EVGA GeForce 8800GTS 320MB ACS3 benchmarks

[MmmSkyscraper]

Text is in Polish, doesn't stop you reading the graphs

Click


Edit: added AT article

 

[sisq0kidd]

Those are some very strange numbers for basically every single benchmark.

It beats out the 640mb version by a good margin in half the tests and in the other half, it tanks by a huge margin, losing even to the 7950gt.

I don't know what to make of this...

 

[MmmSkyscraper]

Maybe I should have used instead

 

[chizow]

Makes sense, the 320MB ACS3 Edition comes stock OC'd @580/850 compared to the standard 640MB version which clocks at 500/800. Also, the Shader core is clocked faster than even the GTX @1400MHz compared to 1200 and 1350 for the GTS and GTX respectively.

When you bump up the resolution though, the 640MB starts taking the lead again. I guess games do need more than 512MB when running high resolution with the graphics cranked up. Not sure about the 7950GT, but I'm going to guess its the 512MB version which allows it to outperform the 320MB GTS in some higher resolution tests.

Interesting part though, if all or most of these 320MB cards are stock OC'd, this will be an even more attractive offering than originally thought. Hard to say though, EVGA could easily have an ACS3/OC'd variant and a stock clocked reference variant.

 

[40sTheme]

I think it's very odd that it is dropping out so much from the higher resolutions... I think there is some optimizing to be done at nVidia. Especially since most benchmarks have shown that 256mb and 512mb versions of different video cards have made a very miniscule difference, and it definitely would not be outperformed by my 7950GT, although their benchmarks for that card were almost on the dot with what I get.

 

[Donuts]

Newegg had one listed now for $339 You can get the 640 version for $15 more after rebate. Hardly worth it at this price.

http://www.newegg.com/Product/Product.asp?Item=N82E16814150173

* Just noticed they have a bunch up at $299

 

[chizow]

AT posted their review and found some of the same results 8800 GTS 320 . Their findings aren't as consistent across the board but it definitely shows the 320MB version is crippled at higher resolutions with AA enabled.

Price point is very attractive imo. The stock clocked parts look to be selling for $299-$309 and the slightly OC'd versions for $320-$330. I'd expect the sticker price to drop in a few weeks along with a few MIRs once RD600 is closer to release putting these cards in the $250-280 price range. At that price, who in their right mind would buy anything else in the upper mid-range?

As it is though, you can basically get 2 of these cards and run them in SLI for the price of a GTX, which imo is much more realistic than SLI for the 640MB versions.

 

[videopho]

This card is a great deal for those who still game < 1680x1050 resolution.
As a matter of fact, based on on that Polish review site, it smokes the 640mb on 3dmark06 default and some other games, with the exception of COD2 where it bombs, so bad?

 

[Avalon]

I am completely unimpressed by the benchmarks I've seen thus far. If you're going to be playing at 1280x1024 or 1600x1200 without AA, or HL2, then sure, this card will work for you. However, I know a lot of gamers that play at 1600x1200 or above, and most spending $300+ on a video card are going to expect to be able to use AA. With those settings on, this card gets beaten by the old 7900GTX. Meh.

 

[munisgtm]

Originally posted by: videopho
This card is a great deal for those who still game < 1680x1050 resolution.
As a matter of fact, based on on that Polish review site, it smokes the 640mb on 3dmark06 default and some other games, with the exception of COD2 where it bombs, so bad?

that means it would be perfect for me .

 

[chizow]

Originally posted by: Avalon
I am completely unimpressed by the benchmarks I've seen thus far. If you're going to be playing at 1280x1024 or 1600x1200 without AA, or HL2, then sure, this card will work for you. However, I know a lot of gamers that play at 1600x1200 or above, and most spending $300+ on a video card are going to expect to be able to use AA. With those settings on, this card gets beaten by the old 7900GTX. Meh.

K, lets just end it before the FUD spreads. The review is a bit difficult to read because its in Polish, but I just looked it over again and there isn't a single instance where the 7950GT beats the 320MB 8800GTS under the same settings. People must not be noticing the 4x AA next to some of the bars in the lower resolution tests where the 7950GT is being compared. The 7950GT isn't even included in the 1920 resolution testing at all.

Anyways, the most interesting and telling part of this review is their use of a memory usage tool that tracks both local and non-local memory usage. Anyone know what tool that is? It shows pretty clearly why the 320MB version falls way behind the 640MB version at high resolutions with 4x AA enabled. The non-local memory (system and page file) get absolutely slammed at these resolutions, greatly decreasing relative performance.

So, conclusion is this card is still going to be the 3rd or 4th fastest card (X1950XT beats it a few times) at high resolutions w/ AA enabled. At 1920 with 4x AA, most frame rates are acceptable and still better than anything else this side of G80, so people who are expecting to game at those resolutions with 4x AA won't be disappointed.

 

[Avalon]

Originally posted by: chizow

Originally posted by: Avalon
I am completely unimpressed by the benchmarks I've seen thus far. If you're going to be playing at 1280x1024 or 1600x1200 without AA, or HL2, then sure, this card will work for you. However, I know a lot of gamers that play at 1600x1200 or above, and most spending $300+ on a video card are going to expect to be able to use AA. With those settings on, this card gets beaten by the old 7900GTX. Meh.

K, lets just end it before the FUD spreads. The review is a bit difficult to read because its in Polish, but I just looked it over again and there isn't a single instance where the 7950GT beats the 320MB 8800GTS under the same settings. People must not be noticing the 4x AA next to some of the bars in the lower resolution tests where the 7950GT is being compared. The 7950GT isn't even included in the 1920 resolution testing at all.

Anyways, the most interesting and telling part of this review is their use of a memory usage tool that tracks both local and non-local memory usage. Anyone know what tool that is? It shows pretty clearly why the 320MB version falls way behind the 640MB version at high resolutions with 4x AA enabled. The non-local memory (system and page file) get absolutely slammed at these resolutions, greatly decreasing relative performance.

So, conclusion is this card is still going to be the 3rd or 4th fastest card (X1950XT beats it a few times) at high resolutions w/ AA enabled. At 1920 with 4x AA, most frame rates are acceptable and still better than anything else this side of G80, so people who are expecting to game at those resolutions with 4x AA won't be disappointed.

As a disclaimer, I do want to mention that the eVGA ACS3 will obviously be faster than stock...I'm just talking about this card in general.

So...I wasn't really focusing on just the Polish review. I'm talking about ALL reviews I've seen, and the AT review in general that was interjected into this thread halfway. I also never mentioned the 7950GT anywhere, I said 7900GTX, but looking again, I am seeing places where even the 7950GT is beating the 8800GTS 320MB. Examples of both:

AT BF2 4xAA Benchmarks, the 8800GTS 320MB loses to the 7900GTX at 19x12 and 25x16. The 7950GT and X1950pro beat it at 25x16.

In Fear 4xAA, it loses to the 7900GTX @ 16x12, 19x12, and 25x16.

In HL2, once you stop becoming CPU bound (older game), the card drops down to 7900GTX levels with 4xAA.

In Quake 4 4xAA, it loses to the 7900GTX and 7950GT @ 16x12 and 19x12 (Yes, this is in Ultra mode, but with 4xAA tacked on, all cards should be reaching their memory limit).

Plus the X1950XT 256MB beats it in some of those scenarios I presented. As you can see by now (if you couldn't already), my main gripe is in this cards AA performance at 16x12 and above. Again, I feel that if you want to game with 16x12 + AA, this card takes such a huge performance hit it either loses out to older and cheaper cards, or barely outperforms them.

If you don't use AA at all, play below 16x12 with AA, or play older games, like HL2 based games, then have a ball. I'm unimpressed, and it should be clear by now why that is. I'm even more worried about how this card will handle future games with AA, which is why I am no longer interested in buying it. It could be a driver issue, or it could be that the G80 architecture really likes memory. Either way, you can get the 640MB version for as little as $50 more after rebate. Definitely worth it.

 

[aka1nas]

Originally posted by: chizow


Anyways, the most interesting and telling part of this review is their use of a memory usage tool that tracks both local and non-local memory usage. Anyone know what tool that is? It shows pretty clearly why the 320MB version falls way behind the 640MB version at high resolutions with 4x AA enabled. The non-local memory (system and page file) get absolutely slammed at these resolutions, greatly decreasing relative performance.

I'm not sure what they actually used in the article, but Rivatuner is capable of monitoring those values.

Edit: Yeah it looks like Rivatuner from the graphs

 

[chizow]

Originally posted by: Avalon
As a disclaimer, I do want to mention that the eVGA ACS3 will obviously be faster than stock...I'm just talking about this card in general.

So...I wasn't really focusing on just the Polish review. I'm talking about ALL reviews I've seen, and the AT review in general that was interjected into this thread halfway. I also never mentioned the 7950GT anywhere, I said 7900GTX, but looking again, I am seeing places where even the 7950GT is beating the 8800GTS 320MB. Examples of both:

AT BF2 4xAA Benchmarks, the 8800GTS 320MB loses to the 7900GTX at 19x12 and 25x16. The 7950GT and X1950pro beat it at 25x16.

In Fear 4xAA, it loses to the 7900GTX @ 16x12, 19x12, and 25x16.

In HL2, once you stop becoming CPU bound (older game), the card drops down to 7900GTX levels with 4xAA.

In Quake 4 4xAA, it loses to the 7900GTX and 7950GT @ 16x12 and 19x12 (Yes, this is in Ultra mode, but with 4xAA tacked on, all cards should be reaching their memory limit).

Plus the X1950XT 256MB beats it in some of those scenarios I presented. As you can see by now (if you couldn't already), my main gripe is in this cards AA performance at 16x12 and above. Again, I feel that if you want to game with 16x12 + AA, this card takes such a huge performance hit it either loses out to older and cheaper cards, or barely outperforms them.

If you don't use AA at all, play below 16x12 with AA, or play older games, like HL2 based games, then have a ball. I'm unimpressed, and it should be clear by now why that is. I'm even more worried about how this card will handle future games with AA, which is why I am no longer interested in buying it. It could be a driver issue, or it could be that the G80 architecture really likes memory. Either way, you can get the 640MB version for as little as $50 more after rebate. Definitely worth it.

I took a closer look at the AT review and you're right, there are instances where the 7950GT and 7900GTX beat the 320MB GTS in resolutions over 1600x1200 with 4x AA. However, AT gives no indication of the spec on those cards, so I'm going to assume they are the 512MB parts. Also, in the resolutions and AA settings the 320MB loses, the difference is within a few frames or borderline unplayable anyways. Splitting hairs between 13 fps and 15 fps at 2560 isn't exactly something I'd base a purchasing decision on.

That being said, its pretty clear why the 320MB GTS is performing so poorly, as evidenced by the benchmarks as well as the RivaTuner local memory usage graphs in that Polish review (thanks for the confirmation aka1nas). AT also mentioned it in their review noting both the 256MB X1950pro and X1900XT parts outperform it in BF2 @ 2560 w/ 4x AA. It could very well be a simple driver problem. The lower memory of the 320MB part isn't correctly reported or registered by the driver leading to inefficient memory caching/clearing of the frame buffer. The driver thinks it should be using 640+MB instead of limiting it to 320MB which causes the card to use system memory and page file to make up the difference, resulting in massive performance hits. The Polish review and RivaTuner graphs would support this.

As for how this card (or any of the current cards) will fare in future games, I wouldn't buy anything today thinking it'll run them at blazing fast speeds at high resolutions with everything cranked up considering the 8800 GTX is already showing its mortality in those instances. Also, the same price comparisons can be made with the other cards in the test. Is a 320MB GTS worth $30-$50 more than a 1900/7900 when it runs everything significantly faster without AA and about the same with AA in resolutions up to 1920? I still think the AA/high-res problems on the 320MB can be cleared up with a driver optimization, but I guess we'll see.

 

[Avalon]

Well, I hope it's a driver issue regarding mis-handling the ammount of memory the card has. If so, it should have a lot of room for AA performance improvement.