Maximum PC

Hey guys, hoping you can help me decide here. I have no loyalty to AMD or Nvidia and am all about bang for the buck and solid performance. Please no red v. green team arguments here. I have everything picked out for the upgrade (parts listed below) accept for the GPU(s). The 4890s have been absolute screamers and still are. I don't really "need" to upgrade anything but would like to be able to run DX11 and just have "the itch" the upgrade. I have no preference between single or dual GPUs and game at 19x12 resolutions and prefer to have every option maxed. If I go with a single GPU, such as the 580 or 570, I would look to go SLI in the future when I look to upgrade again. Don't really have a budged but would like to keep the GPU upgrade under $500.

Current setup is:
Asus Rampage Extreme
Q9650 @ 4ghz 24/7
Full on custom liquid loops (CPU, GPUs, and Ram)
4GB OCZ Flex DDR2 1200
MSI 4890 crossfire
Win 7 64

Upgrading to:
ASRock Z68 Extreme Gen 3
16GB Ripjaws DDR3 1600
2600k

What kind of PSU are you running. Wattage and amperage on the 12V+ please? Number of rails, both 12V+ and aux PCIe?

Good point, I should have had that in the original post. PSU is solid - Silverstone 1200 Gold, single rail 100v. Should be able to run darn near anything without breaking a sweat.
http://www.newegg.com/Product/Product.aspx?Item=N82E16817256069&nm_mc=OTC-Froogle&cm_mmc=OTC-Froogle-_-Power+Supplies-_-Silverstone-_-17256069

Yep. Silverstone makes great PSU's. At that capacity pushing 100A on a single rail you're definitely good there.

Well the baddest card on the planet is the AMD Radeon 6990 HD if you can find one, but it's likely to be a little out of your price range right now since volumes are so low.

The Radeon 6970 HD is also an excellent card. Massively large parallel 5 dimensional processing array and 2GB GDDR5 VRAM.

EVGA 025-P3-1579-AR GeForce GTX 570 (Fermi) HD 2.56GB GDDR5 (on a wider 320-bit bus) is an equivalent card on the nvidia side of the fence.

The Fermi based 580's are right at the edge of your budget, but if I was going to spend that much, I would wait for the Radeon 6990's to become more plentiful.

In the most recent shootout from MPC the 590 won nearly every category. Not trying to argue at all, but saying it's the badest card on the planet is highly debatable. I also don't see the 6990 getting anywhere near a $500 price point for several months still at least...same with the 590.

I was looking at the 6870x2 but the price point is retarded. Nearly $500 for the x2 when you can go 6870 xfire for around $350. Dumb. So I'm looking at:

EVGA GTX 580 - $370 (B Stock) none in stock currently
EVGA GTX 570 - $250 (B Stock) none in stock currently
EVGA 460 2win - $380 - beats the 580 in nearly all benchmarks
6870 Crossfire - $360ish - (soundly beats both the single 580 and 460 2win)
6950 Crissfire - $480 (not sure it's worth it)

The price is nearly the same for the 6970 and 580 (B stock) so with the price being equal, I don't think anyone can seriously argue the 6970 is bet. That is why I have the 580 and 570 as consideration in single card format (for future SLI).

Page 78, September issue. XFX Radeon HD 6990 vs. Asus GTX 590. Out of 14 benchmarks, the 590 only won 5. In most games and 3DMark 2011 (Extreme) the 6990 was superior. You're talking about the quad-GPU benchmarking from the October issue, and at 2560x1600 @ 8X AA the cards split the benches. That is indicative of the fact that quad-SLI performance will drop off faster at higher resolutions than will quad-CrossfireX. AMD does need a driver update to address the Dirt 3 issue, or someone needs to put a stop to the use of Intel mobo's in test beds. Either way though irrelevant.

So the top five single GPU cards in order of performance would be:

1. Radeon HD 6990
2. GTX 590
3. GTX 580 SC
4. Radeon HD 6970 (2GB)
5. GTX 570 (2.56GB)

The GTX 460 is a dual GPU config, with a 2GB frame buffer. Crossfire 6870 (2GB) cards, would give you are larger and more capable processing array with a 4GB frame buffer. If you're looking for the best performance for the price in a dual GPU config, I'd go with the 2 x Radeon 6870 HD (2GB) cards.

There's also energy savings with the Crossfire config vs the Fermi SLI'd. Consider the point that a single Radeon 6870 uses close to 100W less (~300W) under load than a GTX 580 does (~395W).

Shouldn't expect the 6870x2 with 4GB VRAM (2x2GB on 256bit bus) to be close in price to two 1GB 6870's. You're neglecting the fact that the 6870x2 has double the frame buffer, and power users will pay a premium in order to be able to populate that extra PCIe slot with another card. 4 GPU's and 8GB VRAM is a formidable dual card config.

Good point on the frame buffer, hadn't thought about that. Not sure how much that translates into FPS at 19x12 res. The 6870 xfire setup is starting to look pretty appealing. Is the 2gb version worth an extra $50+ per card vs. the standard 1gb? That would put the 6870 2gb xfire at the same price as (or more than) a 6950 xfire setup. And while power consumption isn't a huge factor for me, it is something to consider and the AMD/ATI cards are definitely more energy efficient. Especially vs. the GTX 4XX and 570 series cards. The 580 isn't toooo bad...

Also keep in mind that whatever I end up with, it will have a full body waterblock and will be overclocked to whatever it will safely run 24/7. Temps won't be an issue and the waterblock(s) are not part of the $500 budget.

More and more processes these days are being aimed toward GPU computing as we find more uses for massively parallel arrays. It's not just video transcoding, encoding, and acceleration anymore. Designers are allocating more resources to developing processes that can take advantage of these arrays, and AMD is really pushing the idea with their view and demonstrations of hybridization. There are already tangible examples (with Brazos and Llano) of their vision and it is impressive. If Bulldozer lives up to it's FX moniker, as it looks like it will in the early stages, AMD will finally be able show that their idea of hybridization can increase data throughput exponentially by developing smart software that intelligently offloads work to the GPU on the fly.

One of their early demo's for the Zacate E-350 included a video of the dual 1.6Ghz Bobcat CPU's pinged at 100% running SuperPi, 1080HD video running seamlessly on the UVD shell, and the GPU (dual 40 core SIMD's) processing a physics load, all simultaneously with no lag in any of the processes. All on their flagship netbook processor. It's a far cry from smart-ware that can dynamically access or allocate system resources appropriately based on work type, but it shows the possibilities inherent in the idea. It is something that AMD has talked about since they acquired ATI, and it seems likely that very soon they will deliver.

I apologize for the long winded reply, but I didn't know how to get to this point without the background. The larger the frame buffer available to this increasing number of GPU processes, the more efficiently they will be able to process multiple workloads simultaneously (same as adding RAM to your system and the subsequent improvement in multitask efficiency). The more the GPU can handle the faster the CPU gets to data that it will be responsible for handling, and thus the exponential performance increase (in theory).

I assume you fold with gpu clients? I would go for the 2GB buffers for an extra $100. I don't think Microsoft has any plans on ramming a DX update down our throat anytime soon that will make the 6000 series obsolete before you get your $100 worth of use out of it.

Here's the thing though, if you're willing to spend the whole $500, and the cooling solution is not part of the equation, you can get 2 x 6950's with 2GB VRAM each. You'd have a much larger array, same size buffer but faster video RAM, and you can bring the core clock up to 6870 levels if you're going to be overclocking and water cooling.

When it comes to the AMD side of the fence I usually recommend XFX cards because you can't beat a lifetime warranty. HIS makes a pretty mean Radeon too, but many of their products only have a 2 year warranty.

Quick tip. Don't use Flash acceleration with the Radeons if you want maximum GPU performance. First of all, you don't need it if you have even a half-way decent multi-core CPU. Secondly, it can lock the core speed of the card at 50% under some circumstances because it's not an intensive operation for such a massive, multidimensional array. When work is added to the GPU, a single Flash operation can keep them from spooling up to full speed.