[H] saw a 500-600MB reduction in Crysis when switching from DX9 to DX10. That is quite a bit more than the 200MB that you are calling "generous"
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
No performance increase with SLI. Why?
- Thread starter pardo
- Start date
[H] saw a 500-600MB reduction in Crysis when switching from DX9 to DX10. That is quite a bit more than the 200MB that you are calling "generous"
[H] saw a 500-600MB reduction in Crysis when switching from DX9 to DX10. That is quite a bit more than the 200MB that you are calling "generous"
Here's some of that new fangled Chinese math they taught me:
2123MB (DX9) - 1889MB (DX10) = 234 MB. And that's just on first load. As the game progresses, the difference btw. DX9 and DX10 gets smaller.
Notice also the wonderful 10-13 fps. drop going from DX9 to DX10. Not worth it IMO, but if it makes others happy...that's fine. It still doesn't change my original point(s).
I don't know how or what [H] tested to get their results. Maybe the config. file changes the disparity. Maybe the /3Gb switch influences the disparity. All I know is what I see.
I just did the same run through but with DX9, and it used about 300MB more (1.6GB total) - again, this is with texturestreaming=0 and all of the same config tweaks you are using.
I'm curious what they changed to get Warhead to suck down another 400+MB...
I'm curious what they changed to get Warhead to suck down another 400+MB...
Once again, AA and AF won't affect system RAM usage. AF doesn't even increase VRAM usage...
I didn't use any AA for those screenies. Just 2X edge AA.
I should be getting my second GTX 260 tomorrow. I'll report back my SLI experience.
A few things I noticed:
I'm now running my e8400 @ 4.2 GHz on a 780i....all things being equal, it's 2-3 fps. slower than a P43 motherboard running the same e8400 @ 3.8 Ghz.
The P43 reported a solid 3 GB of system RAM in Windows (what I have installed) with the GTX 260. The 780i is reporting 2.75 GB.
I'm now running XP because my Vista install is tied to the P43 motherboard and I don't feel like haggling with India to get it activated again. If I like 260 SLI enough, i'll get Vista-64.
And btw. kllrnohj, Warhead surprisingly runs the same FPS in DX9 in both Vista and XP.
I haven't activated the /3GB switch in XP and with the same config. file, texturesstreaming set to "0" I started at about 1400 MB in DX9 (it was 2100 MB in DX9 Vista-32) So I don't know if that high RAM usage was due to Vista-32, the /3GB switch or a combination of the two. At any rate, I still find the switch beneficial for 32-bit OS.
I should be getting my second GTX 260 tomorrow. I'll report back my SLI experience.
A few things I noticed:
I'm now running my e8400 @ 4.2 GHz on a 780i....all things being equal, it's 2-3 fps. slower than a P43 motherboard running the same e8400 @ 3.8 Ghz.
The P43 reported a solid 3 GB of system RAM in Windows (what I have installed) with the GTX 260. The 780i is reporting 2.75 GB.
I'm now running XP because my Vista install is tied to the P43 motherboard and I don't feel like haggling with India to get it activated again. If I like 260 SLI enough, i'll get Vista-64.
And btw. kllrnohj, Warhead surprisingly runs the same FPS in DX9 in both Vista and XP.
I haven't activated the /3GB switch in XP and with the same config. file, texturesstreaming set to "0" I started at about 1400 MB in DX9 (it was 2100 MB in DX9 Vista-32) So I don't know if that high RAM usage was due to Vista-32, the /3GB switch or a combination of the two. At any rate, I still find the switch beneficial for 32-bit OS.
Ok, just got my GTX 260 SLI up and running here are some early results:
Specs: E8400 @ 4.2 GHz, 780i, 3 GB RAM, XP
Warhead Benchmark tool (Ambush):
1680 x 1050 DX9 enthusiast 4X AA single 260: 21 FPS. AVG.
1680 x 1050 DX9 enthusiast 4X AA SLI 260: 39 FPS. AVG.
1680 x 1050 DX9 Gamer no AA single 260: 46 FPS. AVG.
1680 x 1050 DX9 Gamer no AA SLI 260: 55 FPS. AVG.
The Ambush average doesn't tell the whole story. Ice levels are seeing a 10-15 fps. boost with SLI over a single card, which, as everyone who has ever played Crysis knows, really makes a difference. Overall, Warhead looks excellent in SLI and really shines with AA enabled. I'm running a bit better than the OP, but I couldn't test in DX10, either so..
I never bought into the "SLI isn't worth it @ 1680 x 1050" camp. ...because it's simply not true.
After rebates and cashback, it was about $300 for GTX 260 SLI which just about puts it in
HD 4850 Crossfire territory with more power and VRAM.
I'll give some more tests and observations with FC2, FO3, COD4, and Clear Sky later for anyone interested.
Newer titles using complex engines may not scale well on multiple GPUs at release because devs are focused on console performance. The nVidia or ATi driver team will be responsible for getting multiple GPUs scaling going in said titles. Their degree of success will vary.
You may add to the mix that simplistic engines that are easy to slice up and spread across multiple GPUs for near linear scaling probably don't need more than a single strong GPU to run well in most cases.
Most games are developed and optimized for the console. 90% of the time it's all about console with PC sales as an afterthought.
A good article below that includes commentary from Carmack about multi-GPU-
"From a developer stand point the uncomfortable truth is that the console capabilities really dominate the development decisions today. If you look at current titles and how theyve done on the console, you know, high end action GPU based things, the consoles are so the dominate factor that its difficult to set things up so that you can do much to leverage the really extreme high end desktop settings.
..its difficult to try and justify any radically different algorithm, something you would really do with 4x the power youd have with a high end PC system.
Its probably making it even harder for the PC card guys because as developers get more sophisticated with the low level access we get on the consoles, the rendering engines are become harder to kind of, behind our backs, automatically split across multiple GPUs. We are doing more sophisticated things on the single GPU - there is a lot more data transfer going back and forth, updated states that have to be replicated across multiple GPUs, dependent sections of the screen doing different things. Its still possible buts its kind of a hairy job and I definitely dont envy those driver writers or their task at all."
In the conclusion-
"Where the trouble comes in though is with the introduction of DX9c and DX10, with features like feedbacks, sub-buffers and conditional queries, slicing up the work for multiple GPUs to address is getting much harder."
Source:
http://www.pcper.com/article.php?aid=532&type=expert&pid=3
You may add to the mix that simplistic engines that are easy to slice up and spread across multiple GPUs for near linear scaling probably don't need more than a single strong GPU to run well in most cases.
Most games are developed and optimized for the console. 90% of the time it's all about console with PC sales as an afterthought.
A good article below that includes commentary from Carmack about multi-GPU-
"From a developer stand point the uncomfortable truth is that the console capabilities really dominate the development decisions today. If you look at current titles and how theyve done on the console, you know, high end action GPU based things, the consoles are so the dominate factor that its difficult to set things up so that you can do much to leverage the really extreme high end desktop settings.
..its difficult to try and justify any radically different algorithm, something you would really do with 4x the power youd have with a high end PC system.
Its probably making it even harder for the PC card guys because as developers get more sophisticated with the low level access we get on the consoles, the rendering engines are become harder to kind of, behind our backs, automatically split across multiple GPUs. We are doing more sophisticated things on the single GPU - there is a lot more data transfer going back and forth, updated states that have to be replicated across multiple GPUs, dependent sections of the screen doing different things. Its still possible buts its kind of a hairy job and I definitely dont envy those driver writers or their task at all."
In the conclusion-
"Where the trouble comes in though is with the introduction of DX9c and DX10, with features like feedbacks, sub-buffers and conditional queries, slicing up the work for multiple GPUs to address is getting much harder."
Source:
http://www.pcper.com/article.php?aid=532&type=expert&pid=3
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
Devices today use sliding windows instead of mapping the entirety of graphics ram to memory space.
The typical window size for a PCIe device is 256MB. So, for an SLI setup, you lose 512MB. In addition to this, you have memory set aside for the PCI bus and other memory-mapped devices.
I've been looking for a good answer to that...it kept from me getting GTX 260 SLI for fear of losing most of my available system RAM.
It's seems to explain what I've noticed. I just got GTX260 SLI today...and with 3GB of system RAM installed, Windows XP has 2.5 GB available. Excellent post. Thank you.
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
Yup. Memory windows have always been a pretty standard concept when you have limited address space. They only fell out of style because of the adoption of IA32, which was more than enough memory for anyone...for two decades at least
Here is actually an excellent article on the subject, if you want to know more
http://www.gamasutra.com/view/feature/3602/sponsored_feature_ram_vram_and_.php?page=1