GPU Advice, GTX590

[antimatter27]

Hey all.

I'm in the process of a new build and am tossed on what to do for graphics despite reading every article ever written on GPU's authored in the last 6 months. I've officially gone nuts trying to make a decision, which has not been helped by things like "Inside the Second" and the comparable Tom's Hardware article on micro-stuttering.

So I'm turning my noob self over to the better advice of others!

What I really want is a rig that can run every current game at absolutely max'd out settings, single 1080p screen only. I'm talking Metro, BF3, Crysis + High res/DX11, and the upcoming Arkham City totally max'd without a single thing lowered one iota. And I want flawless silky smooth gameplay. I'd like to accomplish this with the least power consumption, heat, and noise possible. At the cheapest price would be nice too.

Because I enjoy graphics more than gameplay (I'm a terrible person, I know), I've basically ruled out AMD as I want PhysX. I'm actually willing to pay more money to see glass shatter and cloth ripples in all of three game titles. I have massive OCD attention to detail.

I know that general wisdom is that with a single 1080p screen dual gpu's is wasted, but benchmark evidence suggests otherwise. It seems rather that memory is wasted, e.g. non-reference 3gb 580 SLI at 1080p would be overkill. From all that I've read it seems that a single GPU solution can't deliver on what I described above. Even a superclocked 580 in Metro is getting FPS in the 30-something range, which isn't too hot. There's a small chance I might go 3D in the next year if the price of those screens come down, in which case the performance hit would, I think, destroy any chance of a single gpu option working.

It seems to me that a GTX 590 is the best solution. It can do all of the above, with less heat/power/noise than two 570's or 580's. Granted 570/580's can be OC'd, but I don't care about FPS bragging rights. To me, either a system can deliver acceptable performance (max'd out, smooth), or it can't. Having 80fps instead of 70fps doesn't interest me.

I am however worried that I'm weighing FPS to heavily. Perhaps 35 fps in Metro on a 580 feels smoother than 55fps on a 590 given the issues of microstuttering, but from reading "Inside the Second", looking at every chart, and reading accounts of people who own the card, there doesn't seem to be enough jitter to cause any noticeable stuttering. Future more taxing games that bring it's fps down into the 30-40 range might change that, but by that time it might be reasonable to upgrade to Kepler or Haswell.

Whatever card I get I plan on coupling with a 2500k on an Asus gen 3 board, probably upgrading to ivy bridge and/or kepler in 1-2 years.



So am I being a wasteful idiot here leaning 590 over 580? And on a sidenote, does anyone know how much a 590 can be OC'd (if at all) now that overvolting has been disabled? Have I over thought all of this massively? 😛


Thanks.

 

[bystander]

Those numbers definitely show that I'm dead wrong. I remember reading in so many reviews comparing nvidia's latest offerings to AMD's the reviewers saying that AMD is better at uber resolution because they have more memory, and that some of the latest titles perform better at 5760x1080 on AMD for that reason. I must have misread-misremembered-misunderstood.

/noob mistake.

You were right, when referring to uber resolutions. It's just depends on the game and settings used, but at 1080p it's extremely rare that more than a 1GB is needed.

 

[rolli59]

You were right, when referring to uber resolutions. It's just depends on the game and settings used, but at 1080p it's extremely rare that more than a 1GB is needed.

Yep! http://www.tomshardware.com/reviews/radeon-hd-6950-1gb-benchmark,3041-13.html

 

[Warmacblu]

If the price difference isn't a concern, then why not purchase a card with more VRAM? It's worth it for at least a little future proof potential.

 

[FtsArTek]

Just gonna note here: I have two 6950s in Xfire, and a GT520 as a PhysX card, and it runs great. Not the easiest thing to get workng at first, but soon sorted.

 

[cbrunnem]

dont fool yourself.... kepler is NOT going to bring 70-80% better performance and this is what your relying on being faster so an upgrade would be a good decission but were ever you got those numbers, i want what they are smoking. everything but architecture the same there is NO way kepler will be 80% faster lets get real now.
thats like sandy bridge being 80% faster then the first gen i series which it MIGHT be 30%

 

[bystander]

dont fool yourself.... kepler is NOT going to bring 70-80% better performance and this is what your relying on being faster so an upgrade would be a good decission but were ever you got those numbers, i want what they are smoking. everything but architecture the same there is NO way kepler will be 80% faster lets get real now.
thats like sandy bridge being 80% faster then the first gen i series which it MIGHT be 30%

I thought they were working on a die shrink and more transistors. I'd expect more improvement than the last series, since they did not do a die shrink, that goes for both companies. 70%-80% may still be pushing it, but I'd expect a minimum of 30%, not a max.

 

[cbrunnem]

I thought they were working on a die shrink and more transistors. I'd expect more improvement than the last series, since they did not do a die shrink, that goes for both companies. 70%-80% may still be pushing it, but I'd expect a minimum of 30%, not a max.

should i bring up bulldozers expected performance gains?

 

[spagalicious]

Yeah this may have been an overstatement but a companies cpu performance is definitely not directly related to gpu performance. Just because intel makes a good cpu definitely doesnt mean they make a good gpu...

 

[antimatter27]

Just out of curiosity, how much exactly does throwing in AA add to vram? Or what are the factors that most increase the need for more vram. AF, DOF, SSAO, texture quality, PhysX?

 

[cbrunnem]

Yeah this may have been an overstatement but a companies cpu performance is definitely not directly related to gpu performance. Just because intel makes a good cpu definitely doesnt mean they make a good gpu...

my point was that the performance increase that bulldozer was expected to bring was drastically over stated and that could be the case with kepler.

 

[bystander]

my point was that the performance increase that bulldozer was expected to bring was drastically over stated and that could be the case with kepler.

Bulldozer was definitely a let down, but don't expect the same to happen with the GPU's. Last generation was a let down due to not doing a die shrink, so they couldn't really push the shader count without heat issues, but this next generation will be on a new process.

It's also a lot easier to increase GPU performance through brute force (adding more shaders). The GPU's utilize more and more shaders with ease, unlike CPU's which require new software to take advantage of more cores.

While 70-80% surely is on the optimistic side (The 5870's did achieve this kind of improvement btw), but don't expect it to be lower than a 30% boost. The new shader counts are way up from current cards.

 

[gnomio]

Nvidia's reference cards just lack the memory for uber resolutions, AMD's don't. You can find benchmarks that show this very clearly, especially when 1.5gb 580's in sli are swapped out for 3gb models. 2 x580 3gb models > anything AMD has in sheer performance. But that's talking about eyefinity vs. nvidia surround stuff and I'm a single monitor 1080p person.

I guess I'd need to dig up some benchmark evidence that modern games break 1gb of vram with ease at 1080p. I've read so in multiple forums that games like Metro, Shogun, BF3, etc. can do so easily, especially once AA is thrown in.

As a sidenote on the sli front AMD's tech has no frame metering technology, whereas nvidia's does...I sort of wonder if this is why AMD's current generation scales better than nvidia's, that nvidia is purposely hurting their benchmarks to provide a smoother experience.

Do not thonk of the vram as storage. Think of it as a buffer. It gets filled the gpu clear it, it gets filled the gpu clears it. Remember it has 30 cores that help it do that. On metro and the dx 11 games its not the ram but the processing power that brings it down

 

[antimatter27]

Best answer selected by antimatter27.