Budget 90nm Graphics: NVIDIA GeForce 7300
by Derek Wilson on January 18, 2006 9:00 AM EST- Posted in
- GPUs
What's Under the Hood?
As the title of the article implies, the GeForce 7300 is NVIDIA's very first shipping 90nm part. In order to come out with the 7800 series as early as they did, NVIDIA didn't want to try anything tricky like pushing the process envelope and moving to 90nm. Sticking with the 110nm process proved to be a good decision for NVIDIA this time around. With TSMC bringing out their 80nm process and NVIDIA just now starting to manufacture 90nm parts, it will be interesting to see who takes the next step towards smaller transistors. ATI has been pretty aggressive recently, but the 80nm process should be a fairly simple transition (as far as fabrication process transitions go), so it could go either way.
Aside from transistors with 90nm gate lengths and FP16 framebuffer blending, the 7300 will also carry NVIDIA's TurboCache technology. With relatively few vertex shaders, pixel shaders, and render outputs, a little die area can afford to be spent on beefing up caches to allow for the efficient use of higher latency system memory. This allows NVIDIA to "support" games that require more RAM than is physically present on the card. With the GeForce 7300 sporting 128MB of RAM, TurboCache allows games to treat the graphics subsystem like it has 256MB of RAM. Likewise, the 256MB version of the 7300 will look like it has a 512MB framebuffer to software.
In addition to allowing programs that require more graphics memory to run efficiently, NVIDIA is also able to cut the bandwidth to graphics memory down without taking a large performance hit. The bandwidth gained through TurboCache augments the bandwidth of onboard memory and should provide some good speed advantages for low end parts. The 128MB version of the 7300 will have a 64-bit memory bus, which is actually on the high end for TurboCache parts. It is likely that the extra RAM and onboard memory bandwidth boost in this generation of TurboCache parts is there to enable the use of memory intensive HDR features (like the FP16 blend absent from 6200 parts).
The rest of the NV4x/G70 features will also come along with this new budget part without exception. This includes NVIDIA's PureVideo features for clean DVD viewing and efficient playback of HD content, and the possibility of SLI in the future. SLI is not part of the 7300 series at launch, but, if NVIDIA's track record is any indication, we can all but guarantee SLI support. Shader Model 3 is of course supported, but with fewer pipelines, the efficiency of high end features like looping and conditional rendering will not be optimal. We should also see more efficient handling of math intensive shaders due to the 7 series ability to handle more Multiply-Adds per clock than the 6 series. Here's the full breakdown of what the GeForce 7300 brings to the table.
Clearly, the 7300 will be able to win out over the 6200 TC in any flavor, but we will need to wait until we get our hands on a part to know exactly what flavors of X1300 it will match up against. Listed in our table is our best guess match (the vanilla X1300), but we'll try to test as many flavors as we can when we are able to do a performance comparison.
For now, we can at least take comfort in the fact that a >50% increase in core clock speed, increased memory size and bandwidth, and higher level of architectural efficiency over the 6200 TC will help make the new GeForce 7300 very competitive at a <$100 price point. While it may not get everyone's blood boiling, the 7300 is not just an exciting part for the budget market, but it will help raise the bar for the minimum target that game developers will be shooting for over the next couple of years.
As the title of the article implies, the GeForce 7300 is NVIDIA's very first shipping 90nm part. In order to come out with the 7800 series as early as they did, NVIDIA didn't want to try anything tricky like pushing the process envelope and moving to 90nm. Sticking with the 110nm process proved to be a good decision for NVIDIA this time around. With TSMC bringing out their 80nm process and NVIDIA just now starting to manufacture 90nm parts, it will be interesting to see who takes the next step towards smaller transistors. ATI has been pretty aggressive recently, but the 80nm process should be a fairly simple transition (as far as fabrication process transitions go), so it could go either way.
Aside from transistors with 90nm gate lengths and FP16 framebuffer blending, the 7300 will also carry NVIDIA's TurboCache technology. With relatively few vertex shaders, pixel shaders, and render outputs, a little die area can afford to be spent on beefing up caches to allow for the efficient use of higher latency system memory. This allows NVIDIA to "support" games that require more RAM than is physically present on the card. With the GeForce 7300 sporting 128MB of RAM, TurboCache allows games to treat the graphics subsystem like it has 256MB of RAM. Likewise, the 256MB version of the 7300 will look like it has a 512MB framebuffer to software.
In addition to allowing programs that require more graphics memory to run efficiently, NVIDIA is also able to cut the bandwidth to graphics memory down without taking a large performance hit. The bandwidth gained through TurboCache augments the bandwidth of onboard memory and should provide some good speed advantages for low end parts. The 128MB version of the 7300 will have a 64-bit memory bus, which is actually on the high end for TurboCache parts. It is likely that the extra RAM and onboard memory bandwidth boost in this generation of TurboCache parts is there to enable the use of memory intensive HDR features (like the FP16 blend absent from 6200 parts).
The rest of the NV4x/G70 features will also come along with this new budget part without exception. This includes NVIDIA's PureVideo features for clean DVD viewing and efficient playback of HD content, and the possibility of SLI in the future. SLI is not part of the 7300 series at launch, but, if NVIDIA's track record is any indication, we can all but guarantee SLI support. Shader Model 3 is of course supported, but with fewer pipelines, the efficiency of high end features like looping and conditional rendering will not be optimal. We should also see more efficient handling of math intensive shaders due to the 7 series ability to handle more Multiply-Adds per clock than the 6 series. Here's the full breakdown of what the GeForce 7300 brings to the table.
| Budget Card Features | |||
| Radeon X1300 | GeForce 6200TC | GeForce 7300 | |
| Vertex Pipelines | 2 | 3 | 3 |
| Pixel Pipelines | 4 | 4 | 4 |
| Render Outputs | 4 | 2 | 2 |
| Core Clock Speed | 450 | 350 | 550 |
| Memory Clock Speed | 500 | 350 | 350 |
| Memory Size | 256-512MB | 16-64MB | 128-256MB |
| Memory Bandwidth | 64-128bit | 32-64bit | 64bit |
| GPU Video Decode | yes | yes | yes |
| FP16 Filter | no | yes | yes |
| FP16 Blend | yes | no | yes |
Clearly, the 7300 will be able to win out over the 6200 TC in any flavor, but we will need to wait until we get our hands on a part to know exactly what flavors of X1300 it will match up against. Listed in our table is our best guess match (the vanilla X1300), but we'll try to test as many flavors as we can when we are able to do a performance comparison.
For now, we can at least take comfort in the fact that a >50% increase in core clock speed, increased memory size and bandwidth, and higher level of architectural efficiency over the 6200 TC will help make the new GeForce 7300 very competitive at a <$100 price point. While it may not get everyone's blood boiling, the 7300 is not just an exciting part for the budget market, but it will help raise the bar for the minimum target that game developers will be shooting for over the next couple of years.
Facebook
Twitter
RSS
24 Comments
View All Comments
DerekWilson - Wednesday, January 18, 2006 - link
NVIDIA products hardlaunched in the past year and a half:Go 6800 Ultra (yes, it all started with a notebook launch)
GeForce 7800 GTX
GeForce 7800 GT
Go 7800
GeForce 6800 GS
GeForce 7800 GTX 512 was available on launch day in limited quantities and sold out fast ... so we'll count it as half a hard launch ...
... and that's just off the top of my head. There are probably others that slipped my mind.
DerekWilson - Wednesday, January 18, 2006 - link
Yes, worldwide hardware launches are more difficult than software launches ... But thanks for sticking up for us :-)As I said, we don't currently expect worldwide availability. That's just the direction we are headed.
peldor - Thursday, January 19, 2006 - link
I don't think we're headed toward hard launches for everything. Financially it's not that great for the companies. They have to build up supplies for weeks or months, which leaves money tied up in products in a warehouse. Or the manufacturing facilities have to be built to over capacity which makes even less sense. On the high-end, low-volume parts this should be easier, but on the low-end, not so much.Market timing will continue to be a huge part of the decision to paper launch vs hard launch.
Phiro - Wednesday, January 18, 2006 - link
Diablo: II had a simultaneous worldwide launch.Star Wars: Episode 3 "Revenge Of The Sith" had a simultaneous worldwide launch.
Fantastic Four had a simultaneous worldwide launch.
Phiro - Wednesday, January 18, 2006 - link
Troy was simultaneous. The Frozen Throne expasion for WC3 was simultaneous. The Pretendo Revolution is planned on being simultaneous.Bobby Fisher's new chess game was simultaneous.
Red Alert 2 was simultaneous.
Steam releases from Valve are simultaneous.
Apple's Tiger release was simultaneous.
Phiro - Wednesday, January 18, 2006 - link
ATI maintains to this day that the 9500 & 9700 release and then three months later the PRO release were both simultaneous worldwide launches.ViRGE - Wednesday, January 18, 2006 - link
Without forcing manufacturers to hard launch, they're going to pull shit like the 6800 Ultra Extreme, the X1800XT, and other such launches where they'll announce a product just to steal another company's thunder, and then potentially never release it, fail to release it in decent quantities, or announce a product so far out that we start talking about things that are still a gleam in the eyes of a VHDL. The only solution to this problem is to pressure companies in to hard launching, as they can not commit these abuses if they actually need to have the product ready to go. Gamers deserve real products with real specifications and real benchmarks, so that they know what to buy; soft launches are nothing more than a method of spreading FUD to hurt one's opponent and eventually the gamers themselves.ElFenix - Wednesday, January 18, 2006 - link
no, that's vaporware. not 'soft launches.' the problem isn't that the product comes out a week or two later in large quantities, it's that it never comes out (ultra extreme) or comes out in such a trickle that it isn't available anywhere unless you're really lucky (xt pe, gtx 512). the beef was and always has been with vaporware. yes, the 'hard launch' is a way to fix it, but it brings its own downside.Jep4444 - Wednesday, January 18, 2006 - link
doesnt even solve the problem, the GTX512 was hard launched and its still nearly impossible to findOrSin - Wednesday, January 18, 2006 - link
On board video for ATI Xpress and NV 6100's are more theng ood enought for most business applications and family video games (Wheel of fortune type). Thie card has no place. Most real games this card can't do anything.This quote "Because many game developers write software for the least common denominator" is just not true. Most of the huge selling "real" game of last year, you needed to upgrade your card to play. I say real because I'm not including game like the Sims. But even if include the market for games like the Sims, this card is once again not need when on board video will do. If you not paying $150 and up for card you might was well stick with on board video.