Quad Sli vs. Single Co-Op - All Benchies In! - Overclock.net

brettjv · 07-03-09

So, saw a Co-Op at Fry's, thought it'd only be right to check it out ... a little quad-sli action for the 4th of July Weekend! If I decide I like the Co-Op better, I may decide to keep it and sell my (very nice at overclocking) Sandwich version ...

I doubt seriously that I'll keep the quad-setup, this is just to play around with. But I'm gonna do some benches of my own to find out which way I wanna go afa whether or not I'll keep the Co-Op and sell the sandwich, or just return the Co-Op (in 29 days

)

Coming up tonight and tomorrow, my planned benches will be:
(all run on both single 295 and dual 295)
Vantage - High and Extreme
Stalker CS Bench 1920x1200 DX10 Maxed at 2xAA, 4xAA, 8xAA
Crysis - Warhead Avalanche Flythrough DX10 Maxed at 1920x1200 @ 2xAA, 4xAA, 8xAA
Far Cry 2 DX10 Maxed

I'll consider running other benchies if they're free or I own them ... lets see whether doubling up on 295's gives us some higher playable settings, shall we?

Unfortunately I promised the girlie I'd watch friggin Madagascar 2 with her, so ... it'll be a while before I get benchies rolling, but until them ... some pics!

All results are now in ...

1. Looks like 40% fan on the CoOp is about 500RPM slower than on the sandwich

2. Teh CoOp has no trouble hitting the same OC as the sandwich did. ALL BENCHES ARE AT THESE CLOCKS (cause this ain't StockClock.net, now is it?)

3. The bottom (new) card is running considerably cooler at idle. Part of this is no doubt because it's on the bottom ... we'll see when I run it alone how much cooler it really runs, but it's looking very nice ATM. NB temps and board temps are about 4-5 higher that normal, no doubt because of the four gfx cards and their heat, especially around the southbridge which heats the NB via heatpipe.

4. LET THE BENCHIES BEGIN!!!

All tests were done at 675/1476/1224.

Here we have the results from the Vantage High test, phys-x is off. I didn't run Performance mode because this IS quad-sli ... so who really cares about 19" monitor resolution performance?

Quad-Sli scores:

And on the single Co-Op:

5. Vantage Extreme benchies

Quad-Sli Scores:

Single Co-Op at same clocks ... scaling on quad-sli is pretty phenomenal on the Vantage tests, but I think we knew this already ...

6. And onto the STALKER CS benchies. Here we have quad-sli results with the test completely maxed out. The Min FPS is pretty bad, but these really low FPS only happens in ONE place, for a second or two, and that is first coming into the ClearSky camp near the end of the bench. All that foliage and all the stalkers loading I'd imagine. Stalker with SSAA (i think that's what the second box is saying) is pretty crushing in terms of memory usage ... I'd expect something like Tri-SLI 280's to be quite superior on this particular test

Unfortunately, as it turns out, CS was not scaling on quad-sli, for me at least. Here are the scores for the single Co-Op on the same tests:

7. Here we lowered expectations a bit with CS ...

BTW here's what temps looked like during this test ... the newer card runs 10+ degrees cooler on load. Honestly it is SO much cooler, I almost wonder whether it's actually working or not ... But since the new card doesn't have the Volterra VRM, I can't check amperages (or a bunch of other detailed stats) in the RT graphs like I can which the Sandwich ... DEFINITELY a minus for the CoOp in my book.

And the quad-sli results maxed out with 2xAA:

The 2xAA tests look nearly identical on the Co-Op, so I won't post those results ... I don't know if it's a Win 7 thing, or what, but quad-sli isn't working on the Stalker Bench on my rig ... pity ...

7. And what benchmarking run would be complete w/o some Crysis and Warhead runs...

Warhead Benchies

Quad-SLI 675/1476/2448

DirectX 10 ENTHUSIAST 2X @ Map: avalanche flythrough @ 8 1920 x 1200 AA 2xx
==> Framerate [ Min: 38.69 Max: 72.99 Avg: 59.54 ]
DirectX 10 ENTHUSIAST 2X @ Map: avalanche flythrough @ 8 1920 x 1200 AA 4xx
==> Framerate [ Min: 27.66 Max: 74.33 Avg: 52.68 ]
DirectX 10 ENTHUSIAST 2X @ Map: avalanche flythrough @ 8 1920 x 1200 AA 8xx
==> Framerate [ Min: 25.92 Max: 71.72 Avg: 49.21 ]

Single Co-Op 675/1476/2448:

DirectX 10 ENTHUSIAST 2X @ Map: avalanche flythrough @ 8 1920 x 1200 AA 2xx
==> Framerate [ Min: 24.20 Max: 38.13 Avg: 31.53 ]
DirectX 10 ENTHUSIAST 2X @ Map: avalanche flythrough @ 8 1920 x 1200 AA 4xx
==> Framerate [ Min: 21.72 Max: 38.25 Avg: 29.86 ]
DirectX 10 ENTHUSIAST 2X @ Map: avalanche flythrough @ 8 1920 x 1200 AA 8xx
==> Framerate [ Min: 19.74 Max: 37.03 Avg: 28.65 ]

The scaling on Warhead is pretty impressive in all regards w/quad-sli.

Crysis - Benchies

Quad SLI Score (675/1476/2448)
7/3/2009 11:52:07 AM - Vista 64

Beginning Run #2 on Map-harbor, Demo-Assault_Harbor
DX10 1900x1200, AA=8x, Vsync=Disabled, 32 bit test, FullScreen
Demo Loops=2, Time Of Day= 5
Global Game Quality: VeryHigh

Run #1- DX10 1900x1200 AA=4x, 32 bit test, Quality: VeryHigh ~~ Last Average FPS: 48.51
Min FPS: 26.39 at frame 3940, Max FPS: 82.15 at frame 1749
Run #2- DX10 1900x1200 AA=8x, 32 bit test, Quality: VeryHigh ~~ Last Average FPS: 49.76
Min FPS: 28.04 at frame 3922, Max FPS: 86.12 at frame 2040

Crysis - Co-Op Scores (675/1476/2448):

Run #2- DX10 1900x1200 AA=4x, 32 bit test, Quality: VeryHigh ~~ Last Average FPS: 38.50
Min FPS: 24.89 at frame 2904, Max FPS: 52.39 at frame 3245
Run #3- DX10 1900x1200 AA=8x, 32 bit test, Quality: VeryHigh ~~ Last Average FPS: 44.03
Min FPS: 29.26 at frame 2921, Max FPS: 59.76 at frame 972

Scaling in Crysis is not as impressive here as I saw in Warhead ... strange. Also strange is the way that 8xAA seems to run better than 4xAA with both the quad-sli, and (especially) the single Co-Op.

Crysis - Temps in Quad-Sli (remember I'm in AZ and it's summer)

Here are the temps during the Crysis run ... again we see the CoOp keeping temps down by 10+ degrees vs. the sandwich like we did on the CS benchies ... perhaps those low temps were not a result of quad-sli not working after all?

We should remember though that the CoOp is on the bottom ...

But honestly I don't see how it doesn't work out that this thing is a LOT cooler running no matter what.

Stay tuned for some results from the CoOp all on it's own ... temps, benchies, max overclocking, etc ...

And for those of you who were wondering ... My Enhance-built CM RealPower 850W had absolutely ZERO problems handling quad-sli with a heavily OC'd processor. I never had a crash, freeze up, or any other problems whatsoever during any of these tests.

Except ... some of the benchies did seem kinda stuttery, esp. Vantage Extreme and Warhead. I think the microstutter may be 'for real' ... but I'm gonna have to do some gaming and see if I see it in real life ...

SINGLE CO-OP Basic Fan/Temp/OC Tests

Next up: Graphs from a Furmark Run ... The Co-Op hits 100% fan at a mere 80C, and the card doesn't seem inclined to get any warmer under these conditions.

Thus, the fan ramp-up is MUCH steeper on this card vs. the sandwich, which doesn't hit 100% until at least 90C. Even at full bore, the Co-Op sounds about the same as the sandwich fan does at around 60%.

The sandwich will also hit 90C under the same Furmark test pretty easily, so I'm going to say that the Co-Op loads out at about 10-12C cooler than the sandwich. Idle temps are running at about 6C lower than the sandwich as well (approx 56C vs 50C ... it IS summer in Arizona, so ...)

At idle, the Co-Op is noticeably quieter (I'd almost say 'hugely quieter' ...not quite, but almost) than the sandwich. On load they are close to the same, volume-wise ... only the Co-Op is running much cooler at the same noise level. So ... yeah. It's cooler and quieter overall, without a doubt. And I'm not seeing my other temps being affected by the dump of warm air in the case.

Our friend Zip (from BFG) tells us below that the Co-Op actually dumps about the same amount of warm air into the case as the sandwich, so ...

Max Stable Overclocking on the Co-Op is shown below. An extra strap on the shaders vs. the sandwich, and, even better, a solid 45MHz bump in the core clock speeds. I should add that gpu temp overall was virtually unaffected by this fairly monster overclock from stock speeds.

On Edit: Turns out this core clock wasn't entirely stable. Dropped it down to 712 and played WH for about a half-hour, no problems. So the final stable OC is 712/1512/2484.

Temps from after a two runs of 3x on the Crysis bench at max overclocks ... the one extra bump is from an aborted attempt at a 729MHz core clock, which caused crysis.exe to quit to desktop.

Final Notes:

Well, a few things. Stalker CS seems to have major microstutter when I try to play it. But quad-sli doesn't seem to actually be working, so I'm not too sure what to make of this phenomenon.

I also gamed a bit with Warhead, both in quad-sli and immediately after with the Co-Op. Honestly, it felt smoother on the single Co-Op overall even though the frames were lower ... and they were not as much lower in actual gameplay as they appear on the benches. I can't say I noticed microstutter with quad-sli in WH, not like with CS, but honestly the gaming experience just wasn't any notably better on quad-sli.

All in all, I have to say I'm not impressed with quad-sli. As so many have said, it's great if you wanna break benching records, but if you just want a good gaming experience, you definitely don't need it, in fact, you may it less enjoyable. Esp. with $500 less in your wallet.

The GTX295 Co-Op, in general, though, is a very different story. On the OC, I saw 45MHz (edit: correction, 37MHz) extra on the core, 36MHz more on shaders, and memory is the same. The idle temps are 6-8C lower on the Co-Op, and load temps are a solid 10C lower, and even more when you're playing something 'normal' (i.e. not Clear Sky or the Crysis series). Idle noise is considerably lower, and load noise is slightly lower, even whilst holding temps much lower. I don't see any noticeable increases in temps caused by the heat dumped into the case, either.

Aside from the fact that it seems more cheaply built than the Sandwich (it's a whole lotta plastic is the truth, and thus lacks the solid metal feel and rubberized texture of the Sandwich), along with the lack of the software-programmable Volterra VRM and the nice measurements it provides in RT, the Co-Op is pretty much a straight WIN in my book.

Not sure if I like it enough to try to sell my sandwich and keep it (and the lo$$ that would entail) but FOR SURE if I was starting out with nothing, the Co-Op is seriously worth your consideration.

With STABLE overclocks that approach 25% (!), the Co-Op is pretty sweet.

On Edit : I am gonna throw the sandwich up for sale after all ... it's worth $100 to me for the quietness and OC-ability of the Co-Op vs. the Sandwich ... although I hate to lose the Volterra ... we'll see what happens ... no buyers in 28 days, and the Co-Op goes back ...

P.S. Madagascar 2 is actually pretty hilarious. Sacha Baron Cohen and the Penguins are awesome ...

**grunion** · 07-03-09

Oh

You know the drill, temps, oc capabilities.....etc

Also how much air is dumped into the case?

**metallicamaster3** · 07-03-09

On God's green Earth... O_o...

$530 on a GPU. Sheesh, wish I had that money to burn.

Thanks for the porn

Slightly skewed · 07-03-09

Oh my dear god, finally someone with a pair! Of balls that is. In for results. Also, air flow is balls in that case. I thought you were an experienced system builder?

ltulod · 07-03-09

nice nice.

Slightly skewed · 07-03-09

I hope it's ok to post these as a direct comparison. I just did them. Using latest WHQL driver. Note 5 passes for better average.

Nelson2011 · 07-03-09

Do they work in sli together?

sgdude · 07-03-09

Quote:

Originally Posted by Nelson2011

Do they work in sli together?

yes.

Crooksy · 07-03-09

looks pretty hectic. Nice job

brettjv · 07-03-09

Quote:

Originally Posted by Slightly skewed

Oh my dear god, finally someone with a pair! Of balls that is. In for results. Also, air flow is balls in that case. I thought you were an experienced system builder?

I see ... so ...

1x120mm (in drive cage from antec 900) in the bottom 3x5.25 bays (intake)
1x200mm in front of case (intake)
1x120mm in bottom of case (intake)
1x120mm in side of case (intake)
1x120mm in rear of case (exhaust)
1x200mm in top of case (exhaust)
+ the exhaust from the graphics cards ...

you consider this to be 'balls' airflow, and imply that I don't know what I'm doing?

Interesting ...

Yeah, I'll be posting some benchies ... but this ain't a competition like you've been itching for since the last thread died off

I already know your setup is better.

And I'm really not interested in finding myself convinced in a positive away about quad-sli, truth told, I hope to NOT like it

This Co-Op came from Fry's ... so I got 30 days to play with it, and if I like it better than the my old 295, the sandwich may go on the market

		Overclock.net - Overclocking.net > Graphics Cards > NVIDIA
Quad Sli vs. Single Co-Op - All Benchies In!

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)