Note: UT's Vertical sync option is actually reversed. You have to turn it on to turn it off 😀

and maybe OA 0.8.8 could be used as a benchmark. Not really plugging, i'm suggesting it as it's even more unoptimized and dirtier than Q3A itself with memory-hungry high-poly vertex morphed MD3s, textures that cause a 64MB card to thrash, abuse of multitexture, tons of texture switching, bloom, detail textures and the ever-so-fps-killing readpixel-on-every-frame lensflares that would bring a modern Vulkan-supporting card to their knees (even worse with SLI 😜)

leileilol wrote:

Note: UT's Vertical sync option is actually reversed. You have to turn it on to turn it off 😀

Yeah that's quite weird..and it got me sometimes in the past wondering what's going on. But i guess it was really turned off 😜

This was with the ATI HD 4670.

Also here's a screenshot of GPU-Z running this very card. Seems to be running at 16x?

I'm not really a PCI-E expert but after googling PCI-E 1.1 @ 16x can deliver 4GB\s and AGP 8X does 2.13GB\s, so PCI-E bandwidth shouldn't be the bottleneck here for the newer cards. Most likely it's a driver limitation under XP. or the CPU itself. Also the games weren't the best to test the newer cards i know, but these titles were the games i was really interested in. I bet Crysis would scale up like crazy on the newer cards like 3D mark 2003 did.
I'm still disappointed with Doom 3 here. Also with the performance of my AMD HD 7870, it felt strangulated in every situation but 3D Mark 2003

Another to mention that modern hardware sometimes doesn't report compiled vertex array extensions properly to idtech3 games like Q3 so it won't use the fastest drawing path automatically. You may need to force r_primitives 2 on every card you test

I've theorized myself that at a certain point video cards (and their drivers) as they got newer and newer.. both nvidia and ATI/AMD started progressively omitting optimizations in hardware and software for older DirectX-8 and further back games and the newer cards got progressively slower at older DirectX content, and instead the card manufacturers just started optimizing hardware and software for newer DirectX-9 and DX-10 instead. Almost to the point where in some titles, older cards almost match and sometimes surpass the newer bits.

It's just a theory to me but still.. I think it actually is a "thing", and I think we can sort of see it to some degree in the test results here.

And yeah, it looks like your testing board ran all the PCIE cards at "full speed" PCI-Express 1.1 16x, so that's not the issue with the newer cards, it's something else.

kithylin wrote:

It's just a theory to me but still.. I think it actually is a "thing", and I think we can sort of see it to some degree in the test results here.

This is what i was wondering about. Is there some way we could scientifically prove it?

I was going to comment earlier. There have been claims that Fermi gen cards are faster on DX9 games than later gen cards. Not sure where the supposed "bell curve" would peak on the Radeon side. Drivers must also play a big role on this.

If only we could 'standardize' on a testbed/platform and crowdsource all these GPU variations...

xjas wrote:

Weird how the Radeon HD7870 is almost universally beaten or matched by the 3850 & 4670, except in the final 3DMark. Wonder what's going on there...

Slow CPU and different driver optimizations for each GPU architecture, which are most visible in CPU limited scenarios. 3dmark 03 is not that much limited... but still delivering only half the true performance of HD 7870.

Just mssing dx9 optimisation ?

Texture of UT99 looks washed 😳
ah i remember S3 Metal, it has soo nice textures.

kithylin wrote:

I've theorized myself that at a certain point video cards (and their drivers) as they got newer and newer.. both nvidia and ATI/AMD started progressively omitting optimizations in hardware and software for older DirectX-8 and further back games and the newer cards got progressively slower at older DirectX content, and instead the card manufacturers just started optimizing hardware and software for newer DirectX-9 and DX-10 instead. Almost to the point where in some titles, older cards almost match and sometimes surpass the newer bits.

It's just a theory to me but still.. I think it actually is a "thing", and I think we can sort of see it to some degree in the test results here.

And yeah, it looks like your testing board ran all the PCIE cards at "full speed" PCI-Express 1.1 16x, so that's not the issue with the newer cards, it's something else.

its a function of the card can brute force it without needing any optimizations in drivers. Also the reason older cards would score better than newer would also have to do with cpu overhead. Newer drivers and cards require more CPU overhead. And UT99 is CPU capped on newer cards, pretty much anything newer than say a Geforce FX would CPU bottleneck on UT.

matze79 wrote:

Just mssing dx9 optimisation ?

Texture of UT99 looks washed 😳
ah i remember S3 Metal, it has soo nice textures.

I've used S3 Metal in these tests, you can't notice that because i'm standing as much as i can against a wall and not all textures in UT are replaced with a higher S3 res. counterpart, because there isn't a S3TC version for that particular texture, as it is the case of deck16 textures were this printscreen was taken. Also i couldn't capture the games actual brightness on this printscreen because i was using opengl (i believe it was because of that), that's why it is so darkened.

The game looks astonishing in maps rich with S3TC textures like DM-Codex!

It's interesting to see where the Radeon 3450 fits in. It's 4 pixels per clock like GF3/4 but it has more than double the clock speed to save the day, and I guess the unified architecture is great for Doom3.

"3DMark 01" = 3DMark01 SE ?
Did you ran 3DMark 2000 in compatibility mode ?

agent_x007 wrote:

"3DMark 01" = 3DMark01 SE ?
Did you ran 3DMark 2000 in compatibility mode ?

Also I'm curious what OS was used for these tests. If Windows XP 32-bit, then no compatibility mode is needed for 3dmark 2000, it just runs natively. I think.

kithylin wrote:

agent_x007 wrote:

"3DMark 01" = 3DMark01 SE ?
Did you ran 3DMark 2000 in compatibility mode ?

Also I'm curious what OS was used for these tests. If Windows XP 32-bit, then no compatibility mode is needed for 3dmark 2000, it just runs natively. I think.

It was Windows XP 32-bit.

3D Mark 2001 was SE and i didn't ran 3D Mark 2000 in compatibility mode as i believe it was not necessary. Although the AMD HD 7870 wasn't able to run 3D Mark 2000 with or without compatibility mode.

rick6 wrote:

It was Windows XP 32-bit.

3D Mark 2001 was SE and i didn't ran 3D Mark 2000 in compatibility mode as i believe it was not necessary. Although the AMD HD 7870 wasn't able to run 3D Mark 2000 with or without compatibility mode.

http://www.outfoxed.net/3DMK2000patch4.zip replace your 3dmark2000.exe with this modified one (you'll need to rename it) and it's a modified / patched EXE file for 3dmark 2000 to allow it to run on newer video cards with higher video ram onboard. Works even on my 4GB GTX 770 card.

I did it and it worked, but only when i changed the color depth, otherwise i would get a incredible amount of artifacts. It scored 15000+/-, really awkward 😁

Also i don't get why it's so hard to get a score above 20000\25000 on 3DMark 2000. Maybe it's a limitation of the formula it uses to calculate the score.

rick6 wrote:

I did it and it worked, but only when i changed the color depth, otherwise i would get a incredible amount of artifacts. It scored 15000+/-, really awkward 😁

Also i don't get why it's so hard to get a score above 20000\25000 on 3DMark 2000. Maybe it's a limitation of the formula it uses to calculate the score.

Nope, just need faster processors to go with it.

54,000 with some dual-gtx-260's I have and a i7-3770k @ 4.5 ghz: http://www.outfoxed.net/dual-gtx-260/3dMark%202000.jpg

I should re-do that test some day now that I have the cpu @ 4.8 ghz stable.

rick6 wrote:

Also i couldn't capture the games actual brightness on this printscreen because i was using opengl (i believe it was because of that), that's why it is so darkened.!

F9

kithylin wrote:

rick6 wrote:

Also i don't get why it's so hard to get a score above 20000\25000 on 3DMark 2000. Maybe it's a limitation of the formula it uses to calculate the score.

Nope, just need faster processors to go with it.

54,000 with some dual-gtx-260's I have and a i7-3770k @ 4.5 ghz: http://www.outfoxed.net/dual-gtx-260/3dMark%202000.jpg

Didn't know my Core 2 Extreme is so fast : 33800 pkt. - LINK 😀
And that's with just 3,2GHz on one core, while using VIA chipset/x4 PCI-e speed and a single GPU mode 7900 GX2 graphics card.