Jump to content

ATI or NVIDIA


Recommended Posts

This is a commonly asked question but I'll throw it out anyway. I'm looking to buy a new Graphics card. I'm looking at a choice between an EVGA GeForce GTX 680 1006MHz 2GB PCI-Express 3.0 HDMI or ATI Radeon 7970 HD 925MHz 3GB PCI-Express 3.0 HDMI. The ATI has more on-board memory but the GEFORCE is CUDa enabled.

 

Is it worth getting a CUDA card with less memory then the ATI.

 

I'm running 3DS Max design 2012 & 2013

Link to comment
Share on other sites

I would go with a 580 w/3gb instead and pass on the 6xx series until the next dev cycle.

also, I would stick with nvidia. It's not necesarily the better performer of the two but they have garnered more industry support so you may experience fewer problems with compatibility in the long run

Link to comment
Share on other sites

At this point of time I don't think 79xx has driver compatibility with VRay RT GPU - the only GPU accelerated rendering engine that supports Open CL. If and when it does, it will probably be quite fast as the 79xx tops OpenCL benchmarks easily thus far.

 

GTX680 is not properly utilized by most CUDA applications too, as the simpler shader architecture Kepler utilizes is different than what developers optimized for (based on Fermi architecture). It might get better, it might not. We still don't know if kepler GTX cards where crippled on purpose in order to limit GeForce cards replacing Quadros in GPU calculating tasks, as it was the case for the Fermi generation. For the most bang for your money, the GTX 580 3GB is still the best choice, tho lacks a bit in gaming speed and energy consumption / noise / generated heat in comparison to both the cards you've mentioned.

 

If you don't care about complex scene GPU rendering, VRam won't be an issue...2GBs are enough for the average workstation viewport acceleration and gaming on 2-3 screens at the same time.

The ATI/AMD card has more raw power, but GTX680 is better for games and has CUDA...

I don't know about Kepler, but last 2 AMD generations were also faster/smoother in viewport acceleration than similarly or more expensive Fermi based Geforce and Quadros alike.

Edited by dtolios
Link to comment
Share on other sites

Actually, don't get the 680 if you want 600 series. get the just released 670. They have 4gig vram models and they have a better price to performance. however, Andrew will probably chime in on this, there has not been any testing in regards to how the 600 series relates to architectural/visualization programs. Everything tested has been strictly gaming.

Link to comment
Share on other sites

If you decide for 680 GTX, there are already few models with 4GB on market, only few bucks more expensive than 2GB. Just for future proof, even with poor gp-gpu performance, I would opt for this one.

 

Valid point...

 

If you don't want to over-spend, you might also wait for the GTX 670 that is out already too...I would say it will be 90% or better the performance of the 680 - or more than that. The current Anandtech review for the 570 measures performance difference of 4-10% in OpenCL and computation applications. They expect the 670 2GB around $400 which is an improvement over both 580 and 680. And uses some 10% less power than the 680.

 

Why is power so important? Comparing older 40nm GPUs like the Fermi with newer architectures like the ones AMD or nVidia kepler, means you will save the price difference over a 3GB 580GTX on electricity bills in just a year or so...

 

The consumption difference is about 60-65W - average.

That means that for 8 hours up-time a day, you will have about $40 a year extra running costs just for opting for a GTX 580 instead of a 680! So if you are planning to keep your GPU for more than couple of years, even a $100 price difference is diminished (and you are "greener" too!). $40 is for the US (California in my case) where $/kWh is lower than most European countries i think.

 

Both Kepler cards are also much quieter under load (because the run on less Watts).

It's a big difference having gtx 670 @ 50db under load, a 680 @ 52.5 db and the 580 roaring at 60db. 10db difference is about double the noise (it's not a linear scale).

Normal conversation is about 60db, that means that by having a 580 in a poorly insulated case 4ft away might stop you from actually hearing conversations around you...it's that loud! The 79xx series is around 54db.

 

(today i am posting stuff adding nothing more than what someone posted before me while i was writing, sorry)

Edited by dtolios
Link to comment
Share on other sites

Interesting to read 580 went up to 60db, pretty crazy.

I didn't know 670 came up with 4GB models, that is great news, and I might honestly buy one out of interest.

 

I don't know if 670 is coming out any time soon with 4GB versions and for how long the initial launch quantity will be available...

 

I don't know when 680 will be actually available with that buffer - haven't spotted any...most online vendors here in the US have trouble stocking 680 2GB since its launch - it's back-ordered and as soon as they receive a bunch is already shipped to people pre-ordering it...3rd party amazon sellers are speculating charging a nice premium ontop of the normal price etc etc...

 

For sure the 570 is a good value - still faster than the 580 in whatever the 680 was faster at, and about as fast as the latter for CUDA computation. 20% cheaper, and quieter...sounds like a win-win, given you were willing to spend $400 or so anyways...I don't know how it would compare to a V5900 you can buy for similar $/€ (which is also just 75W and pretty silent) as Andrew suggests in his system builds - being probably the fastest in this $ class for viewport acceleration - but the 670 is a peppy card undoubtedly.

Link to comment
Share on other sites

Those recommendations need an update badly. I can't do it for a couple weeks but I will. Meanwhile, my advice:

 

Do you plan to do GPU rendering? If so, nVidia's got the edge. And you don't want the 600 series - it's got a ton of "cores" (not the right word for it, but it's what they use) but the cores are each less capable, and with the changes in the architecture, the GPU renderers don't run particularly well on it. For example, I read one guy recently comparing the 680 to the 470 in Vray RT-GPU and the 680 was only 30-something percent faster. So given that and the difficulty of even finding a 680 that's not the 2GB model, I'm sticking with the 3GB 580 as my recommendation for GPU rendering.

 

If this isn't for GPU rendering, I wouldn't necessarily bother buying anything at that price point. I'd still take a more modest current generation FirePro or a consumer type video card that's a notch or two under the high end.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

×
×
  • Create New...