NVIDIA GeForce 7800 GT: Rounding Out The High End
by Derek Wilson & Josh Venning on August 11, 2005 12:15 PM EST- Posted in
- GPUs
The Card, Specs and Test
The basic setup of the GeForce 7800 GT is the same as the GTX. We still have one single and one dual-link DVI port and a 6-pin power connector on a x16 PCI Express card. The stock HSF solution is the same, and the silicon hasn't changed either. The GT is still based on the G70 with a few blocks disabled.The G70 on the 7800 GT features 7 vertex pipelines, 20 pixel pipelines and lower clock speeds (400/500 core/mem) than the GTX. This still ends up putting the 7800 GT above the 6800 Ultra in terms of vertex and pixel performance even though it lags in core and mem speed. Here's a quick table of the differences between the current high end NVIDIA parts.
| Card Comparison | |||
| 6800 Ultra | 7800 GT | 7800 GTX | |
| Vertex Pipes | 6 | 7 | 8 |
| Pixel Pipes | 16 | 20 | 24 |
| ROP Pipes | 16 | 16 | 16 |
| Core Clock | 425 | 400 | 430 |
| Mem Clock | 550 | 500 | 600 |
| MSRP | $399 | $449 | $599 |
The 7800 GT comes in the lowest on clock speed and memory bandwidth, but the combination of pipelines and clock speed give the 7 series GT a definite theoretical advantage over the 6800 Ultra in the vertex and pixel processing department. Obviously, the GTX has all the rest beat, but the real question is: how much real world difference will it make? We've already seen that the 7800 GTX performs very well compared to the 6800 Ultra, and, ideally, we would like to see the GT split the numbers. That would leave all three cards with viable performance characteristics.
Price is a very important matter in viability as well, and it is very difficult to tell where the chips will fall. As we can already see with the 7800 GTX, a six hundred dollar suggested price has resulted in street prices mostly between five hundred and five hundred fifty dollars. We don't expect the price of the GT to fall one hundred dollars as quickly as the GTX, but the prices that we see on our Real Time Price Engine show GT products available starting around $400.
Even though NVIDIA haven't recently adjusted their suggested price on 6800 Ultra cards, if the prices don't fall, there will really be no reason to buy a 6800 Ultra (unless it's desperately needed for an SLI update). The suggested retail prices on the 6800 GT and vanilla 6800 have been changed to $299 and $199 respectively. The new pricing layout does give the 6800 Ultra some ability to move down in price a little, if the 7800 GT outperforms it as we expect. It is also possible that the similarity in the price between the 6800 Ultra and the 7800 GT could mean that the 6800 Ultra is being phased out.
Aside from the rather less than spectacular differences between the GT and GTX, we still need to learn about performance. Potentially, this card could be a good buy depending on the numbers that we see. If performance is near that of the GTX, we could save money and still have a very powerful card. If we see numbers near the 6800 Ultra, then we are likely looking at a simple replacement for the 6800 Ultra in NVIDIA's lineup.
The test setup that we used for these performance numbers is the same as the one used in the last few graphics performance articles that we've done:
MSI K8N Neo4 Platinum/SLI motherboard
AMD Athlon 64 FX-55 Processor
1 GB OCZ 2:2:2:6 DDR400 RAM
Seagate 7200.7 120 GB Hard Drive
OCZ 600 W PowerStream Power Supply
We used the NVIDIA 77.77 ForceWare beta driver running on Windows XP SP2.
Facebook
Twitter
RSS
77 Comments
View All Comments
IdBuRnS - Tuesday, November 29, 2005 - link
I just placed an order for a eVGA 7800GT with their free SLI motherboard.This will be replacing my ATI X800 Pro and MSI K8N Neo2 Platinum.
A554SS1N - Tuesday, September 6, 2005 - link
Ok, now for my views on the 7600GT...The 6600GT had exactly half the pixel-pipelines and memory bus of the 6800GT/Ultra, and this makes me think the 7600GT will be the same in relation to the 7800GTX. To add weight to this theory, a 128-bit would be much cheaper to produce, and with a smaller die size, be more economical and cooler. By being cooler, smaller fans can be used, saving more money. Also, NVidia would probably want to keep the PCB smaller for mainstream components (something that I would like myself).
So basically, my suggestion is that the 7600GT would be a 12-pipe, 128-bit card probably with those 12 pipes matched to 8 ROP's (like the 6600GT was 8-pipe matched to 4 ROP's). Around 5/6 Vertex pipelines would sound about right too. If teh core were at 450Mhz or even 500Mhz with 12-pipes, and paired with 1100Mhz memory, it would likely turn out somewhere inbetween a 6800 and 6800GT in performance, but importantly, would be:
- Cooler
- Potentially Quieter
- More energy efficient
- Smaller PCB
- Potentially cheaper
- More easy to produce, therefore able to provide lots of cores to the mass market
Just my opinions, but I can believe in a 12 pipe card more than 16 pipe mainstream card which I consider to be a "pipe dream".
Pythias - Friday, August 12, 2005 - link
You know its time to quit gaming when you have to have a card that costs a much as a house payment and a psu that could power a small city to run it.smn198 - Friday, August 12, 2005 - link
Marketing
DerekWilson - Friday, August 12, 2005 - link
We stand corrected ... After reading the comments on this article it is abundantly clear that your suggestion would be a compelling reason to release lower end G70 parts.Evan Lieb - Friday, August 12, 2005 - link
I think maybe some of you are taking the article a little too seriously. Most hardware articles nowadays are geared toward high end tech for good reason; it's interesting technology and a lot of people want to read about it. It's useful information to a lot of people, and a lot of people are willing to pay for it. You want entry level and mid range video reviewed too? That's fine, but you'll have to wait like everyone else, AT can't force NVIDIA to push out their 7xxx entry level/mid range tech any faster. When it's ready you'll probably see some type of review or roundup.Regs - Thursday, August 11, 2005 - link
Well there is still no AGP for 939 AGP owners and the performance difference between the Ultra and GT this year is a lot more significant from last years. I would hate to spend 500 dollars on a "crippled" 7800 GTX. Not to mention ATI is still a bench warmer in this competition. Just seems like upgrading this year is not even worth it to a 939 AGP owner no matter how much of a gamer you are. I'm disappointed in the selection this year. Performance is there, but the price/value and inconvenience is above and beyond. Last year was a great time to upgrade, while this year seems more like a money pit with no games to fill it over.bob661 - Friday, August 12, 2005 - link
Next year is probably a better time to upgrade for the AGP owners, I agree. For me, I want a 7600GT. If there will be no such animal then maybe a 7800GT at Xmas.dwalton - Thursday, August 11, 2005 - link
I intially agreed with that statement until I thought about 90nm parts. Correct me if I am wrong but Nvidia has no 90nm parts.While nvidia current line of 7xxx and 6xxx provide a broad range of performance. I'm sure nvidia can increase profit margins by producing 90nm parts.
Nvidia can simply take the 6800 GT and Ultra 90nm chips and rebadge them the 7600 vanilla and GT. Since this involves a simple process shrink and no tweaking, these new 90nm can possibly be clocked higher and draw less power while increasing profit margins, without the cost of designing new 7600 chips based off the G70 design. Making everyone happy.
coldpower27 - Friday, August 12, 2005 - link
I would like G70 technology on 90nm ASAP, I have a feeling Nvidia didn't do a shift to 90nm for NV40 for a reason, as that core is still based on AGP technology, and Nvidia currently doesn't have a native PCI-E part for 6800 Line, they are all using HSI on the GPU substrate from the NV45 design.NV40 on 0.13 micron is 287mm2 as pointed out by a previous poster, a full optical node shrink from 0.13 micron to 0.09 micron without any changes whatsoever, would bring NV40 287mm2 die size to ~ 172mm2 as full node optical shrink generally gives a die size of around 60%. This die size may not be enough to maintain a 256Bit Memory Interface,
Hence why Nvidia is rumored to do only a 0.11 micron process shrink (NV48) on the NV40 as that would bring a core down to about 230mm2 which is 80% of the size. Still large enuogh to maintain the 256Bit Memory Interface with little problem.
Making a 90nm G7x part for the mainstream segement directly would be very nice.
Let's say it has 16 Pipelines, and 8 ROP's to help save transistor space, plus the enhanced HDR buffers, and Transparency AA. It would be fairly close to the range I believe of 170mm2, it would probably still be limited at 128Bit Memory Interface, but the use of GDDR3 1.6ns @ 600MHZ could help alleviate the bandwidth problems some. Remember large amounts of memory bandiwdth combined with high fillrate is reserved for the higher segements, very hard to have your cake and eat it too in the mianstream.
Let's faice it for the time being, were not going to be getting fully fucntional high end cores at the 199US price point with 256Bit Memory Interface, so far we have gotten things like Radeon X800, Geforce 6800, 6800 LE, X800 SE, X800 GT. Etc etc. It just doesn't seem profitable to do so.
From what we have seen mianstream parts based on the tweaked technology are usually seen, RV410 Radeon X700, NV36 Geforce FX 5700 are mainstream cores based on the third and second generation of R300 and NV30 technology.
The 6800 @ 199US, 6800 GT @ 299US, 6800 U @ 399US is a temporay measure and production should slow on these cards as Nvidia ramps up the 90nm G7x based parts.