Is this worth it? SLI vs 5xxx series. -Updated-

The_Blood_Raven

Distinguished
Jan 2, 2008
2,567
0
20,790
Hey, I'm thinking of getting my cousin a new 5xxx series card for his birthday in a little less than a month. However, I'm not sure whether the 5870 is worth it over the 5850 or if either are worth it over his current setup.

His rig (made by me):
E6750 @ 3.2 Ghz (possibly 3.6 again if I can get it back up there)
4GB of G.skill DDR2
XFX 680i LT
2x EVGA 8800 GTS 512mb

So am I far off to assume that the 2 8800 GTS 512s in his rig is slightly lower than a single GTX 275? When the GTX 280 came out it wasn't quite as fast but the extra memory and the newer drivers should have pushed the GTX 280 well above them and the GTX 275 is very similar.

Thanks to the limited memory and bandwidth he can't play any game with AA at his native res 1920x1200 and Crysis is only really playable at medium at his native res which has always seemed odd to me. Also Fallout 3 maxed at that res hitches a bit even without Vsync or AA, which always seemed odd to me aswell but I chocked that up to low memory bandwidth/VRAM and low minimum framerates thanks to SLI.

So do you think a 5850 would be a large increase in performance, do you think the 5870 would be a far more substantial upgrade, or do you think neither is much of an upgrade? I have a feeling that it should add a good 30% boost without AA, but far more, in the range of 50%+, with AA enabled. Am I right or am I overestimating the 5xxx series or underestimating the 8800 GTS 512s. I think the low memory bandwidth, VRAM, and the SLI problems have made his setup close to being inbetween a GTX 260 and a GTX 275. Though i have also seen people compare the GTX 260 very favorably to this setup as well, so I don't think it is just me.

Also do you think that processor will bottleneck his cards? At 3.2 Ghz playing at 1920x1200 shouldn't be a problem and the high res gaming should nullify any short comings of the CPU in every game except for GTA 4.

Thanks for any help, I can't make up my mind on this present.
 
His rig has some driver issues or something.

The 8800GTS 512mb is a G92 card that is almost equal to a 9800GTX.

8800GTS 320/640 < 8800GT/9800GT < 8800GTS 512 ~/< 9800GTX/GTX+/GTS250

Two 9800GTXs are better than a GTX285.

Your cousin's rig should be able to handle any game except Crysis at nearly max at 1920x1080.


Realistically, since a 5850 performs maybe 20-30% better than a GTX285, and 2x8800GTS 512's also perform 20-30% better, there wouldn't be any improvement by upgrading to a 5850.

There is something wrong with the drivers I'd say...
 

The_Blood_Raven

Distinguished
Jan 2, 2008
2,567
0
20,790
I see your point and I have wondered at it myself, but when I test it via the Far Cry 2 benchmark it scores the same as a GTX 280 which is about right. You must remember 1920x1200 is a high resolution for 512mb of GDDR3 on a small 256-bit bandwidth. The G200s are limited by a 1GB 512-bit bandwidth in some instances!

When at 1280x768 Crysis can be run on high with very high textures, that is about right.

I thought something was wrong too and kept the PC for weeks of full blown testing until I found nothing wrong. All drivers were fine, re-installations of windows did nothing, upgraded the memory, tested the hard drive, prime95'd the CPU to death, used OCCT and 3dmark vantage to test the graphics, but it all ended up correct.

Also if you look online there are copious amounts of 8800 GTS 512 SLI vs GTX 260 and the general consensus is usually the GTX 260 for more $. That said I don't have a good way to test the performance difference.

Also he went from around 1200 to 1600 points in 3dmark06 with SLI.
 

The_Blood_Raven

Distinguished
Jan 2, 2008
2,567
0
20,790
True but a heavily overclocked 5850 and a 5870 can play Crysis at 1920x1200 on very high with some small AA, that's a hell of a lot better than he can. After thinking a bit, I'm starting to think something is wrong again. Does my listed performance for his setup really sound that low?
 
I agree - the pair of 8800GTS's aren't acting like they should.

My single 8800M GTX is comparable to somewhere in between an 8800GT and a 512 GTS, and it plays Crysis about the same as that setup, and Fallout 3 at a solid 50fps at 1920x1200. Two 8800GTS's should do far better.

Play with the setup for a while, see if you can get it straightened out. If not, I think the 5870 is the better option.
 
Some (random thought) things to consider.

Those 8800s dont have much RAM so you should get much better performance with AA/AF turned off. SLI would effectively divide the RAM by 2 making the problem worse.

Did you try the same games with a single 8800 to see if you get better or worse performance?

It doesnt look like that motherboard is capable of crossfire.

If the motherboard/CPU is limiting the 8800s performance, a 5850 or even a 5870 wont give you much benefit.

If buying a 5850 results in good performance, is it worth spending any more time messing around with the old 8800s? You at least get future capability for multiple monitors and DX11.

 

The RAM per GPU is half as much, which should make the AA/AF resource problem worse. Every x2 GPU made has twice the RAM of the single core GPU for a reason.

My single core 9800 GTX+ 512MB gets large improvements (like going from 30-35 FPS to 45-50 FPS) in many games when I turn the AA off. If I did SLI, it would make an even larger difference. Basically the lack of video RAM creates a severely diminishing return on GPU power. Going into the games custom graphics settings and further reducing features that are RAM intensive helps even more.
 

The_Blood_Raven

Distinguished
Jan 2, 2008
2,567
0
20,790
Yeah I'll have him drop it off tomorrow morning and I should have some opinions by the night. Also that 680i LT is a piece anyway and I'm not worried about him finding another, I have an X38 laying around here somewhere.

I'm thinking I need to test the slave card (2nd in SLI) by itself to make sure it is pulling it's own weight.

Thanks for the help guys.
 



2x9800GTXs with 512mb = 1GB, divide that by 2 and you get 512mb...the same as a single GPU

ATI x2 GPUs actually have the same amount of RAM as two regular GPUs in crossfire.

4850x2 with 1GB basically means each GPU has 512mb, combined equaling 1GB. 4850x2 1GB = same amount of VRAM as 2x4850s w/ 512mb each.

That's why there is a 2GB version of the 4850x2...that is the version with twice the RAM.
 


Yes, I have a pair of 8800GT's and they run flawlessly at 19 x 10, the XFX 680i LT was not one of the better 680i boards IMHO.
 

The_Blood_Raven

Distinguished
Jan 2, 2008
2,567
0
20,790
Also Monk is running a stock E8400 (I think I remember it being stock in an other thread) and he is doing better with 8800 GT's.

To be honest I compared it to 2 4850s and didn't see much of a difference but I guess I expected the 4850 to be a good bit faster and it isn't great in Crysis. Going by benchmarks I can see 2 4850s don't do all that hot in Crysis, but they do a hell of a lot better in other games.
 

HibyPrime

Distinguished
Nov 23, 2006
263
0
18,790
Is the SLi bridge plugged in securely? Even if it's not, most of the time the drivers still let you enable SLi. I know it's a noob thing, but you never know, it might be broken or something. Might even want to examine the traces on the connector.

You said it went from 1200 to 1600 in 3d mark? Do you mean 12000 to 16000? If not, then there is most definitely something seriously wrong lol...

You might also want to check the pcie clock rate in the bios, it's probably a set multiplier of the FSB, but it might be changable, who knows...




What CPU are you using? He's got a dual core, so if you have a higher end C2Q or PII then he has his answer.
 
When the 680i was in it's prime I spoke with a few vendors who refused to stock the XFX 680i boards due to the failure rate they had experienced, which makes me wonder it they are one of the partners that skimped on the amount of layers used in the boards construction.
 
SLI is working, right? My old Asus SLI board got this weird glitch where it would detect the second card only about 10% of the time, though it ran fine otherwise. Of course a quick call to Asus got that fixed, and while this probably isn't your problem, it does go to show that you can have board partial failures without entire computer failures.

Edit: And just to note, even though that board is 5+ years old it is going strong still!