finally becoming a very solid product with good manufacturing. DDR-2 has had a rough ride so far, though it was introduced by Intel a year and a half ago. As you all remember, AMD was a huge proponent of DDR while Intel went the RAMBUS way. Most memory manufacturers preferred the DDR route as it was easier to produce and showed significant performance advantages over RAMBUS at that time. AMD has stuck with DDR as the onboard memory controller for the Athlon 64 was not as hungry for bandwidth as the competing Intel Pentium 4. Now AMD finally sees the time where they will get a boost from DDR-2 rather than trying to push DDR higher. Though unofficial DDR speeds are reaching 300 MHz (600 MHz DDR or PC4800), there is little interest to push JEDEC to get a higher spec. By the time AMD will transition to DDR-2, I believe that they will push the DDR-2 800 as the preferred memory speed. DDR-2 667 isn't so bad, but it isn't necessarily better than fast DDR PC3200. Still, it looks as though we might be seeing a new generation of DDR-2 chips coming out early next year that will make faster DDR-2 a sure thing. I still am not looking forward to investing in new DDR-2 memory though for testing...
October 14, 2005
AMD Opens Fab 36 - Josh
Today AMD announced that it has officially opened Fab 36 in Dresden, Germany. With over $300 million US in funding from the local and Federal German Governments, plus some very nice loans with great terms, AMD broke ground on Fab 36 some 24 months ago. Previous announcements from AMD about Fab 36 were a bit misleading, especially those quoting that it will be the starting point of 65 nm production. Many were assuming (I being one of them) that when Fab 36 would open it would be producing 65 nm parts right off the bat. This is not the case. 90 nm is the word of the day for Fab 36, until AMD can convert it to 65 nm sometime in the 2nd half of 2006. This means no 65 nm parts until at least early Winter.
This changes around the dynamics of the industry quite a bit. I am guessing that the first 90 nm products off the line at Fab 36 (conveniently located right next door to Fab 30) will be the new revision of the Athlon 64 core which will sport the new DDR-2 memory controller. You can bet that AMD did some other changes to the design, but it appears as though Fab 30 will continue to produce the current generation of products. This is merely speculation on my part, but it appears to be fairly logical. Why change Fab 30 to the new core technology when the new fab next door can handle it? Eventually AMD will ramp down DDR based parts at Fab 30 and produce the DDR-2 based cores there as well when demand increases for the new parts.
This does give Intel a window of opportunity with their new parts coming in 2H 06 based on the Conroe cores. If Conroe is everything that Intel hopes it will be, it should surpass the Athlon 64 products in overall performance and power/heat. These should have a fairly quick introduction due to the continued use of the Pentium 4 level FSB. Most likely Intel will need a new chipset to support this processor, but with the rest of the infrastructure remaining the same, it is not a big deal. AMD will still be competitive at this time mainly due to their ability to clock their current Athlon 64's up to 2.8 GHz (and in the case of the FX-59 up to 3.0 GHz). We can probably deduce that AMD will at least release a 2.6 GHz dual core product based on DDR-2 technology. This should help to keep things fairly competitive until AMD can release their 65 nm parts.
All in all, the opening of the massive Fab 36 for AMD is a great event for a company its size. The management and employees have worked very hard together to make all the parts come together, and they have done it on time and on budget.
Just as a note, I have received word from several AMD employees that have spoken how they were benefactors of the bonus. So, it was not just upper management that got the money, but essentially every employee of AMD.
October 12, 2005
AMD Q3-2005 Results and Why Some Analysts May Just be Plain Wrong - Josh
Yet again, what started out as a front page posting turned into an article that was much larger than should be handled here. I took a look at the latest earnings from AMD, as well as the near future of its products and compare it to what Intel has to offer. I also took a shot at the UBS analyst who said that Intel is poised to leapfrog AMD with their next generation of products. While Intel did release their dual core Xeons based on 2 x Cedar Mill Pentium 4 cores (technically the core is called Paxville), as we can see from many benchmark results with the current crop of dual core Pentium D's, these will not have any real performance advantage over AMD's current dual core Opterons. What Intel's gains from Paxville is a lower power dual core product that is easier to produce since it can use two separate dies on a substrate (better yields and bins vs. two large cores on one die). But how exactly does this technology leapfrog AMD, much less what the next generation parts will bring for both Intel and AMD? I try to cover this as best I can, and I hope the results look good! Here is a 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 1999-2005 PenStar Systems, LLC.