If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.
![]() |
|
|
#1 |
|
Member
|
My system:
MSI K8N Neo4 Platinum (PCB Rev1.0) Athlon X2 3800+ (200x10=2000mhz) 2GB DDR 400 (4 dimms) ASUS 7800GTX (reference clocks) 200GB Maxtor HD And here are the benchmarks: Standard 3dmark06 Run, small (16%) gain moving from PCIe2x to PCIe16x. Game breakdown follows: Percent Gain for GT1: 16%,GT2: 15%, HDR1: 15%, HDR2: 33% ~15% drops across all games except HDR2 which shows a relatively large drop of 33%. My guess would be that the 256MB frame buffer was filled making it necessary to move additional data over the slow PCIe2x bus. You can't run the HDR tests with AA enabled but here are game tests 1 & 2 with 4x AA @ 1280x1024. Notice that with AA enabled the difference between 2x and 16x in GT2 nearly triples to 43%, again the added sampling most likely demanded more than the 256MB available in the local frame buffer on my 7800GTX. Please note differences of +/-5% should be ignored due to measurement inaccuracy. Taking this into light we see that oblivion gains absolutely nothing going from PCIe2x to PCIe16x. I was running near maximum settings @ 1280x1024. Benchmarks were run @ 1280x1024 maximum settings. As we can see stalker gains about 50% in the Buildings time demo and 40% in the Outdoors timedemo. Feel free to comment on the data, I'm not quite sure why I'm not seeing a performance drop in oblivion. It could be that at these settings the game is largely CPU limited. I'll do a bit more benching later tonight, I might throw in 3dmark05 too if I have time. And yes I know this is largely pointless but I was curious haha. P.S. I've got a question for the 3D buffs on this site: What exactly is moved across the PCIe bus during game play? |
|
|
|
|
|
#2 |
|
Mord's imaginary friend
Join Date: Jan 2004
Location: PT, EU
Posts: 3,506
|
This is not pointless. In fact, it was quite helpful to me. I'm also surprised by the relatively low performance impact.
__________________
The optimist proclaims that we live in the best of all possible worlds, and the pessimist fears this is true. - James Branch Cabell |
|
|
|
|
|
#3 |
|
Senior Member
|
Not surprised by the numbers from Stalker, as this title is taxing quite heavily the video memory. Try again with 512MB board.
__________________
Apple: China -- Brutal leadership done right.
Google: United States -- Somewhat democratic. Microsoft: Russia -- Big and bloated. Linux: EU -- Diverse and broke. |
|
|
|
|
|
#4 |
|
Darlek ******
Join Date: Jun 2004
Posts: 9,503
|
why do manufacturers insist their cards are x16 only
for example ati insist the x1900 needs 16 lanes even though it works fine in a 4 lane slot the 8800 doesnt work in a 4 lane slot but i suspect its because nvidia has told the card not to and not because it cant |
|
|
|
|
|
#5 | |
|
Senior Member
Join Date: Oct 2006
Location: Germany
Posts: 1,003
|
Quote:
PCI Express Scaling Analysis The problem with some ASRock mainboards could be, that they are cheap. http://www.asrock.com/mb/overview.as...ual-VSTA&s=775 They miss some PEG/PCIe requirements, and it's possible, that 8800GTX doesn't get enough power.
__________________
Hail Brothers and Sisters! Coranon Silaria, Ozoo Mahoke Eta Kooram Nah Smech! Find Chuck Norris. |
|
|
|
|
|
|
#6 |
|
penguins
Join Date: Feb 2004
Posts: 13,978
|
It's an interesting investigation. If you're willing, perhaps it would be worth a look to benchmark Oblivion with those high resolution texture packs.
Hm...I was under the impression that the main use of the PCIE bus was in transferring textures and communication with a second GPU (duplicating textures). |
|
|
|
|
|
#7 | |
|
Member
|
Quote:
![]() I think this makes a little more sense. The bench consisted of an outdoor scene with lots of grass & tress and a little water. As we can see there's a drop in performance with AA but not with HDR, this would make sense given the memory requirements of each. |
|
|
|
|
|
|
#8 | |
|
Darlek ******
Join Date: Jun 2004
Posts: 9,503
|
Quote:
and my board P4Dual-880 http://www.asrock.com/mb/overview.as...l-880Pro&s=478 Last edited by Davros; 25-Apr-2007 at 19:24. |
|
|
|
|
|
|
#9 |
|
Member
Join Date: Jun 2004
Posts: 168
|
Does that mean in theory if i have 1 - 2Gb Video Memory i would get neligble performance gain from PCIe2x to PCIe 16x?
|
|
|
|
|
|
#10 |
|
Member
|
Yep. According to these benchmarks anyway. It also means I could run SLI on my MSI K8N Neo4 Plat without a big performance hit, which is pretty sweet. I'd just need to mod my chipset to NF4 SLI from NF4 Ultra with the pencil trick then I should be good to go.
|
|
|
|
|
|
#11 |
|
Entirely Suboptimal
Join Date: Mar 2003
Location: WI, USA
Posts: 6,848
|
My Gigabyte DS3, like many 965P boards, likes to drop to PCIe 1X on initial POST when overclocked. It needs a reboot to sort itself out. I'll confirm that PCIe 1X is a major bottleneck for an 8800.
|
|
|
|
|
|
#12 |
|
Darlek ******
Join Date: Jun 2004
Posts: 9,503
|
Heres a thought if i were to buy a 16 lane board will i regret it one day because the latest cards perform better with (or even require) 32 lanes
i hope pci-e doesnt end up like agp i hate replacing gfx cards on old mboards because i have to go trawling the net to find out what voltage the agp slot uses and if the card i have is compatable with it |
|
|
|
|
|
#13 |
|
Moderate Nuisance
Join Date: Feb 2002
Posts: 4,654
|
Great post, Panda. FYI, THG investigated the same thing, only with an 8800 and X1900 at x1, x4, x8, and x16: PCI Express Scaling Analysis.
|
|
|
|
|
|
#14 | |
|
MSI Man
|
Quote:
__________________
I miss you CJ, 1976 - 2010 |
|
|
|
|
|
|
#15 | |
|
Member
|
Quote:
|
|
|
|
|
|
|
#16 | |
|
Senior Member
Join Date: Jul 2005
Location: State of Illusionism
Posts: 2,091
|
Quote:
__________________
Everything's Eventual Oedipus On The Orpheum Circuit! |
|
|
|
|
|
|
#17 |
|
Gamerscore Wh...
Join Date: Jan 2002
Posts: 12,956
|
Yes, physically they are the same, however there can be up to twice the bandwidth per pin. PCIe 1 -> 2 will be similar to, say, AGP2x -> 4x.
|
|
|
|
![]() |
| Thread Tools | |
| Display Modes | |
|
|