NVIDIA's First 55nm GPU: GeForce 9800 GTX+ Preview
by Anand Lal Shimpi & Derek Wilson on June 24, 2008 12:00 AM EST- Posted in
- GPUs
A very smart man at Intel once told me that when designing a microprocessor you can either build a new architecture, or move to a smaller manufacturing process, but you don't do both at the same time. The reason you don't do both is because it significantly complicates the design, validation and manufacturing processes - you want to instead limit the number of variables you're changing in order to guarantee a quick ramp up and good yields of your silicon.
NVIDIA followed this rule of thumb with the GT200, building its "brand new" (or at least significantly evolved) architecture on a tried-and-true 65nm process instead of starting at 55nm. Despite AMD building both RV670 and the new RV770 GPU on TSMC's 55nm process, NVIDIA hadn't built anything on a smaller than 65nm process, including the 1.4 billion transistor GT200.
Shortly after the GT200 launched, AMD "responded" with its Radeon HD 4850, a cheap card by comparison, but a far more interesting one from a practical performance standpoint. Priced at $199 and selling for as little as $170, the Radeon HD 4850 managed to invalidate most of NVIDIA's product line. In response, NVIDIA dropped the price of its GeForce 9800 GTX to $199 as well and introduced one more card: a $229 GeForce 9800 GTX+.
Originally we thought the GTX+ was a silly last minute afterthought as it looked like nothing more than an overclocked 9800 GTX. While its clock speeds are higher, it also happens to be the very first 55nm NVIDIA GPU. The specs are as follows:
| 9800 GTX+ | 9800 GTX | |
| Stream Processors | 128 | 128 |
| Texture Address / Filtering | 64 / 64 | 64 / 64 |
| ROPs | 16 | 16 |
| Core Clock | 738MHz | 675MHz |
| Shader Clock | 1836MHz | 1690MHz |
| Memory Clock | 1100MHz | 1100MHz |
| Memory Bus Width | 256-bit | 256-bit |
| Frame Buffer | 512MB | 512MB |
| Transistor Count | 754M | 754M |
| Manufacturing Process | TSMC 55nm | TSMC 65nm |
| Price Point | $229 | $199 |
The core clock went up 9.3%, shader clock went up 8.6% and memory clock stayed the same. The clock speed bumps are marginal and by far the more interesting aspect of the chip is how much less power it consumes thanks to its 55nm process, which thanks to AMD should be quite mature by now.
Here's the full NVIDIA lineup:
| GTX 280 | GTX 260 | 9800 GX2 | 9800 GTX+ | 9800 GTX | 8800 GTS 512 | 8800 GT | |
| Stream Processors | 240 | 192 | 256 | 128 | 128 | 128 | 112 |
| Texture Address / Filtering | 80 / 80 | 64 / 64 | 128 / 128 | 64 / 64 | 64 / 64 | 56 / 56 | 56 / 56 |
| ROPs | 32 | 28 | 32 | 16 | 16 | 16 | 16 |
| Core Clock | 602MHz | 576MHz | 600MHz | 738MHz | 675MHz | 650MHz | 600MHz |
| Shader Clock | 1296MHz | 1242MHz | 1500MHz | 1836MHz | 1690MHz | 1625MHz | 1500MHz |
| Memory Clock | 1107MHz | 999MHz | 1000MHz | 1100MHz | 1100MHz | 970MHz | 900MHz |
| Memory Bus Width | 512-bit | 448-bit | 256-bit x 2 | 256-bit | 256-bit | 256-bit | 256-bit |
| Frame Buffer | 1GB | 896MB | 1GB | 512MB | 512MB | 512MB | 512MB |
| Transistor Count | 1.4B | 1.4B | 1.5B | 754M | 754M | 754M | 754M |
| Manufacturing Process | TSMC 65nm | TSMC 65nm | TSMC 65nm | TSMC 55nm | TSMC 65nm | TSMC 65nm | TSMC 65nm |
| Price Point | $650 | $400 | $500 | $229 | $199 | $280 | $170 - $230 |
Notice something very wrong? The 8800 GTS 512 and 8800 GT both need to drop in price significantly, they are simply uncompetitive at their current price points. I expect one of those two products to go the way of the dodo but it's unclear which one; the 8800 GT is cheaper to make, but perhaps it's easier to produce 65nm parts with 128 SPs so the GTS 512 could stick around at a lower price point as well.
The GeForce 9800 GTX+ will be available starting July 16th.
Facebook
Twitter
RSS
36 Comments
View All Comments
Ytterbium - Tuesday, June 24, 2008 - link
Hi, What are the setting you use in Crysis?Pale Rider - Tuesday, June 24, 2008 - link
Like one full page of new ATi product information smack in the middle of this preview.ashegam - Tuesday, June 24, 2008 - link
Anand or Derek,When can we expect 55nm 200 models since they are apprehensive about doing both a die shrink and new architecture together? 6 month, 1 year?
Anand Lal Shimpi - Tuesday, June 24, 2008 - link
NVIDIA probably started working on the 55nm shrink of GT200 as soon as the chip was done, so you can expect a die-shrunk version of it as soon as 6 months but I'd expect one in early 2009.-A
Mr Roboto - Tuesday, June 24, 2008 - link
Yeah how weird is that? Two companies that are such bitter rivals both use the same manufacturer. I didn't know they both used TSMC until a few weeks ago.Also I know it's just a name but I think Nvidia is retarded for their naming scheme the last few rounds. First the whole 8800GTS 640\320 512\256 then the 8800GSO\9600GSO which is basically a 9600GT and now the 9800GTX+. Really unimaginative and it just shows how much they're trying to capitalize on the 8800 name. They need to fire the guy who jumbled this thing so badly. You would think Jen-Hsun Huang would be on top of things like this. Unless this was his idea?
Anyways my 8800GTX is still kickin strong, but I'm looking forward to seeing the rest of ATI's lineup.
Clauzii - Tuesday, June 24, 2008 - link
In my world, 9800 will always be as in "ATI Radeon 9800Pro" :))feelingshorter - Tuesday, June 24, 2008 - link
Your right and a lot of people would agree. Nvidia's naming is so confusing that most people who aren't so tech savy wouldn't know if a GSO or the GS or the GTS or GT is faster. Not to mention they released the cards with different types of memory (512 vs 640) under the same name with the higher memory being slower, older version of the same card. How many people probably mistakenly bought the 640 version of the 8800GT without knowning the 512 was actually faster? Not everyone reads tech news daily.Only the less tech savy customers are the ones hurt, which is most of the customers. Although no one on anandtech would get confused, I know plenty of friends who are and ask me all the time which cards to buy. But even I have to read tech news daily just to keep up with it. This is, YANL, as one review website coined the term (yet another nvidia launch).
strikeback03 - Wednesday, June 25, 2008 - link
For that matter, the 9600GSO offerings seem to generally have less memory on a smaller memory interface, but more SPs. So maybe NVIDIA is confused over which should be faster as well.And why has Anandtech never reviewed a GSO?
formulav8 - Friday, June 27, 2008 - link
The 9600 GSO is basically the 8800GS video card renamed. Same 192 Bit memory interface and 384MB of memory. I believe 96 Shaders as well.The 9600 GT version has 64 Shaders but the clockspeeds are higher than the GS clocks and carries a 256 Bit memory bus.
Whats sad is the GSO version could beat the GT version in certain things and the GT wins in other things. They are not clearly defined as they should be since they are carrying the same Model #. nVideo has the worse naming so far.
So if you've seen the 8800GS review then you've seen the 9600 GSO reviewed...
Jason
Lonyo - Tuesday, June 24, 2008 - link
Are the Oblivion numbers in the chart really accurate?You're trying to tell me the 9800GTX+ with 9.3% better core and 7.x% better shaders and the same speed RAM gets 15% better FPS in Oblivion?
Seems a little odd, unless there are some other changes under the hood, because I don't see how a <10% change in clocks can get you 15% change in performance without something a bit weird going on.