GeForce GTX 680 3-Way SLI & Radeon 7970 Tri-Fire Review @ [H]

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,634
GeForce GTX 680 3-Way SLI & Radeon 7970 Tri-Fire Review - What do you get when you install three GeForce GTX 680 cards for 3-Way SLI and then three Radeon HD 7970 cards for Tri-Fire? You get insanely fast gaming performance and a gameplay experience that begs to be compared delivered by both. We find out which multi-display configuration is better for gaming in Eyefinity and NV Surround.
 
Awesome review! These reviews need to be required reading for anyone browsing the AMD or nVidia GPU subforums.
 
Nice review, now I just need some more monitors...

Also on the BF3 Multiplayer page, the drop down menu is broken.
 
In the BF3 review, you mention
Turning off MSAA completely, and just using FXAA, GeForce GTX 680 3-Way SLI becomes playable. In comparison, Radeon HD 7970 Tri-Fire is technically faster in raw framerate. However, the game feels completely different between the two. The game actually feels better, and smoother with GTX 680 3-Way SLI.
Click to expand...
How can this be when the 7970 setup gets 63fps min compared to 52fps min for the 680 setup? Also the graph seems to show the 7970 TriFire has higher frame rates at all times.
 
slh28 said:
How can this be when the 7970 setup gets 63fps min compared to 52fps min for the 680 setup? Also the graph seems to show the 7970 TriFire has higher frame rates at all times.
Click to expand...

It's called microstuttering which is inherent to all AFR rendering solutions (which CF and SLI are). Nvidia is employing something they call "frame metering" to combat this, AMD is doing nothing obviously. In this regard, SLI has always been better than CF, just look at the last 10 articles here on [H]. In every single one the same observation.
 
slh28 said:
How can this be when the 7970 setup gets 63fps min compared to 52fps min for the 680 setup? Also the graph seems to show the 7970 TriFire has higher frame rates at all times.
Click to expand...

You have to consider not only the raw frame rate, but also the frame intervals. I'll take it to the extreme just to give you an idea. What would feel better to you?

1) 50fps where each frame is drawn on the monitor every 1/50th of a second?
2) 60fps where 1 frame takes 1/10th of a second to come up followed by 59 frames that come in the remaining 9/10th of a second.

nVidia does a better job at making sure frame intervals are nearly equal. AMD (I suppose?) just blasts them out as fast as they can, and sometimes a frame takes longer to compute and we perceive that as jerkiness in the video.
 
slh28 said:
In the BF3 review, you mention

How can this be when the 7970 setup gets 63fps min compared to 52fps min for the 680 setup? Also the graph seems to show the 7970 TriFire has higher frame rates at all times.
Click to expand...

FPS charts are the total # frames in a second. It does not show sub-second refresh rate. Put it this way...120 frames in a 0.5 seconds and then 0 frames for 0.5 seconds will still average 60 on any FPS chart. Our eyes may not be good an resolving differences in continuous motion (e.g. a pure 45 FPS vs a pure 60 FPS)...however, our eyes are very good at resolving "changes" in FPS at very high rates; none of the FPS tools detail this (which sucks).

Personally, I wish Kyle and crew could find a FPS tool which lists the aboslute time each frame is "posted". Then we could apply a bit of calculus and show that a very high FPS/s with ATI and a much lower FPS/s with nVidia.
 
You can measure frametimes with fraps, but unfortunately it measures at the wrong point in the rendering process, not capturing what Nvidia is doing with the frame metering - which is making all the difference here. So this measurement would be useless, sadly.
 
Trepidati0n said:
Personally, I wish Kyle and crew could find a FPS tool which lists the aboslute time each frame is "posted". Then we could apply a bit of calculus and show that a very high FPS/s with ATI and a much lower FPS/s with nVidia.
Click to expand...

Did a quick google and it seems typing in "microstuttering crossfire" comes up with loads of hits! I don't think I've seen any review sites do this kind of analysis, but from the sound of things it's pretty important.
 
Sorry to be critical, but someone should bring this up:

You use TrSS vs SSAA in apples to apples comaprisons. What? That is most certainly NOT apples to apples. Transparency AA is not full scene, TrSS is NOT full scene SSAA, if you want to use SSAA in these comparisons you have to force it in nvidia inspector. You compare TrSS vs SSAA in several comparisons: Thats kind of annoying because TrSS only covers transparent textures while SSAA covers everything. If you want SSAA use SGSSAA in nvidia inspector and make the comparison, you're comparing 2 different things completely. TrSS is nowhere NEAR as good as SSAA and you pass the comparisons off as if they're the same thing.

If you really want apples to apples you should do these tests by choosing "override application setting" in nvidia inspector, and choosing SGSSAA. That is the only means for an apples to apples comparison. TrSS is nowhere near as intensive or as good as full screen SSAA. The other issue is verifying that the override happens, nvidia inspector override a lot of times doesn't work, thus it should be verified to be working. But you can rest assured that SGSSAA will be a lot slower than TrSS and it will be a more valid comparison, as of right now it is not a valid comparison. This doesn't nullify your findings, nvidia is better for scaling beyond 2 gpu's for sure, but ultimately your ssaa vs trss comparison is flawed unless i'm really missing something.
 
Last edited:
My personal experience using NVIDIA and ATI multicard solutions is that SLI is smoother. When I had 5870 CFX, I would be getting 40-50fps and it would feel like 25-30fps. When I switched to GTX580s, I had no such problems.
 
3 cards? I've not seen even one for sell at Newegg, Tiger, or Frys. Always "out of stock." Paper launch comes to mind.
 
xoleras said:
If you really want apples to apples you should do these tests by choosing "override application setting" in nvidia inspector, and choosing SGSSAA. That is the only means for an apples to apples comparison. TrSS is nowhere near as intensive or as good as full screen SSAA. The other issue is verifying that the override happens, nvidia inspector override does not always work. This doesn't nullify your findings, nvidia is better for scaling beyond 2 gpu's for sure, but ultimately your ssaa vs trss comparison is flawed unless i'm really missing something.
Click to expand...

If the game has MSAA-support, no override is necessary. Just select SGSSAA from the transparency-AA menu and you're good to go. Override can provide better results but costs more performance. As it is unclear where the AA will be applied and a comparison between AMD and NV will be difficult at best, rather let the app decide.
 
deeppow said:
3 cards? I've not seen even one for sell at Newegg, Tiger, or Frys. Always "out of stock." Paper launch comes to mind.
Click to expand...

Yep

I have a nice stack of paper in my system right now. My paper stack runs Skyrim/BF3 amazingly well. :rolleyes:

There is a difference between a "paper launch" where a product is announced and none is shipped out to vendors, and a successful launch where demand outpaced supply. I bought my 680 GTX day 1 when there were plenty at the egg.
 
again thanks for the awesome reviews kyle

Fixed, thanks. - Kyle
 
Last edited by a moderator:
Please please stay on AMD about their crossfire drivers. Someone has to speak for us because they are obviously not getting the message. Its so stupid that my cards cant even idle at the right clock speeds with the newest drivers.

Could the Catalyst Crew Feedback Form become a sticky in the AMD sections?

http://www.amdsurveys.com/se.ashx?s=5A1E27D27E29B0E3
 
boxleitnerb said:
If the game has MSAA-support, no override is necessary. Just select SGSSAA from the transparency-AA menu and you're good to go. Override can provide better results but costs more performance. As it is unclear where the AA will be applied and a comparison between AMD and NV will be difficult at best, rather let the app decide.
Click to expand...

I'm sure nvidia is better at 3 gpu's but I very very highly doubt the comparison is apples to apples. This isn't the first review this has happened: There should be an understanding that Transparency SS is not SSAA - apparently, a lot of people think that TrSS is full screen SSAA. As far as override, if you say so, pretty sure they selected override for SSAA on the AMD card as well. That is the best comparison. Override SGSSAA via nvidia inspector compared to override SSAA with AMD. TrSS vs SSAA isn't a valid comparison.
 
Last edited:
deeppow said:
3 cards? I've not seen even one for sell at Newegg, Tiger, or Frys. Always "out of stock." Paper launch comes to mind.
Click to expand...
It's not a paper launch, I have two :p Supply and demand. There is a lot of demand and unfortunately not a lot of supply at this time. That's more on the fabs than NVIDIA though, I'm sure they'd love to be able to keep the shelves fully stocked.
 
Very nice review.

BUT...........I think the overall attitude and result is somewhat biased by the fact that you are still moaning over AMDs driver problems. Crossfire is as good or even better in some of your results, but quickly downplayed by the "smooth-ness factor" whatever that is.

Let's be honest, it's a problem for about two dozen people......I mean really, all those running triple GPU solutions, please raise your hand.....So should AMD make this an emergency? or should they just take their time and fix it right?

That and the fact that you can't even buy one GTX 680 anywhere, let alone three.

Face it, there are only a hand full of games at present that require two GPUs let alone three.......yeah, I've had three way SLI and CrossfireX in the not so distant past.....and honestly I couldn't find a game that was worthy.

I'm not a big fan of either company, they both have their problems. Sure I'd like to see better AMD driver support, but to down-play their success by consistantly bashing their software support doesn't do anybody any good......who has cards for sale that you can actually buy?;)
 
One thing I do appreciate about [H] GPU reviews is that they go beyond pure FPS measurements. I've been running 3 way 680 SLI for 3 weeks now and I've been very pleased with the setup. It's cooler, quieter, faster and more stable, a complete quadfecta over my 3 580s.

It's also hard to see the issues with AMD drivers in setups like these. I know there will be AMD supporters that will say they are perfectly happy with AMD's drivers but I simply have little confidence in AMD's drivers beyond a single card. I've not run Crossfire personally but there's simply too many complaining about AMD drivers in these exotic setups.
 
xoleras said:
I'm sure nvidia is better at 3 gpu's but I very very highly doubt the comparison is apples to apples. This isn't the first review this has happened: There should be an understanding that Transparency SS is not SSAA - apparently, a lot of people think that TrSS is full screen SSAA. As far as override, if you say so, pretty sure they selected override for SSAA on the AMD card as well. That is the best comparison. Override SGSSAA via nvidia inspector compared to override SSAA with AMD. TrSS vs SSAA isn't a valid comparison. Either do a valid comparison or omit the result, period.
Click to expand...

That is true. What is "adaptive SSAA" anyway? Is it the SSAA-part of AAA, so AMDs transparency AA? Or is it really full screen SGSSAA?

As I said - override is not necessary. The game decides where AA is applied and thinks the driver applies MSAA. The MSAA samples are then just substituted for SSAA without the game knowing anything about it. It works in BF3, Skyrim, Batman AC (sadly with red graphic glitches in DX11) and every other game that provides MSAA. With override, it could be that Nvidia smoothes better than AMD and that would be an unfair comparison. For example, the books in Skyrim in the inventory are only smoothed when override is selected, as are the mini lockpicking games. Otherwise, no AA exists there. However, override can cause MASSIVE VRAM requirement. I measured +50% in Skyrim from 2GB to 3GB. And a fps hit of 10-20% in addition to that.
 
boxleitnerb said:
That is true. What is "adaptive SSAA" anyway? Is it the SSAA-part of AAA, so AMDs transparency AA? Or is it really full screen SGSSAA?

As I said - override is not necessary. The game decides where AA is applied and thinks the driver applies MSAA. The MSAA samples are then just substituted for SSAA without the game knowing anything about it. It works in BF3, Skyrim, Batman AC (sadly with red graphic glitches in DX11) and every other game that provides MSAA. With override, it could be that Nvidia smoothes better than AMD and that would be an unfair comparison. For example, the books in Skyrim in the inventory are only smoothed when override is selected, as are the mini lockpicking games. Otherwise, no AA exists there. However, override can cause MASSIVE VRAM requirement. I measured +50% in Skyrim from 2GB to 3GB. And a fps hit of 10-20% in addition to that.
Click to expand...

What nvidia card are you using? First of all, I don't see why you're complaining about override because thats exactly what was used on the AMD card -- override. Secondly, i'm not at home to test but the nvidia control panel had a specific drop down box for transparencies and it does not substitute SSAA for MSAA. I really have no idea what you're talking about there, but when i'm at my PC i'll check it again. The only way to get true full screen SSAA was by nvidia inspector. AFAIK that has never changed, but I could be wrong.

The fair comparison is override SGSSAA via inspector. Either make the comparison EQUAL or leave it out of the review, that is my opinion. As the review stands now -- unless i'm really misunderstanding what happened, and I apologize in advance if thats the case - they might as well compare MSAA on 1 card to SSAA on the other. TrSS vs SSAA doesn't make sense because they're not remotely the same thing. Can anyone clarify on what the situation really is here?
 
magoo said:
Very nice review.

BUT...........I think the overall attitude and result is somewhat biased by the fact that you are still moaning over AMDs driver problems. Crossfire is as good or even better in some of your results, but quickly downplayed by the "smooth-ness factor" whatever that is.
Click to expand...
It's called microstuttering, and it's a real problem. Google it. It's been a problem for both NVIDIA and ATI, but NVIDIA seemingly has been able to work around it better in drivers than AMD has.

Let's be honest, it's a problem for about two dozen people......I mean really, all those running triple GPU solutions, please raise your hand.....So should AMD make this an emergency? or should they just take their time and fix it right?
Click to expand...
So you don't think this is a salient point in a comparison review for 3x GPU setups? Maybe it's a low priority issue overall, but it's sure as hell a big issue if you are shopping for 3 GPUs right now, and trying to decide between NVIDIA and AMD.

Face it, there are only a hand full of games at present that require two GPUs let alone three.......yeah, I've had three way SLI and CrossfireX in the not so distant past.....and honestly I couldn't find a game that was worthy.
Click to expand...
Triple monitor resolutions would like to have a word with you.

I'm not a big fan of either company, they both have their problems. Sure I'd like to see better AMD driver support, but to down-play their success by consistantly bashing their software support doesn't do anybody any good......who has cards for sale that you can actually buy?;)
Click to expand...
When you are looking to spend $1500 on a GPU setup, I would wager that a lot of those people want it to work spectacularly out of the box. It's not impossible to find GTX680s you just need to shop around.

I would not go with an AMD multi-GPU solution right now.. the driver support for the 7xxx series has been garbage when it comes to Eyefinity/CFX in comparison to the 6xxx series.
 
Interesting review.

I was planning to go back to 3 screen (or maybe up to 5 with 6GB 7970s) gaming with CF 7970s, but my main games currently don't need me to upgrade & 6GB versions are not out yet, so I've been holding off.

Sad to hear about those AMD driver issues...
 
Your Skyrim SSAA results are invalid. Sparse Grid SSAA is far more taxing than Transparency AA. You need to use the official Nvidia SSAA tool (or inspector) to enable SGSSAA for a head to head comparison. Or alternatively, bench Transparency MSAA vs Adaptive MSAA.

Apart from that, good review :)
 
xoleras said:
What nvidia card are you using? First of all, I don't see why you're complaining about override because thats exactly what was used on the AMD card -- override. Secondly, i'm not at home to test but the nvidia control panel had a specific drop down box for transparencies and it does not substitute SSAA for MSAA. I really have no idea what you're talking about there, but when i'm at my PC i'll check it again. The only way to get true full screen SSAA was by nvidia inspector. AFAIK that has never changed, but I could be wrong.

The fair comparison is override SGSSAA via inspector. Either make the comparison EQUAL or leave it out of the review, that is my opinion. As the review stands now -- unless i'm really misunderstanding what happened, and I apologize in advance if thats the case - they might as well compare MSAA on 1 card to SSAA on the other. TrSS vs SSAA doesn't make sense because they're not remotely the same thing.
Click to expand...

GTX580 3GB SLI.
Override is not always necessary and doesn't work in DX10/11 anyway.

Okay I'll explain it again:
The application itself has no idea what kind of AA is used. It only knows where and how many samples. If it is SSAA, MSAA, CSAA - it doesn't care.

If you select SGSSAA from the transparency dropdown box in the Inspector, the MSAA that is selected ingame is replaced by SGSSAA. This is what I meant. No other adjustments (except LOD) whatsoever are required. You need the Inspector, though, to select SGSSAA. That is correct (except OpenGL where the bug was never fixed).

Afaik, with Nvidia there are 3 settings:
1. app control
2. enhance
3. override

How does that look on the AMD side?
With Nvidia, override is only necessary to force MSAA in the first place. MSAA is the basis for SGSSAA as both use the same sample positions. Therefore, if MSAA is supported by the game and is already applied, no override is necessary for SGSSAA.

I agree with you that Kyle and Brent should be more careful here and really verify that SGSSAA is working in full screen mode. This is easily seen on the models shown during the loading screens. SGSSAA = less shader aliasing = less shimmering.
 
"This issue is this, you cannot run with any driver version greater than the 8.921.2 RC11 driver that was initially released back in January of 2012 if you want to run games with Tri-Fire + Eyefinity enabled"

Anybody read that line at the very beginning and NOT see NVIDIA coming out on top?
 
slh28 said:
In the BF3 review, you mention

How can this be when the 7970 setup gets 63fps min compared to 52fps min for the 680 setup? Also the graph seems to show the 7970 TriFire has higher frame rates at all times.
Click to expand...

What it truly is, is "frame time." How long did you see that from compared to the flow of others.
 
Hexen525 said:
"This issue is this, you cannot run with any driver version greater than the 8.921.2 RC11 driver that was initially released back in January of 2012 if you want to run games with Tri-Fire + Eyefinity enabled"

Anybody read that line at the very beginning and NOT see NVIDIA coming out on top?
Click to expand...

Yes, its hardly surprising. RC11 had a lot of crossfire issues even for dual GPU. Its AMDs fault however -- why doesn't eyefinity surround work with 12.3?

Whatever.
 
Trepidati0n said:
Personally, I wish Kyle and crew could find a FPS tool which lists the aboslute time each frame is "posted". Then we could apply a bit of calculus and show that a very high FPS/s with ATI and a much lower FPS/s with nVidia.
Click to expand...

NVIDIA has been working on this and we hope to have this very soon. Requires some fairly expensive monitoring tools, but we have committed to NVIDIA that we will purchase the hardware necessary to do this.
 
magoo said:
Very nice review.

BUT...........I think the overall attitude and result is somewhat biased by the fact that you are still moaning over AMDs driver problems. Crossfire is as good or even better in some of your results, but quickly downplayed by the "smooth-ness factor" whatever that is.
Click to expand...

I've never run a Crossfire setup but I wouldn't discount the "smooth-ness factor". I've run enough SLI setups to understand exactly the idea behind smoothness and even between 3 580s and 3 680s there is a noticeable difference I'd say in the smoothness department. Most of what I'm seeing is probably just a matter of better performance for the 680s but there is a difference in the way games feel even when both the 580s and 680s have high frame rates.

magoo said:
Let's be honest, it's a problem for about two dozen people......I mean really, all those running triple GPU solutions, please raise your hand.....So should AMD make this an emergency? or should they just take their time and fix it right?
Click to expand...

While I do agree with your point in general people who buy and build multi-GPU setups over a lifetime are probably going to spend many tens of thousands of dollars on GPUs. We're talking about a very loyal and repeat base that will spend a lot of money over a lifetime. So while perhaps not a top priority certainly one that neither nVidia or AMD can ignore and not satisfy if both companies are going.to build $500 GPUs and develop multi-monitor and multi-GPU technology.

magoo said:
That and the fact that you can't even buy one GTX 680 anywhere, let alone three.
Click to expand...

True they are in short supply but I got mine in about 8 days. Pad a little bit over MSRP but I've paid more in the past for the latest and greatest parts at launch. Nothing really new but nVidia certainly needs to get Kepler going faster.

magoo said:
Face it, there are only a hand full of games at present that require two GPUs let alone three.......yeah, I've had three way SLI and CrossfireX in the not so distant past.....and honestly I couldn't find a game that was worthy.
Click to expand...

It's pretty easy to suck up the power of even 3 680s running 3D Surround.

magoo said:
I'm not a big fan of either company, they both have their problems. Sure I'd like to see better AMD driver support, but to down-play their success by consistantly bashing their software support doesn't do anybody any good......who has cards for sale that you can actually buy?;)
Click to expand...

As you just said, we're talking about very exotic setups here that few people use. I don't see this review bashing AMD drivers in general but in very specific and rare cases. The criticisms in this article simply don't apply to 99% of PC gamers.
 
boxleitnerb said:
GTX580 3GB SLI.
Override is not always necessary and doesn't work in DX10/11 anyway.

Okay I'll explain it again:
The application itself has no idea what kind of AA is used. It only knows where and how many samples. If it is SSAA, MSAA, CSAA - it doesn't care.

If you select SGSSAA from the transparency dropdown box in the Inspector, the MSAA that is selected ingame is replaced by SGSSAA. This is what I meant. No other adjustments (except LOD) whatsoever are required. You need the Inspector, though, to select SGSSAA. That is correct (except OpenGL where the bug was never fixed).

Afaik, with Nvidia there are 3 settings:
1. app control
2. enhance
3. override

How does that look on the AMD side?
With Nvidia, override is only necessary to force MSAA in the first place. MSAA is the basis for SGSSAA as both use the same sample positions. Therefore, if MSAA is supported by the game and is already applied, no override is necessary for SGSSAA.

I agree with you that Kyle and Brent should be more careful here and really verify that SGSSAA is working in full screen mode. This is easily seen on the models shown during the loading screens. SGSSAA = less shader aliasing = less shimmering.
Click to expand...

Why would you select it from the transparency drop down? Am I misunderstanding that anything in the transparency drop down only affects transparency textures?

If you want SGSSAA you should select SGSSAA from the AA mode drop down in nvidia inspector. Not transparency. Or is that incorrect? Honest question. Anyway, i'll look into this later today.
 
xoleras said:
Why would you select it from the transparency drop down? Am I misunderstanding that anything in the transparency drop down only affects transparency textures?

If you want SGSSAA you should select SGSSAA from the AA mode drop down in nvidia inspector. Not transparency. Or is that incorrect? Honest question. Anyway, i'll look into this later today.
Click to expand...

SGSSAA was born due to a bug. When applying transparency SSAA, instead you got SGSSAA. This bug was fixed but because the customers wanted to keep SGSSAA, it was added as a separate option - still in the transparency AA menu, because that is where it originally came from:
 
You must log in or register to reply here.
Back
Top