[H] - GTX680 3-Way SLI vs. 7970 Tri-Fire review

[Silverforce11]

So would I but the problem I have is that it's more than 6 months since they released this card and AMD typically stop making improvements to the performance of a card once it's successor arrives on the scene so we have until this autumn probably before the 7900 is 'retired' and we likely see no further improvements. I hope I'm wrong but that appears to me to be the case (I've been with ATI/AMD since 4870 replaced my 8800 gtx but the green team's driver support is pushing me back towards their camp despite myself). AMD need to spend a little more on driver engineers for their GPU's so stuff works off the bat not 6 months after launch.

I recently moved on from my 5850, and saw improvements all the way through. It's generally many generations ago, ie. 48xx that isn't worked on anymore. Same with NV, they dont up performance for gtx2xx series lately.

 

[cmdrdredd]

I recently moved on from my 5850, and saw improvements all the way through. It's generally many generations ago, ie. 48xx that isn't worked on anymore. Same with NV, they dont up performance for gtx2xx series lately.

No, but the GTX200 series did get adaptive vsync options and that's at least something. They didn't have to make it work on older GPUs like they have.

I think that after a while we just have to accept the fact that we need to replace older cards if we want to have the latest updates and support. Nature of the beast I suppose.

 

[MrMuppet]

One thing I want to know. Btw a oc 7970 and oc 680, assuming similar minimum fps and similar average fps and same settings, is a single 680 smoother or 7970 smoother or both about equal or what? Assuming the same config, same minimum fps, same average fps and same settings. Which would feel far more smoother? Or are they very close? Will input lag be affected? Will image quality vary as well?

Nvidia is smoother even in a single-card configuration this generation, according to the numbers. Relatively speaking a single HD 7970 is about as uneven or "unsmooth" as GTX 680 SLI. But either shouldn't really be a problem compared to HD 7970 CF.

See the graphs here for details:
http://www.sweclockers.com/recension/15381-geforce-gtx-690-varldens-snabbaste-grafikkort/18#pagehead

And this chart for the summary:

In case hot-linking doesn't work:
http://www.sweclockers.com/image/diagram/3038?k=32aca361ccc3d21b6f19fd3afbebe5fb

 

[cmdrdredd]

what is that chart showing? Can anyone translate it?

 

[SlowSpyder]

Looks like AMd really needs to sell more GPU's than they do:

http://www.theinquirer.net/inquirer/news/2169235/amd-reports-usd580m-loss-graphics-sales-fall

Did you have a signature in the past that mocked the credibility of sites like theinquirer, semiaccurate, etc.?

 

[MrMuppet]

what is that chart showing? Can anyone translate it?

Sorry, it's Swedish (SweClockers), so use Google translate on the page and you'll probably understand:
http://translate.google.com/transla...690-varldens-snabbaste-grafikkort/18#pagehead

Basically it's SweClockers' relative measure of micro stuttering. The title of the chart reads (literal translation):

'Battlefield 3, evenness of frame frequency
"Comrades", 1920x1080 Ultra, relative difference compared with local mean value'​

The x-axis label at the bottom reads:

'Lower is better (Percent)'​

Keywords:
jämnhet = evenness/regularity/uniformity/smoothness
bildfrekvens = frame frequency/frame rate
millisekunder = milliseconds (for the graphs)


The chart once more (for convenience):

Source: http://www.sweclockers.com/recension/15381-geforce-gtx-690-varldens-snabbaste-grafikkort/18#pagehead

 

[RussianSensation]

So would I but the problem I have is that it's more than 6 months since they released this card and AMD typically stop making improvements to the performance of a card once it's successor arrives on the scene so we have until this autumn probably before the 7900 is 'retired' and we likely see no further improvements. .

HD7900 is based on the new GCN architecture. VLIW-4/5 were used since AMD launched 2900XT. HD8000/9000 will probably still be based on GCN, just enhanced/improved version. The underlying fundamental architecture should still be GCN (but they might call it GCN 2.0 or something). So when the drivers will be optimized for HD8000 series, they should still impact 7900 series.

For some reason I thought you are a big ArMA player? If so, HD7970 GE is almost as fast as a GTX690 in those games!

If you don't play ArMA though, overall I'd pick SLI over CF as just about every review keeps confirmation that SLI feels smoother even when the framerates are lower.

Nvidia is smoother even in a single-card configuration this generation, according to the numbers. Relatively speaking a single HD 7970 is about as uneven or "unsmooth" as GTX 680 SLI. B

MrMuppet, are you really gaming on HD4000? -

That BF3 testing is a bit odd. How can HD6970 be "smoother" than HD7970?

I like SweClockers but I think I am going to trust TechReport's review on that one since it more coincides with GTX680 > HD7970 > HD6970 for BF3 (which is the more common theme from reviews):

In triple-monitor gaming though, HD7970 CF seems to be faster and smoother than a GTX690/GTX680 SLI. Strange. Who to believe? ?

 

[Silverforce11]

It's probably due to NV's frame time optimization not being effective (or less effective) at 3 monitor resolution??

 

[boxleitnerb]

Techsreport percentile method cannot catch frame metering. I personally find it unsuitable for testing AFR. You have to look at actual frametimes, at the raw data. And even then the frame metering is not accounted for because fraps cannot measure it. So you have to trust the reviewers and people that can actually compare the two setups in gameplay.

 

[Jacky60]

HD7900 is based on the new GCN architecture. VLIW-4/5 were used since AMD launched 2900XT. HD8000/9000 will probably still be based on GCN, just enhanced/improved version. The underlying fundamental architecture should still be GCN (but they might call it GCN 2.0 or something). So when the drivers will be optimized for HD8000 series, they should still impact 7900 series.

For some reason I thought you are a big ArMA player? If so, HD7970 GE is almost as fast as a GTX690 in those games!

If you don't play ArMA though, overall I'd pick SLI over CF as just about every review keeps confirmation that SLI feels smoother even when the framerates are lower.

You're right re:Arma so my choice would probably be 7970 trifire next but I have a cpu upgrade to do first.

 

[guskline]

Just installed 2 EVGA GTX 670 FTWs in SLI in rig 1 below. Very fast and very smooth on a 3 monitor 5760 x 1080 setup.

 

[MrMuppet]

MrMuppet, are you really gaming on HD4000? -

That BF3 testing is a bit odd. How can HD6970 be "smoother" than HD7970?

No, well, I have a GTX 670 and a HD 7970 Dual-X (Sapphire). I just haven't fully decided yet on which to return. To make matters worse, I won't be able to return the 7970 if i unbox it, so there's your answer. I'll probably keep the GTX 670 (not too keen on upgrading to StutterFire in the future, but perhaps GTX 670 SLI?).

Why wouldn't HD 6970 be able to be smoother than a HD 7970, relatively speaking? They're different architectures. In addition to that, since the difference was so small, you need to take possible measurement error into account (especially when one GPU is so much faster and more easily bottlenecked by other things), so that could explain it in part.

Those charts you linked to were of absolute numbers and not relative ones, so that's an apples and oranges comparison (since 6970 is a much slower card).

SweClockers have published their raw data on the page I linked to, if you want to analyze it yourself.

 

[guskline]

MrMuppet: Just installed 2 GTX 670 FTWs in SLI in my first rig below. With your hardware, they will perform incrdibly well. Got 8929 in 3Dmark11 with performance setting with 1 card and 14219 for 2 in SLI. That's with no OCing of the GPUs

 

[jacktesterson]

MrMuppet: Just installed 2 GTX 670 FTWs in SLI in my first rig below. With your hardware, they will perform incrdibly well. Got 8929 in 3Dmark11 with performance setting with 1 card and 14219 for 2 in SLI. That's with no OCing of the GPUs

What is your "graphics" score. Thats really whats important with 3DMark11. Example - Having HT can add 2000-2500 more points on your total score, yet in real world gaming, HT does very little. CPU's can fluctuate the score a lot. I do know you have a 2500k though, so very nice score!

Just for comparison, here are my 2 7970's with a mild overclock on stock voltage:

Graphics: 17736

http://3dmark.com/3dm11/3925027

 

[cmdrdredd]

My system scores 15531 with a graphics score of 20424

 

[MrMuppet]

MrMuppet: Just installed 2 GTX 670 FTWs in SLI in my first rig below. With your hardware, they will perform incrdibly well. Got 8929 in 3Dmark11 with performance setting with 1 card and 14219 for 2 in SLI. That's with no OCing of the GPUs

Well, I'm much more interested in actual game performance than synthetic benchmarks tbh (haven't been interested in those for a long time, other than to make sure nothing's wrong with my system).

How much has your gaming experience improved? How well does it scale? How's the micro stuttering? Etc.

You game at 1200p I presume? (Judging from your HZ281.)

edit: Oh, and I fear my system will overclock pretty poorly (4.1 Ghz is unstable at stock vcore and maxed at slighly above 80c before failing IBT). The ambient/air intake temp is over 30c even at idle in the closet (31c ambient atm). Still I fear I may have applied too little TIM (too small a pea) and perhaps I should consider a remount (sigh)? It's "idling" (5-10% CPU usage for reasons) at 40c atm. (I use RealTemp.)

 

[RussianSensation]

edit: Oh, and I fear my system will overclock pretty poorly (4.1 Ghz is unstable at stock vcore and maxed at slighly above 80c before failing IBT). The ambient/air intake temp is over 30c even at idle in the closet (31c ambient atm). Still I fear I may have applied too little TIM (too small a pea) and perhaps I should consider a remount (sigh)? It's "idling" (5-10% CPU usage for reasons) at 40c atm. (I use RealTemp.)

Even with 31*C ambient those temperatures don't look right for an NH-D14. It's one of the best air coolers, and you are running your 3570k at stock voltage. You should be getting mid 50s to low 60s at 4.1 Ghz I would imagine.

Here is a thread with a couple videos to help you out.
http://www.overclock.net/t/1229728/cooling-the-fx-8120/10

 

[MrMuppet]

Will check it out, thanks. Everyone's saying how not to use too much TIM all the time, so I used a pretty conservative amount of NT-H1 and let the NH-D14 do the spreading (supposedly manual spreading can introduce air pockets).

If attempting a remount, should I clean the cpu and HSF base first, even if I only want to apply some more of the same TIM compound? I have some ArctiClean and iirc correctly coffee filters work well (I don't have any lint-free cloth)? Too much may be bad, but should at least be a safer bet than too little, no? (Thicker layer of TIM > air layer.)

I also have some tubes of AS5 and Arctic Céramique remaining from earlier builds, if either of those are easier to get right than the NT-H1? (But AS5 is said to have to cure.)

 

[MrMuppet]

If attempting a remount, should I clean the cpu and HSF base first, even if I only want to apply some more of the same TIM compound?

Thinking some myself, I realized not cleaning may be equivalent to spreading manually and introduce air pockets. So one probably should clean, I hope ArctiClean is fine since that's what it's made for.

snip

Forgot to quote. In the How Thermal Compound Spreads video it seems a lot of TIM was used, much more than I applied and had imagined. Blob, line or cross hmm.

If Ivy Bridge basically is a line, the line method is probably best. Should the line be vertical or horizontal? I think I'll have to look at the backside ("pinside") while I'm at it (can't remember).

 

[cmdrdredd]

When I installed my D14 I put a dot right in the middle, about the size of a pea or kernel of corn. Mashed the HSF down with the screws and booted up.

 

[MrMuppet]

When I installed my D14 I put a dot right in the middle, about the size of a pea or kernel of corn. Mashed the HSF down with the screws and booted up.

That pea again! It's a bit akin to the length of a rope. I'm pretty sure I didn't use a full-sized one though. I'll probably have to check, clean+reapply TIM and reseat it one of these days.

Did you wriggle and twist the D14 to spread it better? I only really put it in place and fastened the screws.

 

[cmdrdredd]

That pea again! It's a bit akin to the length of a rope. I'm pretty sure I didn't use a full-sized one though. I'll probably have to check, clean+reapply TIM and reseat it one of these days.

Did you wriggle and twist the D14 to spread it better? I only really put it in place and fastened the screws.

no, wiggle and twist can accidently lift it up and create air bubbles, just press it down and screw it in.