SLI Performance Scaling

[exyia]

funny how nobody noticed that ALL of the 4 titles used are "AMD Gaming Evolved" titles. lol

 

*before internet nerds rage, it's just a funny observation - nothing more, get over it

[Faa]

Well you just can't go yell that 3-4Way SLI scales bad without having any idea if the CPU was the limiting factor or the profile being bad. You need a lot of CPU horsepower for 3-4Way.

Giving an example;

No difference, has nothing to do with 2Way SLI scaling bad at all but the CPU was a massive bottleneck for 2Way SLI and even for a single card. If I move somewhere else the GPU's crank up to 99% each with twice as much fps because at that point there's no CPU bottleneck. You won't see a performance gain from 2Way SLI if you bottleneck a single card already. Same thing no gain from 3Way SLI if you bottleneck 2-Way SLI. Again no gain from a 4th card if you bottlenecked 3Way SLI. 

GPU's perform best at their maximum load, not at 30% load but at 99%. Runs at maximum load -> 99%. If you have 4-Way SLI at full load and the scaling is bad, then it's the profile/scaling being bad but if we don't get any information at what load the GPU's were running at then you can't just yell SLI scaling is bad because it could be a CPU bottleneck instead.

If all of this sounds BS, well I'll just drop a link where a mod at the geforce forums explains what I've been saying: https://forums.geforce.com/default/topic/532913/sli/geforce-sli-technology-an-introductory-guide/post/3749687/#3749687

Any specific reasons power consumption wasn't tested in Watch Dogs, but in FC3 where we've seen nearly lineair scaling? Looking at the FC3 results, nearly lineair, so I assume 99% -> 99%/99% -> 99%/99/99% -> 99%/99%/99%/99% (perhaps a small CPU bottleneck in 4way but whatever), if the power consumption test was done in WD where I assume was CPU bound, there would barely been a difference in the power consumption results.

I've played Watch Dogs a lil bit, not long enough, but with my 780's in 2Way SLI and my 3930K at stock there was a shitload of CPU bottlenecking with average gpu loads of 50/50% meaning I wasn't anything with a 2nd card. Honestly this video lacks a bunch of details, it's just another classic review that just benches and drops the results. Tell us if the profile/scaling was bad or if the CPU was causing it or just don't benchmark games youre being bottlenecked by the CPU. I'm not going to benchmark 4Way 980 SLI at 720p low settings and complain about scaling being bad when its freaking obvious it had nothing to do with scaling at all and with a lil bit more thinking you'd realize there won't be any difference between 1 980 & 2 980's. On the other hand they are testing CPU's with the GPU being a bottleneck showing us no difference, why the hell would we bench gpu's with a CPU bottleneck?

[The Gaige]

This video just tells me what I already know & that is buy the highest 2 cards you can get for 2 way SLI & I'm getting 2 970's or 780's at tax time!

[MAXXHEW]

And if you want even better performance for an SLI set up... switch out that Intel processor.

http://www.tweaktown.com/tweakipedia/56/amd-fx-8350-powering-gtx-780-sli-vs-gtx-980-sli-at-4k/index.html

[Shepard_Pro_]

Hello everyone! I have gtx 780, but I am thinking about an upgrade some time later. What is better: adding another gtx780 in sli of go for 980?

[Faa]

And if you want even better performance for an SLI set up... switch out that Intel processor.

http://www.tweaktown.com/tweakipedia/56/amd-fx-8350-powering-gtx-780-sli-vs-gtx-980-sli-at-4k/index.html

Kidding me? Performance would be up to 80% lower.

[mosin40]

 

> MFW i7-4930k + X99

 

comic sans because why not

I was confused while watching the vid, especially on this part

@nicklmg Instead of i7-4930k (x79 and ddr3) it should say i7-5930k

[DeViLzzz]

Linus c'mon man.  I am still waiting to see how well the newer cards and even some older cards handle games with PhysX maxed and other features maxed at 1080p, 1440p, 1600p and so on.  A reason to get SLI times 2, 3 or 4 might be to help out with games that have PhysX with it being so demanding.

 

Also like I said in the comments on YouTube I think there are situations where people might want 3 or 4 cards for their personal rig when they are playing single player game or a non competitive multiplayer game but then when not doing that they could take out 2 or even 3 of the cards and put them in other pcs in the home for say when friends come over for a LAN party or whatever.  Investing in 3 or 4 cards doesn't have to be a bad thing.

[MAXXHEW]

Kidding me? Performance would be up to 80% lower.

http://www.tweaktown.com/tweakipedia/56/amd-fx-8350-powering-gtx-780-sli-vs-gtx-980-sli-at-4k/index.html

Did you read that link? Maybe I don't understand English well, or read graphs too good... do you? Looked pretty convincing to me though.

[Faa]

http://www.tweaktown...t-4k/index.html

Did you read that link? Maybe I don't understand English well, or read graphs too good... do you? Looked pretty convincing to me though.

So yeah there's no way a 8350 outperforms a 4930K. It can outperform it by 1 FPS because its GPU bound but run it again and a 6300 will top out, run it again a 4670K will top out - it's just the variation of the GPU's performance.

[jumbobreakfast]

Any chance you could do a video about frame pacing? Is this what causes problems with SLI in VR?