Geforce GTX 1070 - Specview 12 Benches

[Dimitris Tolios]

Initial numbers are impressive imho. Easily tops the charts vs. anything Maxwell had to offer and apparently strong OpenGL drivers in-tune with AMD's R9 (in many cases including fast Quadros like the K4200 in Catia viewport).

 

 

 

Full set of results @ pcfoo.com

 

 

Card used was a Gigabyte GTX 1070 G1 Gaming, "downclocked" to FE/Reference specs.

 

Also note that numbers vary greatly from other sources because of my choice of running an overclock CPU on by bench rig: some of these GPU tests are actually CPU bound. I had no Titan X to test myself apples to apples, but paired with a stock 5960X the X scores around 123 score for Maya 2013 & 48 for Solidworks 2013, values the 1070 tops easily. In Solidworks that apparently is not GPU bound, the 970 also topped the Titan X having that 4.5GHz CPU advantage, another hint on how to invest your money smart if you can afford separating your modeling & rendering box(es).

 

Easy to guess the 1080 being even faster, but already we are looking at 30% increase in many cases over the 980Ti / Titan X. Yes, the 970 was a better value @ $100 less, but the 1070 is a monster.

 

Soon to be updated with Specapc 3DS 2015 numbers (pretty good).

Did not get Specapc Maya 2015 to complete. Will update in time.

Edited July 5, 2016 by dtolios

[RyderSK]

Where did you source all the cards :- ) ? Great work.

 

I could alternatively complete this with gtx 1080, Titan-X and even friend's M6000, but for 2016, 2017 alternatively 3ds Max.

[numerobis]

Titan P is coming

 

http://vrworld.com/2016/07/05/nvidia-gp100-titan-faster-geforce-1080/

[Dimitris Tolios]

There is no doubt that we will see a "big Pascal" and its binning derivatives at the near future. The way pricing is working though makes me worried.

[RyderSK]

Titan P is coming

 

http://vrworld.com/2016/07/05/nvidia-gp100-titan-faster-geforce-1080/

 

What ? No 24gb memory as leaks suggested ? Hm.

 

I am surprised it's bottlenecked by CPU, looking forward to tests. But yeah, this is impressive upgrade path. Testing Redshift as we're speaking, the 2.0 runs quite nice on my single Titan-X.

 

The way pricing is working though makes me worried.

 

Pricing is getting ridiculous on everything.

[Dimitris Tolios]

Doubt the could do 24GB...afaik the only GDDR5x chips in production are 8Gb (1GB) and even that production rate is a bit short: allegedly that's why we see shortages in the 1080 supply.

 

I mean, they would probably not fit 24 chips on a single board...makes sense to fit 6~8x on each side of the PCB as they did with previous generations.

 

I am perplexed with the 16GB tho...will that mean the GP1xx will be 512 bit bus? Or 384bit bus like previous Titans / Gx110 chips with 4GB running on half the bus (tricks they were pulling with the 660 3GB)? The 12GB makes perfect sense with a 384bit memory interface & 128bit GDDR5 chips.

[numerobis]

At least the 16GB version features HBM 2 with 4096bit -> 4 HBM stacks of 1024bit. It seems max would be 8 stacks = 32MB

I think it's basically a Tesla P100 (GP100 and also 16GB) with higher clock rates (maybe up to 1TB/s instead of 720GB/s for the RAM).

Maybe the 12GB version is the 1080 TI.

Edited July 9, 2016 by numerobis

[Jason Matthews]

Thanks for posting this. I was on the fence between the 980TI and the 1070. I ordered both from Newegg yesterday. I am planning on returning the 980TI unopened.

http://www.newegg.com/Product/Product.aspx?Item=N82E16814126109

http://www.newegg.com/Product/Product.aspx?Item=N82E16814487160

 

I am curious though. The 980TI has 2816 cores and the 1070 has 1920 cores. Is this going to impact Vray RT? I mainly bought it for the 8GB memory (Lumion), but it would be a nice bonus if the performance is as good in Vray RT as the 980TI.

[xjapanese]

....and how about Radeon RX 480 8gb ..looking forward to someone sharing some thoughts or experience on how it will perform in vray rt compared to gtx 1070 gtx 1080..

[Dimitris Tolios]

Most of the cards in the charts above are "old-er" cards as I source most of them out of eBay on prices I am confident I can resell them for (losing shipping and or few $ here and there) or borrow them from friends. Some of them are cards for personal use, including the 1st gen Titan, the 970, 1070 etc.

 

The Vray RT benches are complicated with newer cards, as often there is lack of optimization or even lack of compatibility with newer models, especially so to be honest I've stopped trying. I also don't get any freebies on the software side of things and I Don't have the latest VRay anymore, so I just rely on Chaosgroup's forums for hints of performance with newer cards if someone asks. I would recommend expanding your search for RX 4xx performance there (did not see much being reported as of this day tbh, the 10xx cards do seem to work off the bat tho with good results).

 

 

In Luxmark V2 OpenCL render - again, dated as V3 is out, but my database to date was for V2 and I cannot repeat the test for more than a handful of cards that haven't been sold already - the 1070 again is doing very well.

 

Room Scene:

970 SLI : 3344

1070: 2437

970: 1516

R9 270: 811

750Ti : 626

Titan 6GB (Kepler) : 615

 

Sala Scene:

970 SLI : 5877

1070: 3734

970: 2783

R9 270: 1545

Titan 6GB (Kepler) : 1374

750Ti : 1069

Edited July 11, 2016 by dtolios