Guru3D 970 SLI review

[3DVagabond]

Guru3D

Guru3D has done a GTX 970 SLI review. Scaling and driver support seem really good. Especially for a new SLI architecture. While we've had Maxwell for a while, GM204 is the first to offer SLI.

Two problems though, performance really tends to tank @ 4K and, according to FCAT results they are dropping frames like crazy. Is hardware frame pacing broke in Maxwell?

FCAT

@4K

I wish they had included 290 CF results since that would be the closest price comparison.

 

[desprado]

Guru3D

Guru3D has done a GTX 970 SLI review. Scaling and driver support seem really good. Especially for a new SLI architecture. While we've had Maxwell for a while, GM204 is the first to offer SLI.

Two problems though, performance really tends to tank @ 4K and, according to FCAT results they are dropping frames like crazy. Is hardware frame pacing broke in Maxwell?

FCAT

@4K

I wish they had included 290 CF results since that would be the closest price comparison.

U right stuttering is every where with 970 or 980 SLI and because they are new right know.
but did u see the Interview of Nvidia Toms.He said on Pcper that they are coming with something new for SLI which will generally increase more performance and smoothness he also said that maybe it will be more smooth than a Single GPU.

 

[3DVagabond]

U right stuttering is every where with 970 or 980 SLI and because they are new right know.
but did u see the Interview of Nvidia Toms.He said on Pcper that they are coming with something new for SLI which will generally increase more performance and smoothness he also said that maybe it will be more smooth than a Single GPU.

Strange that Maxwell would be a step backwards. I thought nVidia beat this years ago? So the cheap 970 SLI setups don't seem like such a good idea.

 

[desprado]

Strange that Maxwell would be a step backwards. I thought nVidia beat this years ago? So the cheap 970 SLI setups don't seem like such a good idea.

No 334.07 was just an testing driver.I searched and i cannot find any Issues with 970 or 980 SLI with 334.11 or 334.16 which solved issues of Mirco pause.

 

[Keysplayr]

For the games tested with FCAT:
Tomb Raider
"SLI card performance wise there are no stutters recorded, BUT there is good and bad news, the bad news is that you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops at 6 to 8 ms like these can not be seen with your eyes (that's the good news)."

Thief
"The first 15 seconds of the Thief benchmark sequence are difficult with the rain and everything. That results in small stutters and hitches, all cards tested show this visibly on screen. Once it passes the first 15 seconds we no longer detect a significant enough glitch or stutter in frametime scaling (chart wise) that is extremely significant or long-lasting.

But again, you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops like these can not be seen with your eyes.

For those that do not understand what you are looking at, the above is a gametime scene recorded for 28 seconds. With this chart, lower = better. Huge spikes above 40 ms to 50 ms can be considered a problem like a stutter or indicate a low framerate. "

Bioshock Infinite:
"On this 30 Second run the graphics cards all manage a good result. Despite one or two very small spikes we cannot detect a glitch or stutter that is significant or long-lasting. For those that do not understand what you are looking at, the above is a game-time scene recorded for 30 seconds.

Here again, you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops like these can not be seen with your eyes."

It appeared that SLI drivers (as per Tom Petersen) for the 9xx (344.07) series needed some tuning. That came as it always does. The good news was it couldn't be seen.

Maybe Guru3D should re-run this test with newer driver. 344.16 was released Tuesday the 23rd of Sept.
344.11 were out before that.

Thanks 3D for the info though.

 

[Silverforce11]

Frame-drops at 6-8ms are half frames, would indicate runt frames?

That would imply the displayed FPS obtained is not real compared to the frames you observe, thus, its artificially inflated like back on AMD's Tahiti CF. Its been said that such a drop of frame means you need much higher reported fps to appear smooth (if you're sensitive to that).

It is a shock because Maxwell is a better Kepler, and they had plenty of time to work on drivers, its not like they are in a hurry to launch Maxwell given the current competition and market.

 

[iiiankiii]

I guess until this gets fixed, there's no point in getting these cards if you value your gaming experience at 4k. I would rather have smoother gameplay at the cost of a few FPS. Those runt frames can impact gameplay negatively to gamers. I'm sure these will be fixed someday, though.

For a single card solution, the GTX 970 is hands down the best overall gpu Nvidia has put out in years.

 

[parvadomus]

wow NV and AMD exchanging positions in smoothness in only 1-2 years :/

 

[wand3r3r]

You'd think, considering NVidia developed the FCAT to demonstrate the competitions poor dual card solutions at that time, that they'd perfect this before shipping.

I guess it'll be fixed, it just shows they are doing software smoothing too. I wonder if SLI FPS will drop when they smooth this out?

 

[Enigmoid]

The problem is that the release drivers likely don't support this kind of stuff.

http://www.nvidia.com/Download/Find.aspx?lang=en-us

Do a review with the 344.16 drivers.

 

[wand3r3r]

They don't support SLI with release drivers?

I would expect after the big NV marketed slower but smoother campaign, proven using NV's developed tool fcat, that they would be on top of that. (In the end it benefited consumers)

I guess we have to see if the newer drivers fixed it, then it's pretty much a non-issue.

 

[Attic]

That's a very surprising result. nVidia should have this issue on lockdown though so personally I wouldn't wait to get 970SLI based on the poor SLI results shown here if that's what I wanted. It's a bit troubling that nVidia has let this side of things slip a bit though, particularly when faced with the competition XDMA and the 290/x brought to the front here.

I'd like to see a few more sites touch on this with new driver (not just a PR word from nVidia), particularly [H] who is very good at giving a gameplay breakdown.

Review sites must have been really pressed on time with cards before releasing reviews. I'm seeing very late followups for 970 and SLI results after the initial 980 reviews dropped from a lot of sites.

 

[Subyman]

I'm not too worried. First driver on launch is sure to have some issues.

 

[3DVagabond]

For the games tested with FCAT:
Tomb Raider
"SLI card performance wise there are no stutters recorded, BUT there is good and bad news, the bad news is that you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops at 6 to 8 ms like these can not be seen with your eyes (that's the good news)."

Thief
"The first 15 seconds of the Thief benchmark sequence are difficult with the rain and everything. That results in small stutters and hitches, all cards tested show this visibly on screen. Once it passes the first 15 seconds we no longer detect a significant enough glitch or stutter in frametime scaling (chart wise) that is extremely significant or long-lasting.

But again, you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops like these can not be seen with your eyes.

For those that do not understand what you are looking at, the above is a gametime scene recorded for 28 seconds. With this chart, lower = better. Huge spikes above 40 ms to 50 ms can be considered a problem like a stutter or indicate a low framerate. "

Bioshock Infinite:
"On this 30 Second run the graphics cards all manage a good result. Despite one or two very small spikes we cannot detect a glitch or stutter that is significant or long-lasting. For those that do not understand what you are looking at, the above is a game-time scene recorded for 30 seconds.

Here again, you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops like these can not be seen with your eyes."

It appeared that SLI drivers (as per Tom Petersen) for the 9xx (344.07) series needed some tuning. That came as it always does. The good news was it couldn't be seen.

Maybe Guru3D should re-run this test with newer driver. 344.16 was released Tuesday the 23rd of Sept.
344.11 were out before that.

Thanks 3D for the info though.

That's because they were running very high framerates for those tests. It's dropped frames and lots of them. If you remember when AMD had exactly the same results it was reported that you needed higher FPS to feel as smooth. I'm sure you remember. Those tests were at +100fps by Guru3D. There's also the possibility that Hilbert isn't sensitive to MS, but that would just apply at slower fram rates like we are seeing @4K, which would require SLI.

 

[x3sphere]

Haven't had any issues with stuttering on my 980 SLI, but I've not played the games tested in that review.

Mainly spent time with Witcher 2 and Crysis 3 this weekend at 4K and both ran great.

Crysis 3 runs noticeably smoother than it did on my R9 290 CF setup, which is interesting as I felt quite a bit of stutter in that game with the 780 SLI setup I had previously. Could have been a lack of VRAM, though.

Edit: Should add I was also on z87 when I had the 780 SLI setup, which was limited to x8 PCI-E in dual GPU mode... wonder if that had something to do with it as well. I'm on X99 now so have dual x16.

 

[iiiankiii]

For the games tested with FCAT:
Tomb Raider
"SLI card performance wise there are no stutters recorded, BUT there is good and bad news, the bad news is that you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops at 6 to 8 ms like these can not be seen with your eyes (that's the good news)."

Thief
"The first 15 seconds of the Thief benchmark sequence are difficult with the rain and everything. That results in small stutters and hitches, all cards tested show this visibly on screen. Once it passes the first 15 seconds we no longer detect a significant enough glitch or stutter in frametime scaling (chart wise) that is extremely significant or long-lasting.

But again, you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops like these can not be seen with your eyes.

For those that do not understand what you are looking at, the above is a gametime scene recorded for 28 seconds. With this chart, lower = better. Huge spikes above 40 ms to 50 ms can be considered a problem like a stutter or indicate a low framerate. "

Bioshock Infinite:
"On this 30 Second run the graphics cards all manage a good result. Despite one or two very small spikes we cannot detect a glitch or stutter that is significant or long-lasting. For those that do not understand what you are looking at, the above is a game-time scene recorded for 30 seconds.

Here again, you can see a lot of framedrops. This issue continued persistently through all FCAT results. We are still investigating what is going on there. These are not stutters, frame-drops like these can not be seen with your eyes."

It appeared that SLI drivers (as per Tom Petersen) for the 9xx (344.07) series needed some tuning. That came as it always does. The good news was it couldn't be seen.

Maybe Guru3D should re-run this test with newer driver. 344.16 was released Tuesday the 23rd of Sept.
344.11 were out before that.

Thanks 3D for the info though.

I think that the reviewer just isn't able to detect those drop frames. Some others might detect those frame drops. The irony is apparent b/c people who owned AMD cards were saying the exact same thing; those frame drops didn't result in stutters. But some insisted it was a real problem b/c the hard data suggested runt frames. As some stated, others might be more sensitive to such frame drops. Personally, I couldn't tell.

 

[DigDog]

god forbid NVidia being allowed to make mistakes at launch. in bold.

 

[tviceman]

Looking simply at FPS, techpowerup came to a different result than Guru3D, showing that 970 SLI is unnoticeably slower than the 295x2 at 4k. TPU tests way more games than Guru3D as well.:

Also, throw in a third GTX 970 for the same price as the 295x2, get lower power consumption and better performance.

 

[Keysplayr]

n/m no sli testing.

 

[Skurge]

snip

I would get 2 980s long before i ever get 3 970s or 3 of any card for that matter.

 

[tviceman]

I would get 2 980s long before i ever get 3 970s or 3 of any card for that matter.

Yes I would too, I was just making the point that TPU showed 970SLI to be faster under a wider range of games that Guru3d, and that any complaints about possibly slower perf can be rectified and then some if willing to pay a similar price of 295x2.

Anyways, I am sure nvidia will fix any SLI issues Maxwell has in a fashion much more timely than it took AMD to implement a frame pacing solution to begin with.

 

[Soccerman06]

So why exactly does the 970 and 980 do so poorly in sli relative to the 780 series?

 

[rgallant]

So why exactly does the 970 and 980 do so poorly in sli relative to the 780 series?

no because it's a mid range low cost cards vs high end cards that were even neffed with low vram \clocks?[funny how b1 could not hit 1500+ on the same process]

 

[3DVagabond]

Looking simply at FPS, techpowerup came to a different result than Guru3D, showing that 970 SLI is unnoticeably slower than the 295x2 at 4k. TPU tests way more games than Guru3D as well.:

Also, throw in a third GTX 970 for the same price as the 295x2, get lower power consumption and better performance.

There's no denying the relative performance drops @4K. As I said, I wish we had 290 Crossfire results from Guru3D to compare. That would be a more similar price comparison.

As far as 3x cheaper cards compared to 2x higher end cards, nobody saw that when you could buy 3x 290's for less than 2x 780 ti, or heaven forbid, 2x Titan.

Don't ignore the runt/dropped frames, either. @4K they will be far more important because of the lower FPS.

Try not to cloud the situation with other comparisons that we don't have data for, like 3way SLI. And we know Maxwell uses less power. Just in case you might think differently, we know Maxwell uses less power.

god forbid NVidia being allowed to make mistakes at launch. in bold.

Kind of like 7970 release drivers? I can't tell you how many times I've read that AMD didn't care about their customers by releasing Tahiti before the drivers were up to speed.

 

[Silverforce11]

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.