8800GT @ PCIe x8

[AMDZen]

I have a DFI nF4 motherboard with two so called "x16 PCIe" slots for SLI. In the default configuration, the top slot is x16 and the bottom is x2. There are large jumper blocks between the two slots, when changed to the other configuration, each slot becomes a x8 slot for SLI.

OK so I just bought the EVGA 8800GT SC at some 9" long. When installed in my top slot, the thing is so long it blocks two of my SATA ports. I only have 4 and I use all of them so I figure I'll install in the bottom slot where it doesn't block ports. After horrible performance and using GPU-z I figured out about the whole x16 and x2 thing. So I jumped the board to the x8 and x8 configuration for the time being.

My question is, do you think performance will matter much? I ordered a PCI SATA card to use, thinking it would matter on performance - however people on the DFI forums seem to think that x16 and x8 I will see very little if any performance difference at all. I find this kind of difficult to believe, true with G80 cards - perhaps, but this is brand new tech on PCIe 2.0 and all that so I'm thinking times may have changed and that the new cards are utilizing more of the slots bandwidth.

What do you guys think? Does anyone have any direct experience with this issue?

 

[customcoms]

I will have to say I agree with the other posts over at DIY-Street. I don't think there is a drop so to speak, considering this card is much more CPU limited by our dual core AMD processors than by the available PCIe bandwidth. If it helps, I am running my card in the bottom slot, x8 (same exact board, same exact issue, I need 3 SATA ports) and getting 9500 3dMark06 points, which is higher than some configurations with an 8800gtx on 3DMark's database. No, its not the 12,000+ marks this card is capable of, but its close enough considering the rest of my system to say x8 isn't a serious issue. I am also going to run the crysis benchmarks, since I have comparison numbers to use (benchmarked my 8800GTS 320mb in crysis only, that was in the top slot). If I see a similar performance increase as the review sites, it will at least somewhat confirm x8 isn't the bottleneck. I am going to try to run real x8 vs. x16 tests but that might not happen depending on my laziness.

 

[AMDZen]

I appreciate your response, especially since we are running very similar configs. I am going to do some testing as well and see if I notice any difference from x8 and x16 in gaming. I won't use 3dMark since its not real world, however I think the HL2, FEAR, Crysis and other games that have the benchies built in show a bit more real world gaming type numbers

 

[Denithor]

That's exactly what I was going to recommend, benchmark the card in the x16 slot and then compare at x8 speed as well. Please post your results so the rest of us will benefit from your work.

PCI-E x4 provides slightly lower bandwidth than AGP 8x (comparison on ASRock 4CoreDual) so at x2 you are certainly screwed but x8 should be just fine (current cards barely saturate the bandwidth of AGP 8x let alone PCI-E x16).

 

[customcoms]

My numbers in Crysis, all settings at medium, 1680x1050: 8800GTS 320mb (x16) (163.75): Average of around 32 FPS
8800GT (169.02): Average of 47 FPS

Thats a 15 fps increase. Yes, the GT drivers are newer but are the oldest drivers compatible with the GT and I can't re-run the GTS benchmark. Both cards are/were overclocked to stable limits (no crashes in long gaming sessions, ATITool stable for hours). Basically, I'm seeing the same percentage increase as these guys: http://www.tweaktown.com/artic...arks_crysis/index.html

I am also getting practically the same FPS (if you interpolate the 1600x1200 scores) at the same settings, and they are running a Q6600 @3ghz, which is a significant processing speed increase over my lowly Opteron 165 @2.6ghz, which shows that games are still gpu limited as usual.

And I agree that 3DMark scores are useless for all but comparative benchmarking (aka, determining if x16 vs x8 matters). Based on my results, I am temporarilt concluding that x16 vs x8 has little tangible benefit. If your actual test results show otherwise (like more than a 5fps increase) I may be tempted to switch- did you order any right angle SATA connectors to see if they would fit under the GT?

 

[AMDZen]

I did order 2 right angle SATA cables from SVC yesterday as well as the SYBA 4-port PCI card from Newegg. I'll let you know if the right angle cables help at all.

Just to clarify, you said

My numbers in Crysis, all settings at medium, 1680x1050: 8800GTS 320mb (x16) (163.75): Average of around 32 FPS
8800GT (169.02): Average of 47 FPS

Does this mean you had the GTS at x16 but the GT at x8? Just want to clarify

Also - we both must have 22" monitors so thats good as well, I am running at 1680x1050. I installed 169.13 and one issue I'm having is in CSS I can't select 16:9 and 1600x900 and windowed. I play this game in a windowed environment because I'm always multitasking. Now with this videocard 16:9 is grayed out and I can only select 16:10 and 1680x1050 res. I don't suppose you have CS:S or HL2 or anything that you could test this for me? I am thinking it might be the drivers, but I don't want to revert back to 169.02 and then back to 169.13 if it doesn't help - and its something to do with the 8800GT

My 7800GT did this fine.

 

[dgevert]

I don't believe that the current graphics technology is fast enough to be bandwidth limited by the PCI-E bus yet. x8 should not be any slower than x16 right now.

 

[Tempered81]

AMDzen, have you noticed a difference in x8 vs. x16?

 

[AMDZen]

I don't know because I haven't tried it at x16 yet. I'll be doing this tomorrow probably or this weekend (like sat or sunday)

I'll update this thread when I do test and I'll advise

 

[vanvock]

This is great as I'm in the same boat. I've seen a few threads here & there on this with opinions going both ways but no actual testing. I have a custom cooler on the chipset that I really don't want to replace but if it means a performance hit I would. I'll be looking forward to any results or further input.

 

[Denithor]

Here's what someone with this setup should do: test 3dmark06 (I know, I know...bear with me) with the card in x2, x8, and x16 mode just to see how they compare. That would give us a great idea of how much bandwidth these cards actually use and therefore an idea of whether there is any point at all to PCI-E 2.0 (I personally doubt it).

 

[customcoms]

To clarify: Crysis timedemo, average of the three scripted runs, all settings on medium, 1680x1050,rest of rig in sig.

A) 8800GTS, 163.75 drivers, x16, 612 core/1600 shaders/900 memory, min of 25fps, max of 40fps, average of 32 fps.

B) 8800GT, 169.02 drivers, x8, 675 core/1674 shaders/950 memory, min of 40 fps, max of 53fps, average of 47 fps.

As you can see, thats a 15fps average increase and the lower drop is significant. And my 8800GTS had a pretty darn significant overclock on it. I played crysis on three cards: the 8800GTS (beggining levels), 6600GT (middle levels...don't ask, just don't rush out and buy a 6600GT for Crysis), and now the 8800GT (later levels). I had to turn settings to low by the end of the game with the GT...I can say that you won't enjoy this game without AT LEAST GT performance levels (and my GT performs nearly on par with GTX's).

AMDZen,

I am also running a 22" monitor. If you could post the rest of your specs that would be great (cpu speed namely)! Mine are in sig! I also play CS:S, and on the 169.02 and 163.75 drivers you have to select widescreen 16:10 (not 16:9 like I thought as well) to allow the 1680x1050 option...its great to be able to crank everything to the max and still be in the 100+ fps regions! And I can play windowed. I have considered migrating to the 169.21 drivers but I am no longer so quick to jump on new beta drivers with Vista issues, rivatuner compatibility, and overclocking thrown into the mix...hence I tend to wait for WHQL certified drivers, apparently the 169.21's are WHQL candidates but sometimes nVidia makes a last minute change.

Vanvock,

I too have an aftermarket chipset cooler, the Evercool VC-RE. Running FSB speeds in excess of 290 24/7 demands extra cooling...we are also probably beta users for running 2 year old hardware with such massive overclocks. Anyway, the clearance is tight but I think it will fit, didn't play around too much with the x16 slot because I would need to rearrange my HDD's SATA arrangment, and even though one of my HDD's is not being occupied atm, its a pita because I have vista and xp dual booting with boot.ini files spread between two physical drives and 3 partitions total.

Denithor,

That doesn't really answer the question of PCIE 2.0 speeds, since a 2.0 slot is theoretically double the speed of a 1.0/1.1 slot. It would only answer that question if no difference is found between x8 and x16...I wouldn't bother with x2 as that test has been done at diy-street.com and found to be a significant degradation on 7 series hardware.

I will maybe try and do my own tests if I can find the will to play with my hdd's to get the card in the x16 slot and boot into windows at the same time. Now I know what it is like to benchmark hardware for a living....a major PITA.

 

[vanvock]

customcoms,
Mine is actually a Zalman passive sink that I attached a duct to with a remote fan so if I use the x16 slot I would have to go back to the low profile type, which isn't that big of a deal but I would like to keep the all exhaust setup if I can.

 

[AMDZen]

Originally posted by: customcoms
I am also running a 22" monitor. If you could post the rest of your specs that would be great (cpu speed namely)! Mine are in sig! I also play CS:S, and on the 169.02 and 163.75 drivers you have to select widescreen 16:10 (not 16:9 like I thought as well) to allow the 1680x1050 option...its great to be able to crank everything to the max and still be in the 100+ fps regions! And I can play windowed. I have considered migrating to the 169.21 drivers but I am no longer so quick to jump on new beta drivers with Vista issues, rivatuner compatibility, and overclocking thrown into the mix...hence I tend to wait for WHQL certified drivers, apparently the 169.21's are WHQL candidates but sometimes nVidia makes a last minute change.

Specs:

Gaming Rig
22" Samsung 226BW "S"
Antec Nine Hundred Case
AMD Opteron 170 E6 @ 2.82 Ghz
Thermalright Ultima-90 w/ 92mm Panaflo
DFI LanParty UT nF4-D w/ SLI and SATA II Mod
EverCool VC-RE NorthBridge Fan
2x1GB G.Skill PC-4000 DDR500 @ 256 Mhz
Corsair 620HX v2.2 PSU
EVGA 8800 GT SC @ 680/1960/1700
SoundBlaster X-Fi XtremeGamer

I got my CS:S issue fixed.


My SATA card is schd for delivery today so I definetely plan on doing some testing. I'll let everyone know my results. It probably won't be tonight because I live in Denver and I'm a huge Denver Broncos fan, should be enough said. But I'll probably get to it on Saturday or Sunday. Hopefully my angled cables will come in stock too tomorrow and I can test those as well

EDIT: Oh yea and I forgot to mention, as you can see I have the EverCool as well, and the clearance on that is fine.

Originally posted by: Denithor
Here's what someone with this setup should do: test 3dmark06 (I know, I know...bear with me) with the card in x2, x8, and x16 mode just to see how they compare. That would give us a great idea of how much bandwidth these cards actually use and therefore an idea of whether there is any point at all to PCI-E 2.0 (I personally doubt it).

I might do that. I think Crysis will be the best test to see. I haven't even looked but Crysis has a test in it right? customcoms says "demo" and all I have is the full version but I assume its in there. I've OC'd my card now and its running like a champ. I probably won't do testing at x2 although I can understand why that might be useful. I'll probably do 3dmark though as well as Crysis. Stay tuned guys

 

[AMDZen]

OK so I've completed my testing. Contrary to what I thought, and very strangely - when the card was running at x8 it was actually faster. By only a bit on Crysis but fairly significant on the Source engine. I'm running 169.21

Running Crysis at 1680x1050 with all settings on medium I had an average FPS faster on x8 with 47-48 over multiple tests of the Island. The x16 tests came out at around 46-47 over multiple tests of the Island. The minimum was at 28-29 on x8 and 27-28 on x16. The Highest FPS was 68 on x8 and 65 on x16. Weird eh?

The CS:S video stress test with everything maxed out, Vsync was off, and the AA was AF was at 4x each turned out 218 FPS on x8 and 200 on x16. A rather noticable difference on this one. Still quite weird.

So you probably think I would not only say running x8 is fine, but actually has an advantage over the x16 slot. I don't know if this is just a DFI type thing (my entire specs are in the above post) or what but I did verify over multiple tests of these two games. I didn't do 3dmark just these two tests. But NO hold on, I noticed some other things during my testing.

On x8 my max OC was 680/1960/1700 which is shown above. Whenever I went over this, most every game still ran fine except the CS:S stress test as well as CS:S, HL2, Ep2 and Portal would all freeze when playing.

On x16, I am currently running 700/2000/1744 which is the same as SSC with no problems at all. I don't know if this is a driver thing, as I am running a so called "beta" driver however it is an WHQL candidate. But I have no doubts at all that I could overclock this card even further but just haven't done so yet. I also noticed something else - games load faster. I didn't go back to x8 and time anything to see how much quicker, but when the Crysis demo was loading for example - it was noticably quicker to load. I notice this with all the games as well. Also, every game would not identify my max settings correctly on x8, I had to raise them manually which is no big deal at all - however on x16 games more correctly identify max settings.

Take what you want from this testing but I'm leaving mine in x16, the OC being 100% stable at higher speeds and the quicker load times off set what was really just a 1 fps advantage in Crysis for x8. And of course, as you might expect - the OC took my FPS up around 50 average FPS in Crysis. I also got my SATA card, so its no biggie for me. I didn't get my angled cables yet but will probably just keep them for extra cables. If someone needs me to I can test these early next week when I get them and advise.

Bottom line. If your motherboard has issues with this card blocking ports, don't hesitate to use the x8 port instead.

 

[RanDum72]

I have a MSI P6N Diamond with 2 PCIe 16x slots and 2 PCIe 8x slots. Tried my EVGA 8800GT 512mb superclocked edition on the 16x and 8x slots. Difference? Only ran 3dmark06 and the 16x slots gave me about 25-30 points higher on multiple runs. Virtually no performance benefits that you will notice. But being anal, I like 8800GT running on a PCIe 16x slot so I rearranged my harddrives.

 

[vanvock]

Thanks for the feedback

 

[customcoms]

AMDZen,

Thanks for running these tests and confirming my thoughts! I really think we are cpu bound, considering anand is cpu bound on Crysis with a quad @3.3 and 3 8800Ultra's (yeah, that test rig is insane but so is the cost/performance ratio). I'm looking at better cooling to push for 3ghz's-I'm pretty sure my cpu is capable if I can keep the temps in check (aka under 55 C full load both cores). Motherboard stability might become an issue, so I'll be testing that.

Interesting to here about the card overclocking limitations, as your bottom slot limits are pretty much the same as mine. This makes me want to drop her in the top slot and see what she's really capable of, although once again I'm hitting 71 C under load and I think these run happier in the 60's under load-time for more aftermarket cooling perhaps. Also, I think this is an SLI driver issue, or possibly the bottom slot just can't give the card as much juice as the top slot (I'm assuming you have all the funky DFI power cables plugged in, considering the overclock et all.). What would be interesting is if you hit the same clockspeed limits on the card in the top slot with it configured as an x8 slot...

If you could do the right angle SATA cable test for me it would be GREATLY appreciated!!!

 

[aussiestilgar]

I guess its safe to assume if you have a board that can only do PCIe x16 and x4, SLI is not a good option?

 

[customcoms]

Anandtech did crossfire tests on P965 asking exactly that. Technically SLI won't work on that board without outdated, hacked drivers.

 

[Spike]

Thanks for all the info guys. After reading this I am now worried since I finally get to open my new 8800GTS 512 at Christmas (wife took it away since it's a present) and I did not even think about it having issues with my mobo and SATA wires. Like alot of you I have SATA drives (though only 2) and the evercool VC-RE on my northbridge so clearance is an issue. Since I have not done the mod my board is not SLI so I can't just put the card in the bottom slot without a signifigance decrease in performance...

I guess I'll just have to give it a try and see if everything fits... things are all-ready tight enough in my P180 so I am not looking forward to this.

 

[Karandar]

Just don't do what I did...replaced my Nbridge cooler with a 1st gen Thermalright hr-05 passive cooler (works great btw!) It blocked the PCIE 1 slot - near the CPU, so i installed my vid card into the second PCIE slot, and ran with that...recently upgraded to the superclocked EVGA 8800gt, I was remarkable underwelmed at the improvement in performance over my old X800 GTO2 ...Just found out today, I have been running my video cards on PCIE slot 2 at 2x speeds!! Doh!! changed the jumpers over to SLI,(yes i only have 1 card) and left the SLI option in the BIOS to auto(this makes both PCIE slots 8x speed), reinstalled the video drivers and my PCmark3 scores went up 45% from 24545 to 35429!! The PCmark3 tests the gpu's more than the cpu's that 3dMark6 does. That doesn't help us cpu bound folks with athlons much...
My chipset is actually a dfi nforce4 ultra-d with the SLI mod done.
Also, anyone wanting to use slot one, but have issues with the Northbridge cooler being too big, get a thermalright HR-05 SLI edition - 25 bux, and it should work for ya. weren't avail when i 1st bought mine. they run 5+ degrees cooler than the stock dfi cooler/fan did on the northbridge...

Cheers! Kar