[Lezmaka]

Dell Dimension XPS and Dell Dimension 8400

That's right, a 6800 GTO

If you click on "Help Me Choose" it simply says it's using a GeForce 6800 chip, no GT suffix even though other cards based on chips like X800 XT and X800 SE do have the correct respective suffixes.

If it is a 12 pipe 6800, it's even more of a mutant than the Asus 12 pipe GT. 12 pipes (6800 standard), 350MHz core (same as GT), 256MB GDDR3 (same as GT/Ultra) @ 900MHz (lower than GT/Ultra) yet dual DVI (Ultra) but no TV out.

$330 to upgrade from X300 SE, $150 to upgrade from X800 SE, but $120 more to upgrade to X800 XT.

BTW, these are all PCI-E cards.

Edit: Fixed a link

[Dave Baumann]

Its got 450MHz RAM. I think its just a GT with 50MHz off the ram speeds.

[Guden Oden]

Egads, there's TOO MANY CARD VERSIONS these days to keep track of, and now the midrange stuff is coming too in several more flavors/performance levels, some of which will be comparable pricing/performance-wise with the low-end side of the high-end spectrum. :P

It's a mess, particulary on the NV side it seems...

I can't imagine the card OEMs are too happy about this current situation either, seems to me they could suddenly get stuck with old, unattractive products with the way things are now, but maybe that is part of why ATi and NV want to extend the chip generations more and more too. Give their customers more time to offload their stuff onto us.

*shrugs*

[saf1]

Guden Oden,

Particulary on the NV side? Both parties. Go check out pricewatch.com, select video cards, and look at both NV and ATI. Both have too many.

[max-pain]

http://gzeasy.com/itnewsdetail.asp?nID=16077

[PatrickL]

So that GTO is a 12 pipes card ?

[Mariner]

Originally Posted by PatrickL

So that GTO is a 12 pipes card ?

Sort of a 6800NU Pro, I suppose. Kind of reminds me of the various ATI cards that Medion keep selling in their PCs in supermarkets here in Europe - half way between one card and another and with a peculiar name.

[Xmas]

Originally Posted by Mariner

Sort of a 6800NU Pro, I suppose. Kind of reminds me of the various ATI cards that Medion keep selling in their PCs in supermarkets here in Europe - half way between one card and another and with a peculiar name.

Yep, like the Radeon 9800TX. Such "special OEM editions" aren't unusual, and it's an easy way for the IHVs to sell their between-two-specs-chips at a better price.

That card should be quite good for antialiasing, as it has more bandwitdth per pipe than the other 6800 models.

[Chalnoth]

Antialiasing doesn't really take more bandwidth anymore, due to framebuffer compression.

[Ailuros]

Originally Posted by Chalnoth

Antialiasing doesn't really take more bandwidth anymore, due to framebuffer compression.

When it comes to AA performance comparisons between a 6800 and a 6800GT in high resolutions, the difference is hardly only due to the amount of onboard ram alone. The GT has nearly 50% more raw bandwidth and fill-rate is the last thing that's relevant to Multisampling anyway (especially since all ROPs are still operational on the 6800, despite the disabled quad).

[Chalnoth]

There is a difference in fillrate with multisampling, though, at every triangle edge. And yes, a large part of the difference is due to memory size, at least in newer games. Memory size constraints are pretty much the only reason I don't run in 4x AA in all of my games.

If you want to make a claim that memory bandwidth is still the primary limiting factor, you'd need to take a look at benchmarks in older games (heck, I guess I'll go look now).

Edit: Hrm, didn't find any suitable benchmarks after a quick look....

[Ailuros]

Originally Posted by Chalnoth

There is a difference in fillrate with multisampling, though, at every triangle edge. And yes, a large part of the difference is due to memory size, at least in newer games. Memory size constraints are pretty much the only reason I don't run in 4x AA in all of my games.

If you want to make a claim that memory bandwidth is still the primary limiting factor, you'd need to take a look at benchmarks in older games (heck, I guess I'll go look now).

Edit: Hrm, didn't find any suitable benchmarks after a quick look....

http://www.asus.com/prog/spec.asp?m=...E&langs=01

http://www.hothardware.com/viewartic...?articleid=569

http://techreport.com/reviews/2004q3...t/index.x?pg=1

[Ailuros]

I just bumped into that one (haven´t paid digit-life a visit recently):

http://www.digit-life.com/articles2/...43-p01.html#p1

If one doesn´t get confused or gets a headache from those 22 pages, of course...

[Chalnoth]

What, specifically, are you looking at in these benchmarks?

[Ailuros]

Originally Posted by Chalnoth

What, specifically, are you looking at in these benchmarks?

At everything and I still don't see how framebuffer size should be the primary limiting factor here. The V9999 is clocked at 350/500MHz (which gives it 32GB/s bandwidth) and has 256MB ram on board. I don't see it making a significant difference to a vanilla 6800 to be honest, as I have severe doubts that if a 6800 would have by default 256MB ram that it would be able to be playable with high resolution AA (4x to be more exact).

I have to note that you didn't claim that framebuffer size is the only limiting factor either, yet I don't agree either with the standpoint that it's the most important culprit on the 6800. It's a bundle of factors.

Would we have a V9999 owner right we could have him downclock to 325/350, I'm willing to bet that it'll run on par with a 6800.

[Chalnoth]

From owning a GeForce 6800, I can tell you that the only reason I don't run all of my games at 1280x960 with 4x AA is that texture management issues crop up in some of the games I play. If you've ever seen texture management problems, you know exactly what I'm talking about.

[Ailuros]

Originally Posted by Chalnoth

From owning a GeForce 6800, I can tell you that the only reason I don't run all of my games at 1280x960 with 4x AA is that texture management issues crop up in some of the games I play. If you've ever seen texture management problems, you know exactly what I'm talking about.

I do have a 6800 too and no I haven´t seen yet texture management problems. Slideshows if the game is too demanding at maximum settings yes.

[Chalnoth]

Hmm, well, I just went back and tested one of the games that I was forced to run with only 2x AA before (UT2004), and now it appears that performance is just fine with 4x AA. Hrm, there must have been something really wrong with their texture management algorithms with early drivers. It was so bad in this game that I didn't think it'd be fixed to the degree it has been. So, I suppose that particular problem with the 6800 (128MB of onboard RAM) may have been mostly fixed through updated drivers. I'd have to do some more checking.

Regardless, back to the discussion, where is a benchmark that specifically shows that the 6600 GT suffers particularly in AA scenarios from its lack of bandwidth?

[Ailuros]

Originally Posted by Chalnoth

Hmm, well, I just went back and tested one of the games that I was forced to run with only 2x AA before (UT2004), and now it appears that performance is just fine with 4x AA. Hrm, there must have been something really wrong with their texture management algorithms with early drivers. It was so bad in this game that I didn't think it'd be fixed to the degree it has been. So, I suppose that particular problem with the 6800 (128MB of onboard RAM) may have been mostly fixed through updated drivers. I'd have to do some more checking.

I never ran UT2k4 with 4xAA in 1280, because I´d have to flip back to 2xAA for ONS maps. I do have quite a few results from the 61.45 drivers (optimisations off) and should really give them a run again today in UT2k4 to see how much has actually changed.

***edit: errr nope with high quality results are nearly identical. Quality being an entire different chapter...

CTF Face3 flyby/1280*1024, 4xAA/8xAF

High quality : 37.65 fps
Quality : 69.76 fps

*cough*

I suppose you´ve also tried FarCry with =/>65.xx or not? (performance increases should come here from better tuned lighting calculations - I suppose - for either SM2.0 or 3.0).

Regardless, back to the discussion, where is a benchmark that specifically shows that the 6600 GT suffers particularly in AA scenarios from its lack of bandwidth?

I´m not going to re-read that 22-page headache from digit life, but I recall while reading it, detecting more than a few cases where there were indications that it lacks in bandwidth with AA enabled.

http://techreport.com/reviews/2004q3...t/index.x?pg=4

NoAA/ 1600*1200

6600GT = 62.3 fps
6800 = 66.2 fps

4xAA/ 1600*1200

6600GT = 28.8 fps
6800 = 42.4 fps

(also notice how the 9800XT scales). Not a very clear indication frankly because the 6800 has 4 times the ROPs of a 6000.

CS:s beta:

http://techreport.com/reviews/2004q3...t/index.x?pg=6

VST and Far Cry about the same trend.

Again it´s a combination of factors, but I don´t see any clear indications that the size of the framebuffer is more important than the amount of raw bandwidth.

[Chalnoth]

The different number of ROPs sounds much more likely to be the cause of the performance disparity. With framebuffer compression in use, i.e. when FSAA is enabled, the 6800 can actually make use of all of its ROPs. The 6600 doesn't have that advantage. This would explain why the performance difference between the two architectures with 4x AA is more than the difference that memory bandwidth alone would indicate.

Now, what this should mean is that the 6600 will still have high AA performance as long as a lot of time is spent in processing each pixel. Doom 3, for instance, isn't a very good indicator because of the large fillrate demands of shadow volumes. Far Cry should be a much better indication of how the 6600's AA will perform moving forward, as we are moving towards more time spent in the shader for each pixel.

Granted, there is still the problem with shadow rendering, which the 6600 will be relatively slow at no matter what, but if the games use long enough shaders, and thus the time spent rendering shadows (either shadow maps or shadow volumes) is relatively small, then it shouldn't be a huge drawback.

So I still don't feel that memory bandwidth is the deciding factor with AA performance, not any more.

[Ailuros]

Originally Posted by Chalnoth

The different number of ROPs sounds much more likely to be the cause of the performance disparity. With framebuffer compression in use, i.e. when FSAA is enabled, the 6800 can actually make use of all of its ROPs. The 6600 doesn't have that advantage. This would explain why the performance difference between the two architectures with 4x AA is more than the difference that memory bandwidth alone would indicate.

Amount of ROPs are a factor between 6600 and 6800, especially in Doom3, still not the only one.

Now, what this should mean is that the 6600 will still have high AA performance as long as a lot of time is spent in processing each pixel. Doom 3, for instance, isn't a very good indicator because of the large fillrate demands of shadow volumes. Far Cry should be a much better indication of how the 6600's AA will perform moving forward, as we are moving towards more time spent in the shader for each pixel.

Fill-rate demands as in AA fill-rate I suppose? If yes then I agree.

There were Far Cry links in my former post or where you just too bored to read the reviews supplied?

http://techreport.com/reviews/2004q3...t/index.x?pg=8

FC-Cooler/1600*1200

noAA/AF:

6600GT = 36.86
6800 = 43.28

4xAA/8xAF

6600GT = 17.19
6800 = 22.42

FC- Pier/1600*1200

noAA/AF:

6600GT = 36.27
6800 = 45.07

4xAA/8xAF

6600GT = 9.36
6800 = 18.98

FC- Volcano/1600*1200

noAA/AF

6600GT = 48.29
6800 = 58.89

4xAA/8xAF

6600GT = 21.18
6800 = 31.66

As of course the fact that the 9800XT (albeit limited to the highly unoptimized SM2.0 path) seems to be in most occassions very close to the 6800. I'm sure you know well enough how many quads, at what level the artithmetic efficiency and bandwidth amount of the R360 is.

Granted, there is still the problem with shadow rendering, which the 6600 will be relatively slow at no matter what, but if the games use long enough shaders, and thus the time spent rendering shadows (either shadow maps or shadow volumes) is relatively small, then it shouldn't be a huge drawback.

If games start using long enough shaders 6800GT and Ultra accelerators will be rather mainstream equivalent and despite that the 6600GT will still prove a significant muscle there, I have severe doubts that there's even a chance of AA with that in one high resolutions for future applications.

The longest shaders we're going to get in the foreseeable future, seem to be on the HL2 level; albeit preliminary tests and not 100% representative for final game performance, there are CS:s and VST scores in the links above (+68% for the 6800 in CS:S and +53% for the 6800 in VST -1600+4xAA/8xAF).

So I still don't feel that memory bandwidth is the deciding factor with AA performance, not any more.

I never said it is or that it's only one; it's still important though and yes I refuse to accept that framebuffer size takes a higher importance than bandwidth.

If bandwidth would be so irrelevant with AA performance and especially with arithmetic calculations increasing, then not only would be the 32-35GB/sec bandwidth on today's high end accelerators virtually a waste, but there shouldn't be much reason to increase bandwidth in future accelerators either.

Finally I'd like to have a bit more of documentation than just simple feelings. I won't consider following the wrong logic, unless proven wrong.