Today's Hard|Forum Post
Today's Hard|Forum Post
Date:
Thursday , April 28, 2011
Author:
Brent Justice
Editor:
Kyle Bennett
Share:

NVIDIA GeForce 3-Way SLI and Radeon Tri-Fire Review

We've seen what a Radeon HD 6990 can do when paired with a Radeon HD 6970 for "Tri-Fire" performance. Now it is time to find out what three NVIDIA GeForce GTX 580 video cards in 3-Way SLI game like in comparison. We will look at A2A performance comparisons and discuss which setup offers the best gameplay experience.

Introduction

On April 11th we evaluated a unique triple-GPU configuration supported by AMD, which also happened to be price competitive with NVIDIA GeForce GTX 2-Way SLI. In that review we looked at the dual-GPU Radeon HD 6990 paired with a Radeon HD 6970 for triple-GPU "Tri-Fire" CrossFireX performance. This was still a "two card" solution, even though there are three GPUs in play.

Comparing costs, we found that two NVIDIA GeForce GTX 580 video cards came the closest in price to our Tri-Fire configuration. So while we had a two-card to two-card comparison, and a $ for $ comparison, many NVIDIA loyalists cried foul at us comparing NVIDIA's 2-GPUs to AMD's 3-GPUs. Our conclusion came down to the AMD "Tri-Fire" solution offering the best overall performance and value for the price. This comparison was price competitive and made perfect sense for us to test, but not to all our readers.

NVIDIA also supports a triple-GPU solution, but in order to achieve it you must use three separate video cards for 3-Way SLI. Currently you cannot utilize a GeForce GTX 590 plus a GeForce GTX 580 for a "two card" triple-GPU configuration. This means that for 3-Way SLI you will need three separate GeForce GTX 580 video cards. The caveat with that is the price, GeForce GTX 580's are still around $500. That means that three of these video cards will cost you $1500. At that price GTX 580 3-Way SLI is about $475 more expensive than a Radeon HD 6990/6970 configuration. The gap is even wider when you compare three Radeon HD 6970 video cards for "Tri-Fire" which come out to $945, making GTX 580 3-Way SLI $555 more expensive!

Even given this price difference, we know that many of you are curious as to how the gaming performance compares. So the goal of this evaluation is to find out how GeForce GTX 580 3-Way SLI (3 separate cards with 3 GPUs total) compares to Radeon HD 6990/6970 "Tri-Fire" (2 separate cards with 3 GPUs total.) In order to test this, we of course tested at no less than a triple-display setup at 5760x1200, anything less would be irresponsible on our part to show you meaningful performance.

Hopefully this review will satisfy many of our readers that like to think we purposefully put NVIDIA at an unfair disadvantage.

Test Setup

Article Image

We are using the exact same test setup as we used in our AMD Radeon 6990/6970 CrossFireX / "Tri-Fire" Review. The exact same driver versions are being used, which is the latest Catalyst 11.4 Preview driver and CAP. We are using ForceWare 270.51 Beta for the NVIDIA video cards. We that ForceWare 270.61 WHQL was recently released, and we installed this driver and found no performance differences compared to 270.51 Beta. There are no differences that affect us between the two driver versions, so we stuck to the same driver used in the previous article so we could make comparisons.

Note that the Radeon HD 6990 is running in its default stock BIOS operation, meaning the GPUs are running at 830MHz. Therefore the Radeon HD 6970 is also running at 830MHz, and the subsequent down-clocked memory speeds along with he Radeon HD 6990. So in reality, three separate Radeon HD 6970 "Tri-Fire" video cards will actually be even faster than what we have here, but as you will see it's not really needed to compete.

All tests are done in apples-to-apples configuration at the exact same settings. Please note that the resolution and AA settings used were not a bottleneck to any setup here. We made sure we were not running into VRAM capacity limitations at these settings, so performance is being correctly compared and we are not hitting any VRAM walls at these tested settings.

Please click on to the next page for all the graphs with discussion.