nVidia GT200 Series Review Thread

[ChaosDivine]

Guess I'll continue waiting patiently on the sidelines with my trusty 8800GTS 320MB (only game at 1280x1024).

 

[Extelleron]

We shall see. GT200, unlike RV770, was very close to a true doubling of G92 with all major vitals. It fell short with TMUs and clockspeeds and as a result, its performance isn't quite 2x that of G92 GTX or even GX2. RV770 isn't a true doubling of RV670 so your predictions are incredibly optimistic imo. All the early indicators and leaks have pointed at 4850 being equivalent to a G92 GTX at best and the 4870 being 20-30% faster.

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

Let's look at HD 3870 vs HD 4870

Shading Performance:
HD 3870 = 320 * .775 *2 = 496
HD 4870 = 800 * .750 *2 = 1200
HD 4870 = 2.42x HD 3870

Texture Performance:
HD 3870 = 16 * .775 = 12.4
HD 4870 = 32 * .750 = 24.0
HD 4870 = 1.94x HD 3870

Pixel Performance
HD 3870 = 16 * .775 = 12.4
HD 4870 = 16 * .750 = 12.0
HD 4870 = 0.97x HD 3870

Memory Bandwidth
HD 3870 = 2250 * .032 = 72 GB/s
HD 4870 = 3600 * .032 = 115.2 GB/s
HD 4870 = 1.60x HD 3870

Comparing GTX 280 vs. 8800 Ultra

Shading Performance:
GTX 280 = 240 * 3 * 1.296 = 933.1
G80 Ultra = 128 * 3 * 1.500 = 576
GTX 280 = 1.62x G80 Ultra

Texture Performance
GTX 280 = 80 * .602 = 48.16
G80 Ultra = 64 * .612 = 39.168
GTX 280 = 1.23x G80 Ultra

Pixel Performance
GTX 280 = 32 * .602 = 19.264
G80 Ultra = 24 * .612 = 14.688
GTX 280 = 1.31x G80 Ultra

Memory Bandwidth
GTX 280 = 2200 * . 064 = 140.8 GB/s
G80 Ultra = 2160 * .048 = 103.68 GB/s
GTX 280 = 1.36x G80 Ultra

There you have it. The HD 4870 is a much larger improvement over the previous generation than GTX 280. The differences are not even close; the GTX 280 looks like a meager upgrade to G80 in comparison to RV770 vs. RV670. The only area where GT200 is a bigger upgrade over the previous generation is in pixel performance (ROPs). This is not an important factor in performance at the resolutions RV770 parts will target and RV670 already has more pixel performance than G92 based parts.

 

[Martimus]

Originally posted by: Extelleron
I would consider the GTX 280 to be mildly preferrable to the 9800 GX2 if the two cards were the same price. The 9800 GX2 is a bit faster but the GTX 280 is better with AA/AF, uses less power, and doesn't rely on SLI. But the problem is the GTX 280 costs ~$200 more than the 9800 GX2. I can get a 9800 GX2 right now for $450: http://www.newegg.com/Product/...x?Item=N82E16814143128

Actually, the Anandtech review showed that it used more power than the 9800GX2 (313W at load vs. 289W). In fact, they couldn't even run two GTX 280's in SLI with any PSU they had on hand. They had to jury-rig a second PSU to actually run benchmarks with the SLI's system.

 

[trajan2050]

G280 will just get better and better. For a high end monitor this is the only way to go.

 

[chizow]

Originally posted by: flexy
And is it really the case that SLI/GX2 is "such a problematic issue"..so it would be worth going for single GPU instead of GX2/SLI? Wouldnt all recent games support the current SLI/GX2? I want value....but i dont see it with this card.

I guess it would be good to define some of the problems no one talks about when touting SLI/CF at the high-end. Here's a few with a brief discussion, but by no means is it comprehensive.

• Profiles/Scaling- SLI/CF rely on driver profiles for their performance and in the case of ATI, you can't change these yourself. So if your particular game doesn't have a pre-defined profile you may see no benefit or even *worst* performance than with a single card. In the case of relying on two individually slower cards than your single card, you can see that you may actually be paying more for *worst* performance which is unacceptable to me.
  
• Micro-stuttering- Pretty heated debate about the significance of this problem on this board and others although it pops up infrequently. Basically the timing of each frame from the different GPU in AFR can be erratic, leading to this effect. Apparently some people are very sensitive to it and some aren't. I don't know as I have never used SLI, but I certainly wouldn't be happy if I spent $400-600 for SLI/CF only to find I couldn't stand micro-stutter.
  
• Heat/Power/Space - Typically not an issue for most enthusiasts, but it can become a problem when you have 2 or even 3x the power draw and heat from high-end cards. The PSU issue can be a total W issue, but also a power connector issue with so many high-end parts needing 6 or even 8-pin PCI-E connections. Many cases can also have problems accomodating 1x9"+ card, much less 2+.
  
• Multi-Monitor (NV only) - NV multi-GPU solutions do not support multi-monitors. I don't know if this is a superficial driver limitation to prevent desktop cards being used in professional workstations or a truly technical issue, but I'm leaning towards driver limitation as I'm assuming the Quadro GX2 would support more than 1 monitor..... Multi-Monitor support is important to me as I play full screen on my 1920 and use my 2nd monitor for various monitoring tools, surfing the web, etc.
  
• Bandwidth/Frame Buffer - Not as big a deal at 1920, but one of the major reasons to upgrade to the fastes GPU is for ultra high resolutions with AA. With a GX2 or SLI/CF solution, you're still limited to the same bus width and frame buffer as the individual cards even if you have more rendering horse power. This limitation is apparent in the higher resolutions with AA when comparing a GTX 280 with a true 512-bit bus and 1GB frame buffer to the X2/SLI solutions with a 256-bit bus and 512MB buffer.

I'm sure there's other things I can't think of right now but you get an idea.....

 

[HOOfan 1]

Originally posted by: Martimus

Actually, the Anandtech review showed that it used more power than the 9800GX2 (313W at load vs. 289W). In fact, they couldn't even run two GTX 280's in SLI with any PSU they had on hand. They had to jury-rig a second PSU to actually run benchmarks with the SLI's system.

I have seen several GTX 280 SLI Power comsuption graphs today and most of them had the entire system at around 510 Watts...that is about 8800GTX Sli power consumption....so Anandtech must have some crappy PSUs laying around.

If we assume the 510W is actual AC power draw, and the PSU was probably at most 85% efficient (streching it by a long shot, probably more like 78%)

then the entire system would only be pulling 433 Watts of actual DC power. A 650 Watt PSU with the proper connectors could run that.

 

[ajaidevsingh]

Originally posted by: ViRGE

Reviews:
AnandTech (It's our site, read it ) ///LOL, For real!!

Anyways the GTX 280 is really a sad story was waiting for a powerhouse now will order acard 2moro either a GTX260 or a 4870 'GTX280 is out..its too expensive and too hot i dont want another 9800 GX2 fiasco'

 

[AzN]

Originally posted by: SniperDaws

Originally posted by: Azn

Originally posted by: Extelleron

Originally posted by: tuteja1986
Reading beyond3d article , the gpu design team focused more on making a CUDA monster instead of monster for gaming. Leaves me to wonder what have nvidia have doing since the release of G80. Because all we are seeing from both ATI and Nvidia is incremental update in gaming aspect.

http://www.beyond3d.com/content/reviews/51/8
"Because GT200 doesn't implement a brand new architecture or change possible image quality compared to G80 or G92, we've been able to skip discussion of large parts of the chip simply because they're unchanged. There's nothing new to talk about in terms of maximum per-pixel IQ, because the crucial components of the chip that make that all happen have no improvements or changes to speak of. It's purely a question of performance and how that's derived."

As the R&D budgets for new GPUs continues to increase, nVidia and AMD are going to have to offset those increased costs by selling cards for more purposes than gaming. This is especially true since now GPUs like GT200 and RV770 have so much processing power. And GPGPU will become even more important in the face of Intel's Larrabee, which is going to be more general purpose and more suited toward such applications than current GPUs from nVidia and AMD.

It seems that nVidia took a page out of ATI's book with GT200, and they prioritized the shader core over the rest of the chip. GT200 has 87.5% more stream processors than G80 while it has only 25% more texture units and 33% more ROPs. The ratio of SP : Tex has increased from 2:1 in G80 to 3:1 in GT200. If we had more texture power in GT200, we would likely see more performance.

The other reason GT200 isn't such a monster is clocks.... the shader clock on the GTX 280 is lower than on the original 8800GTX, and the core/shader are much lower than on the 8800 Ultra and 9800GTX.

To be honest I'm wondering why GT200 has so many transistors..... no where in the chip do we see 2x G80, yet we see over 2x G80 in the number of transistors. Where are all these transistors going?

And about increases in gaming performance, AMD will certainly be providing that with R700.

Good insight here.

Fillrate is king and GT200 isn't so much better than G92 where it was starved for bandwidth. Although ROP helps you in high resolutions and AA it doesn't help you when there are multiple textures.

Same reason why 9600gt performs relatively close to 8800gt because it was more balanced than a full G92 with massive fillrate but not enough bandwidth.

Yeah im sure Nvidia have overlooked this, they must be kicking themselves.

Here we go again AZN and ihs magic fillrate. i wish you would phone the head of Nvidia and tell him your amazing facts that way we could have a really fast card.


Cheapest card at www.scan.co.uk is a whopping £433, that is disgusting and no way will i ever pay that for a graphics card.

They sure did overlook with 280gtx and 260gtx didn't they.

Here we go again with sniperdaws clueless about hardware comments as usual.

 

[chizow]

Originally posted by: Extelleron

We shall see. GT200, unlike RV770, was very close to a true doubling of G92 with all major vitals. It fell short with TMUs and clockspeeds and as a result, its performance isn't quite 2x that of G92 GTX or even GX2. RV770 isn't a true doubling of RV670 so your predictions are incredibly optimistic imo. All the early indicators and leaks have pointed at 4850 being equivalent to a G92 GTX at best and the 4870 being 20-30% faster.

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

I don't have it backwards compared to G92 which is what GT200 is derived from, not G80. RV770 isn't a doubling of RV670, not in transistors and not in major vitals. If it comes close to doubling RV670 I'd be shocked and would be the first to congratulate AMD on a major win.

 

[SunnyD]

ArsTechnica's review sums it up:

It's hard to justify buying a $649 GTX 280 when it's outperformed by a Palit 9800 GX2 that happens to cost $141 less, and I know which card I'd personally pick.

The GX2 wins a good portion of the tests over the GTX280. The 280 is no slouch, but at that price point... most definitely NOT worth it.

 

[ajaidevsingh]

I was just wondering if we throw GX2 and 280 for Dx10 CRYSIS @ 1920*1200 who will win???

 

[AzN]

Originally posted by: chizow

Originally posted by: Extelleron

We shall see. GT200, unlike RV770, was very close to a true doubling of G92 with all major vitals. It fell short with TMUs and clockspeeds and as a result, its performance isn't quite 2x that of G92 GTX or even GX2. RV770 isn't a true doubling of RV670 so your predictions are incredibly optimistic imo. All the early indicators and leaks have pointed at 4850 being equivalent to a G92 GTX at best and the 4870 being 20-30% faster.

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

I don't have it backwards compared to G92 which is what GT200 is derived from, not G80. RV770 isn't a doubling of RV670, not in transistors and not in major vitals. If it comes close to doubling RV670 I'd be shocked and would be the first to congratulate AMD on a major win.

They are actually all derived from the same architecture. By the looks of the rop and texture count it derives more from G80 than G92.

 

[ajaidevsingh]

Originally posted by: SunnyD
ArsTechnica's review sums it up:

It's hard to justify buying a $649 GTX 280 when it's outperformed by a Palit 9800 GX2 that happens to cost $141 less, and I know which card I'd personally pick.

The GX2 wins a good portion of the tests over the GTX280. The 280 is no slouch, but at that price point... most definitely NOT worth it.

But i did read that 280 SLi is better than a GX2 SLi in overall scaling...

 

[ajaidevsingh]

Originally posted by: Azn

Originally posted by: chizow

Originally posted by: Extelleron

We shall see. GT200, unlike RV770, was very close to a true doubling of G92 with all major vitals. It fell short with TMUs and clockspeeds and as a result, its performance isn't quite 2x that of G92 GTX or even GX2. RV770 isn't a true doubling of RV670 so your predictions are incredibly optimistic imo. All the early indicators and leaks have pointed at 4850 being equivalent to a G92 GTX at best and the 4870 being 20-30% faster.

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

I don't have it backwards compared to G92 which is what GT200 is derived from, not G80. RV770 isn't a doubling of RV670, not in transistors and not in major vitals. If it comes close to doubling RV670 I'd be shocked and would be the first to congratulate AMD on a major win.

They are actually all derived from the same architecture. By the looks of the rop and texture count it derives more from G80 than G92.

I was about to write the same point AZN.. I recall that indian chick said that GT200 is a fusion of two G92 cores and woo she was rite!!

 

[SunnyD]

Originally posted by: ajaidevsingh
I was just wondering if we throw GX2 and 280 for Dx10 CRYSIS @ 1920*1200 who will win???

From Ars:

In keeping with that theme, I've made a few changes to our test methodology as well. High Quality (HQ) settings are still the same, and are run at a resolution of 1680x1050 with 4xAA and 8xAF enabled. Game settings are adjusted to "High," either by selecting the appropriate in-game option, or by manual tuning. I've also added a new Ultra High Quality (UHQ) setting designed to push cards even further. UHQ tests are run at a resolution of 1900x1200, at 8xAA (not 8xQ) and 16xAF. As for in-game settings, they're either set for "Very High" (when available), or hand-tuned to maximum levels. All tests are run 3x and the results averaged.

The GX2 wins ALL of the UHQ Crysis benchmarks.

 

[SunnyD]

Originally posted by: ajaidevsingh

Originally posted by: SunnyD
ArsTechnica's review sums it up:

It's hard to justify buying a $649 GTX 280 when it's outperformed by a Palit 9800 GX2 that happens to cost $141 less, and I know which card I'd personally pick.

The GX2 wins a good portion of the tests over the GTX280. The 280 is no slouch, but at that price point... most definitely NOT worth it.

But i did read that 280 SLi is better than a GX2 SLi in overall scaling...

That may be so, but are you planning on dropping $1200 for "improved scaling"? I'm not.

 

[SniperDaws]

Originally posted by: trajan2050
G280 will just get better and better. For a high end monitor this is the only way to go.

Because if ATI doesnt kick the shit out of Nvidia this time then youll have no other choice will you.........im sure Nvidia could release a much better card than this just ask Azn.

 

[Genx87]

The 9800 GX2 will most likely cease to ship soon. Or maybe it already has. This is no different than the 7900GTX2. It was a niche product for a few months until the G80 showed up and provided similar performance in a single GPU. The 7900GTX2 stopped being sold and its single GPU cousins worked their way down the price points.

One of the biggest problems as somebody else showed. We are clearly running into CPU limitations. The only way to see huge differences now is to run at insanely high resolutions. I have been at 1600X1200 for a decade. In the fall I am actually going down a notch when I go widescreen with a 22".

 

[ajaidevsingh]

Originally posted by: SunnyD

Originally posted by: ajaidevsingh

Originally posted by: SunnyD
ArsTechnica's review sums it up:

It's hard to justify buying a $649 GTX 280 when it's outperformed by a Palit 9800 GX2 that happens to cost $141 less, and I know which card I'd personally pick.

The GX2 wins a good portion of the tests over the GTX280. The 280 is no slouch, but at that price point... most definitely NOT worth it.

But i did read that 280 SLi is better than a GX2 SLi in overall scaling...

That may be so, but are you planning on dropping $1200 for "improved scaling"? I'm not.

For 280 SLi it is req. to have a small nuclear plant, which was pointed out in this forum somewhere but i sadlly dont and neither will i give 1K for some half baked cake...!!!

Nvidia did not fair well this time around lets hope they soon bring out a 55nm GTX 240 "8800GT Type" which is cheap and a hot performer!! But i can not wait till then i am buying either 4870 or GTX 260 asap!!

 

[Extelleron]

Originally posted by: chizow

Originally posted by: Extelleron

We shall see. GT200, unlike RV770, was very close to a true doubling of G92 with all major vitals. It fell short with TMUs and clockspeeds and as a result, its performance isn't quite 2x that of G92 GTX or even GX2. RV770 isn't a true doubling of RV670 so your predictions are incredibly optimistic imo. All the early indicators and leaks have pointed at 4850 being equivalent to a G92 GTX at best and the 4870 being 20-30% faster.

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

I don't have it backwards compared to G92 which is what GT200 is derived from, not G80. RV770 isn't a doubling of RV670, not in transistors and not in major vitals. If it comes close to doubling RV670 I'd be shocked and would be the first to congratulate AMD on a major win.

You do realize G92 and G80 are exactly the same thing in terms of specs, except G92 has a 1:1 ratio of TA:TF and integrates the display chip on die?

RV770 is a doubling of RV670, as I just showed you. Did you read my post? Is shading performance, texture performance, and memory bandwidth not enough? Transistor numbers are not important. The R600 design takes up much less space than G80 per ALU and that's why AMD can fit so much more in not so much more space.

 

[Genx87]

Originally posted by: ajaidevsingh

Originally posted by: SunnyD

Originally posted by: ajaidevsingh

Originally posted by: SunnyD
ArsTechnica's review sums it up:

It's hard to justify buying a $649 GTX 280 when it's outperformed by a Palit 9800 GX2 that happens to cost $141 less, and I know which card I'd personally pick.

The GX2 wins a good portion of the tests over the GTX280. The 280 is no slouch, but at that price point... most definitely NOT worth it.

But i did read that 280 SLi is better than a GX2 SLi in overall scaling...

That may be so, but are you planning on dropping $1200 for "improved scaling"? I'm not.

For 280 SLi it is req. to have a small nuclear plant, which was pointed out in this forum somewhere but i sadlly dont and neither will i give 1K for some half baked cake...!!!

Nvidia did not fair well this time around lets hope they soon bring out a 55nm GTX 240 "8800GT Type" which is cheap and a hot performer!! But i can not wait till then i am buying either 4870 or GTX 260 asap!!

Fair well compared to whom? Themselves?!?!?!?!?

Lets wait on what AMD has before claiming Nvidia didnt fair well.

 

[JPB]

Originally posted by: Genx87
The 9800 GX2 will most likely cease to ship soon. Or maybe it already has. This is no different than the 7900GTX2. It was a niche product for a few months until the G80 showed up and provided similar performance in a single GPU. The 7900GTX2 stopped being sold and its single GPU cousins worked their way down the price points.

One of the biggest problems as somebody else showed. We are clearly running into CPU limitations. The only way to see huge differences now is to run at insanely high resolutions. I have been at 1600X1200 for a decade. In the fall I am actually going down a notch when I go widescreen with a 22".

7950GX2

Fixed :thumbsup:

 

[AzN]

I don't doubt 280gtx will hold the crown for 2x the price and 1.5x the power consumption over RV770.

AMD is sure going to kick in the price performance though even take the crown back with R700.

 

[Wreckage]

Originally posted by: Extelleron

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

You should wait for actual benchmarks prior to making such statements.

Right now it looks like ATI has fallen completely out of the high end and that's probably how it's going to be from now on.

Most people did not think AMD would want to spend the money on R&D to compete and would be happy putting out mid to low end parts.

So far this plan is coming to fruition.

 

[Genx87]

Originally posted by: Extelleron

Originally posted by: chizow

Originally posted by: Extelleron

We shall see. GT200, unlike RV770, was very close to a true doubling of G92 with all major vitals. It fell short with TMUs and clockspeeds and as a result, its performance isn't quite 2x that of G92 GTX or even GX2. RV770 isn't a true doubling of RV670 so your predictions are incredibly optimistic imo. All the early indicators and leaks have pointed at 4850 being equivalent to a G92 GTX at best and the 4870 being 20-30% faster.

You have it backwards. GT200 isn't anywhere close to a doubling of G80, meanwhile RV770 is truly double RV670 or more.

I don't have it backwards compared to G92 which is what GT200 is derived from, not G80. RV770 isn't a doubling of RV670, not in transistors and not in major vitals. If it comes close to doubling RV670 I'd be shocked and would be the first to congratulate AMD on a major win.

You do realize G92 and G80 are exactly the same thing in terms of specs, except G92 has a 1:1 ratio of TA:TF and integrates the display chip on die?

RV770 is a doubling of RV670, as I just showed you. Did you read my post? Is shading performance, texture performance, and memory bandwidth not enough? Transistor numbers are not important. The R600 design takes up much less space than G80 per ALU and that's why AMD can fit so much more in not so much more space.

They arent an exact replica. In Anandtechs review they speak of many differences within the core of the chips. AMD's ALU's also do less work as evidenced by their lower performance with more of them compared to Nvidia.

/shrug