Adding a new GPU - GTX 770

[unclefarkus]

So I'm looking to upgrade my GPU. Part of the reason is for games, the other to start taking Max rendering to a different level. I've researched a little bit and asked advice from gamer friends and they have all pointed me to the GTX 770. I currently have a 570. Their suggestion has been to use the 570 as a dedicated PhysX card and the 770 as a main.

 

For gaming, this makes complete sense. But how will it fair with rendering? The 1536 CUDA cores of the 770 vs the 480 of the 570 seems like a good improvement not to mention going to 2gb of VRAM from 1gb. So. Two questions:

 

1) Will I experience any benefit when rendering from having both cards or will it just utlize the 770?

 

2) Is the 770 enough of an upgrade in terms of rendering?

 

For reference:

 

My 570: http://www.newegg.com/Product/Product.aspx?Item=N82E16814121432

The 770 I'm considering: http://www.newegg.com/Product/Product.aspx?Item=N82E16814121770

 

My rig: Intel Core i7-3930K Sandy Bridge-E 3.2GHz, 16gb RAM, GTX 570

 

Thanks

[Tommy L]

This guy is the authority on GPU rendering with Max:

http://jeffpatton.net/2010/11/gtxquadrotesla-my-opinion-on-todays-gpu-selections-for-rendering/

[Dimitris Tolios]

The figures are from the VRay RT Benchmark scene, in Chaosgroup's forums.

 

The 770 is a GK104 card, based on the Kepler Architecture. The 570 is a GF110 card, based on Fermi. Yes, they call them "CUDA Cores", or shaders in both cases, but comparing them directly is a bit apples vs. oranges.

 

I don't have figures for the 770 per se, but the 770 is not much more than a 680 with 7GHz Ram instead of 6 GHz Ram. Faster in some stuff, but GPU rendering is not hugely affected by RAM speed past one point. I would expect the 770 to be very slightly faster than the 680, so give or take 40% faster than your 570 in GPU rendering tasks.

 

 

 

Notice that the 580 is still faster than the fastest GK104 card.

Apples for apples, each CUDA core in the Fermi architecture of the 580, is more than 2.5 times faster than each of the Kepler architecture CUDA cores. My estimation is that to match the raytracing performance in VRay RT GPU of the full 512 cores in the GK110, a kepler card with similar clocks with the 680 has, should have around 1880 cores. Between Kepler cards, the performance is much more predictable, as for GPU rendering we see a nearly linear corelation of real performance vs. core*Mhz count.

 

In gaming, the difference will be much more pronounced, and in favor of the 6xx and 7xx series.

Keeping the 570 as a "PhysX" accelerator is a silly gimmic imho...I've been playing Borderlands II, perhaps the most "PhysX" heavy game just fine on a single overclocked 670 and 1440p just fine.

 

Also the 570 is horrible as far as power consumption and heat generation goes (ok, the 470/480s are worse) and this doesn't change much when idling (unlike 79xx Radeons and Titan/780s that are also power-hogs, but at least idle @ low watts).

 

So, for gaming, I would say the 570 is kinda of a waste as a secondary accelerator...

For GPU rendering, it will add a good chuck of speed in your setup, given you can fit your scenes in the 1,2GBs of available VRam.

 

Also - some thoughts on coolers:

 

The good: When you have a sigle card in your system, the open shroud ones like both of those Asus DCIIs, MSI twin force, Gigabyte Windforce 3, EVGA ACX etc are better. Fresh air is sucked in from more places, and pushed out sooner than the conventional (for the last few generations of GTX and Radeon cards alike) blower type coolers, where one large blower @ the rear end of the card, sucks air and pushes it all-along the card, to exhaust it in the back. The "open" type coolers, work better than that - for single cards.

 

The bad: When you start adding more than one card, one or more of them is forced to breathe partially the "pre-heated" air exhausted in the immediate vicinity of its intakes. Most cases have their intake case fan(s) @ the lower front, so usually the GPU lower in the stack gets the fresh air, while the ones above start getting a mix of fresh and "hot offshoot" by the cards below them.

 

Thus the cooling potential of open-type coolers is diminishing, and the blower types become the preferred choice for multi-card setups.

 

　

Of course, the more air flow your case has (the more intake fans directly facing your GPUs @ the front and sides), the better, and perhaps the less affected the open shroud coolers will be by neighboring cards, but the blowers still offer more consistent results.

 

That's a rant cause I was bored @ work. Doesn't mean that you you won't be able to use what you have, just hits for making choices if you were building something from scratch.

Edited August 13, 2013 by dtolios

[unclefarkus]

Both of these posts have been very helpful and I have learned a lot

My current setup was pretty much built by a computer science friend and I just took him at his word for everything. He isn't around anymore, so now I'm trying to learn myself. Very helpful -- thanks!