Guru3D 970 SLI review

[Madpacket]

I wonder if this has to do with using a bridge rather than the pcie bandwidth... also curious that launch reviews ignored SLI testing. Hmmm.

 

[3DVagabond]

There's a few 3-4 way SLI reviews which show horrible scaling beyond 2 cards for Maxwell.

So it seems to be immature drivers for this architecture that is causing some multi-gpu issues.

The scaling is fine. It's just the dropped frames and performance drop off at 4K, which is an ongoing issue for nVidia (4K performance) not a Maxwell/new arch problem.

 

[Black Octagon]

Early drivers, fully expect this to be fixed.

Not sure about sli being tweaked to the point where frame delivery is BETTER than single-gpu, though.

 

[ams23]

The scaling is fine. It's just the dropped frames and performance drop off at 4K, which is an ongoing issue for nVidia (4K performance) not a Maxwell/new arch problem.

Well you aren't making sense here. 4K SLI performance and 4K SLI performance scaling is quite competitive overall with Maxwell if you look at a variety of different games.

From Guru3D, comparing GTX 970 SLI and 290X Crossfire performance improvement at 4K resolution vs. a single GPU and vs. each other:

Hitman Absolution
GTX 970 SLI vs single GPU: +84.6%
290X Crossfire vs single GPU: +76.5%
Multi-GPU 4K Performance Advantage: +25.0% (in favor of 290X Crossfire)

Bioshock Infinite
GTX 970 SLI vs single GPU: +68.6%
290X Crossfire vs single GPU: +66.7%
Multi-GPU 4K Performance Advantage: +18.0% (in favor of GTX 970 SLI)

Medal of Honor Warfight
GTX 970 SLI vs single GPU: +73.1%
290X Crossfire vs single GPU: +75.9%
Multi-GPU 4K Performance Advantage: +13.3% (in favor of 290X Crossfire)

Tomb Raider
GTX 970 SLI vs single GPU: +73.7%
290X Crossfire vs single GPU: +72.5%
Multi-GPU 4K Performance Advantage: +4.5% (in favor of 290X Crossfire)

Metro Last Light
GTX 970 SLI vs single GPU: +48.1%
290X Crossfire vs single GPU: +42.3%
Multi-GPU 4K Performance Advantage: +8.1% (in favor of GTX 970 SLI)

Crysis 3
GTX 970 SLI vs single GPU: +55.0%
290X Crossfire vs single GPU: +81.8%
Multi-GPU 4K Performance Advantage: +29.0% (in favor of 290X Crossfire)

Thief
GTX 970 SLI vs single GPU: +66.7%
290X Crossfire vs single GPU: +68.0%
Multi-GPU 4K Performance Advantage: +5.0% (in favor of 290X Crossfire)

So out of a modest sample size of seven games, GTX 970 SLI leads outright in two games vs. 290X Crossfire while being within 5% of the performance of 290X Crossfire in two other games.

Considering the great price and the frugal power consumption of GTX 970 SLI in comparison to 290X Crossfire, this is a very good result for GTX 970 SLI.

 

[3DVagabond]

Well you aren't making sense here. 4K SLI performance and 4K SLI performance scaling is quite competitive overall with Maxwell if you look at a variety of different games.

From Guru3D, comparing GTX 970 SLI and 290X Crossfire performance improvement at 4K resolution vs. a single GPU and vs. each other:

Hitman Absolution
GTX 970 SLI vs single GPU: +84.6%
290X Crossfire vs single GPU: +76.5%
Multi-GPU 4K Performance Advantage: +25.0% (in favor of 290X Crossfire)

Bioshock Infinite
GTX 970 SLI vs single GPU: +68.6%
290X Crossfire vs single GPU: +66.7%
Multi-GPU 4K Performance Advantage: +18.0% (in favor of GTX 970 SLI)

Medal of Honor Warfight
GTX 970 SLI vs single GPU: +73.1%
290X Crossfire vs single GPU: +75.9%
Multi-GPU 4K Performance Advantage: +13.3% (in favor of 290X Crossfire)

Tomb Raider
GTX 970 SLI vs single GPU: +73.7%
290X Crossfire vs single GPU: +72.5%
Multi-GPU 4K Performance Advantage: +4.5% (in favor of 290X Crossfire)

Metro Last Light
GTX 970 SLI vs single GPU: +48.1%
290X Crossfire vs single GPU: +42.3%
Multi-GPU 4K Performance Advantage: +8.1% (in favor of GTX 970 SLI)

Crysis 3
GTX 970 SLI vs single GPU: +55.0%
290X Crossfire vs single GPU: +81.8%
Multi-GPU 4K Performance Advantage: +29.0% (in favor of 290X Crossfire)

Thief
GTX 970 SLI vs single GPU: +66.7%
290X Crossfire vs single GPU: +68.0%
Multi-GPU 4K Performance Advantage: +5.0% (in favor of 290X Crossfire)

So out of a modest sample size of seven games, GTX 970 SLI leads outright in two games vs. 290X Crossfire while being within 5% of the performance of 290X Crossfire in two other games.

Considering the great price and the frugal power consumption of GTX 970 SLI in comparison to 290X Crossfire, this is a very good result for GTX 970 SLI.

You might try reading my post more clearly. I never complained about SLI scaling.

 

[ams23]

You might try reading my post more clearly. I never complained about SLI scaling.

At the end of the day, looking across a variety of different games, GTX 970 SLI is very competitive with 290X Crossfire with respect to both 4K Multi-GPU performance and 4K Multi-GPU performance scaling, while consuming MUCH less power and at a lower price too.

 

[3DVagabond]

At the end of the day, looking across a variety of different games, GTX 970 SLI is very competitive with 290X Crossfire with respect to both 4K Multi-GPU performance and 4K Multi-GPU performance scaling, while consuming MUCH less power and at a lower price too.

You still didn't read my post. You look at lower res and it's actually faster than 290X crossfire. @4K performance drops off. And in spite of what people want to admit to, those dropped frames are a problem. When AMD suffered from them it made buying Crossfire out of the question compared to SLI. When nVidia exhibits the same issue though it's, nothing to see here, move along." Maxwell SLI is as broken now as Crossfire was.

If you want to compare perf/$ we need to compare the 290 to the 970 and the 290X to the 980. Both the 290 and the 970 are better values of the four. And yes, we know Maxwell uses less power... for about the 100th time.

 

[ams23]

You still didn't read my post. You look at lower res and it's actually faster than 290X crossfire. @4K performance drops off.

Think logically about what you are saying. GTX 970 SLI has higher performance than 290X Crossfire at lower-than-4K resolutions (including 25x14 resolution), and is quite performance competitive with 290X Crossfire at 4K resolution, all at MUCH lower power consumption and at a lower price. So GTX 970 SLI is simply a better buy for most people no matter how you try to spin it.

As for FCAT, what it shows is that GTX 970 SLI has lower latency over time than 290X Crossfire at the 25x14 resolution tested by Guru3D, and as a result should have better and smoother performance in comparison at this resolution. When Guru3D measured Tomb Raider using FCAT at 25x14 resolution, there were numerous frametime latency spikes as high as ~10-15ms on 290X Crossfire, while GTX 970 SLI only sees frametime latency spikes as high as ~ 8-10ms. And in Bioshock Infinite, 290X Crossfire has some frametime latency spikes as high as ~ 50-60ms, while GTX 970 SLI has frametime latency spikes as high as only ~ 25-30ms.

 

[Keysplayr]

Sounds like even with dropped frames, performance on 970 SLI is still smoother than 290X crossfire. At least according to Guru3D and what ams23 is showing.

 

[Black Octagon]

ams23...you forgot to mention Maxwell's power efficiency

 

[Grooveriding]

ams23...you forgot to mention Maxwell's power efficiency

It does look as if the Maxwell launch drivers are a wreck. Dropped frames in SLI, DSR does not work in SLI, MFAA is not available at all etc. I'm surprised that they feel so unpolished as there was no obvious reason to rush these cards out leading to unfinished drivers at launch.

 

[Silverforce11]

Sounds like even with dropped frames, performance on 970 SLI is still smoother than 290X crossfire. At least according to Guru3D and what ams23 is showing.

Those charts don't show smoothness, just raw recorded FPS which NV themselves have said in the past aren't representative of perceived smoothness due to frame latency and runt frames.

You should know that since it was the NV Mantra for a long time already.

 

[Keysplayr]

Those charts don't show smoothness, just raw recorded FPS which NV themselves have said in the past aren't representative of perceived smoothness due to frame latency and runt frames.

You should know that since it was the NV Mantra for a long time already.

Gotcha. They don't show choppiness either. Raw recorded FPS is something Fraps does. Not FCAT. Lower latency means smoother gameplay. Or does it not mean that in this case? Lower latency means rougher gameplay?

 

[3DVagabond]

Gotcha. They don't show choppiness either. Raw recorded FPS is something Fraps does. Not FCAT. Lower latency means smoother gameplay. Or does it not mean that in this case? Lower latency means rougher gameplay?

Dropped frames. We need to subtract those reported frames from the actual viewable ones. I believe when AMD suffered from it it was referred to as cheating, and AMD not being honest and not caring about their customers. Crossfire was terribly broken. How could they have not possibly known and must have tried to simply ignore it? I know you remember.

 

[Keysplayr]

Dropped frames. We need to subtract those reported frames from the actual viewable ones. I believe when AMD suffered from it it was referred to as cheating, and AMD not being honest and not caring about their customers. Crossfire was terribly broken. How could they have not possibly known and must have tried to simply ignore it? I know you remember.

LOL we are talking about release drivers for a new architecture. AMD had their issues forever, long after any architecture was released. (Refer to Scott Wasson's first reports and look how long the architectures he tested were out.) I think you need to give Nvidia a little more time than two weeks to sort out any driver discrepancies. Unless you feel that is out of the question and deplorable and wish to go on slandering them and how they don't care about their customers.

 

[3DVagabond]

LOL we are talking about release drivers for a new architecture. AMD had their issues forever, long after any architecture was released. (Refer to Scott Wasson's first reports and look how long the architectures he tested were out.) I think you need to give Nvidia a little more time than two weeks to sort out any driver discrepancies. Unless you feel that is out of the question and deplorable and wish to go on slandering them and how they don't care about their customers.

Slandering? What have I said that's untrue? You sir are the one who is slandering me by falsely accusing me. I'm simply pointing out facts and backing them with review charts. History isn't what we're discussing. It's the here and now.

 

[Keysplayr]

Slandering? What have I said that's untrue? You sir are the one who is slandering me by falsely accusing me. I'm simply pointing out facts and backing them with review charts. History isn't what we're discussing. It's the here and now.

You never actually say anything 3D. You allude to, and then say "I never said".
And actually, yes indeed history is what you have brought up. Haven't you.
You asked me if I remembered something. Was that the here and now? Or was that history? Or, did you never say such things?
Your attempt to tailor the conversation to your ends has been noted, and expected.

 

[3DVagabond]

You never actually say anything 3D. You allude to, and then say "I never said".
And actually, yes indeed history is what you have brought up. Haven't you.
You asked me if I remembered something. Was that the here and now? Or was that history? Or, did you never say such things?
Your attempt to tailor the conversation to your ends has been noted, and expected.

I'm sure you'll be right there to tell us when nVidia fixes SLI with Maxwell. Until then...

 

[Skurge]

LOL we are talking about release drivers for a new architecture. AMD had their issues forever, long after any architecture was released. (Refer to Scott Wasson's first reports and look how long the architectures he tested were out.) I think you need to give Nvidia a little more time than two weeks to sort out any driver discrepancies. Unless you feel that is out of the question and deplorable and wish to go on slandering them and how they don't care about their customers.

You would think SLI wouldn't have these issues anymore. There was all the talk about hardware frame metering. I guess that was not true if they need drivers to fix it.

This late in the game, new arch or not, you can't be having issue like these and we need to pressure nVidia over this, we can't just say "give them a break". We need to be on them. You as a focus group member should not be telling us to give them time, you should be relaying our frustrations to them.

After all the bad press AMD got, Crossfire on Hawaii is the best implementation of mulit-GPU tech around. That happened cause we raised out voices and showed our displeasure. Nvidia should be held to the same standard.

Do you not agree?

 

[Keysplayr]

I'm sure you'll be right there to tell us when nVidia fixes SLI with Maxwell. Until then...

When it's improved, I, or someone else will let you know.

 

[Keysplayr]

You would think SLI wouldn't have these issues anymore. There was all the talk about hardware frame metering. I guess that was not true if they need drivers to fix it.

This late in the game, new arch or not, you can't be having issue like these and we need to pressure nVidia over this, we can't just say "give them a break". We need to be on them. You as a focus group member should not be telling us to give them time, you should be relaying our frustrations to them.

After all the bad press AMD got, Crossfire on Hawaii is the best implementation of mulit-GPU tech around. That happened cause we raised out voices and showed our displeasure. Nvidia should be held to the same standard.

Do you not agree?

All hardware works in conjunction with some sort of software, or firmware. What is frame metering hardware without any software? Nothing but silicon and transistors.
Your frustrations are pre-mature. GM204 has been out about two weeks.
(N/M I thought you had 970 SLI for some odd reason.)
AMD got bad press with an already long in the tooth arch when TR unleashed it's frame latency/pacing review methods article. Difference is, Nvidia will improve this on their own. You'd have to hammer AMD for months and maybe years for that to happen.

 

[Leadbox]

You would think SLI wouldn't have these issues anymore. There was all the talk about hardware frame metering. I guess that was not true if they need drivers to fix it.

This late in the game, new arch or not, you can't be having issue like these and we need to pressure nVidia over this, we can't just say "give them a break". We need to be on them. You as a focus group member should not be telling us to give them time, you should be relaying our frustrations to them.

After all the bad press AMD got, Crossfire on Hawaii is the best implementation of mulit-GPU tech around. That happened cause we raised out voices and showed our displeasure. Nvidia should be held to the same standard.

Do you not agree?

The original term was hardware based frame metering. The fanboys in their haste to talk up their brand took the "based" out of it and it became hardware frame metering coz that sounded sooo much better. As for focus group members, I think the brief is to talk up the positives and downplay the negatives.

 

[Skurge]

All hardware works in conjunction with some sort of software, or firmware. What is frame metering hardware without any software? Nothing but silicon and transistors.
Your frustrations are pre-mature. GM204 has been out about two weeks.
(N/M I thought you had 970 SLI for some odd reason.)
AMD got bad press with an already long in the tooth arch when TR unleashed it's frame latency/pacing review methods article. Difference is, Nvidia will improve this on their own. You'd have to hammer AMD for months and maybe years for that to happen.

So you don't agree? So we should just sit here quietly and hope it's fixed. This should not be happening even on new cards. I thought we were past this since Hawaii did not have any crossfire issues on launch, you would expect nvidia to at least do as well if not better.

I was thinking of swapping cards as the 970s are much cheaper than my 290s and save on heat and power, but the 4K performance is not what I was hoping.

 

[SirPauly]

It's still good to see FCAT investigations -- with third party findings and their awareness may help improve the quality moving forward.

 

[Keysplayr]

So you don't agree? So we should just sit here quietly and hope it's fixed. This should not be happening even on new cards. I thought we were past this since Hawaii did not have any crossfire issues on launch, you would expect nvidia to at least do as well if not better.

I was thinking of swapping cards as the 970s are much cheaper than my 290s and save on heat and power, but the 4K performance is not what I was hoping.

And then we take a vacation from the illusion that we live in a perfect world, and come back to reality. New architecture. Normally, that would be all a person needs to hear. Normally. By your logic, no GPU should ever have an issue again. Do agree that will not happen?
I'm all for FCAT testing in all GPU reviews. I think its a fantastic and extremely useful testing method.