Tom's Hardware's Three-Part, 3-Way Graphics Scaling Series
Part 1, The Cards: Triple-GPU Scaling: AMD CrossFire Vs. Nvidia SLIPart 2, The Slots: GeForce And Radeon On Intel's P67: PCIe Scaling Explored
Part 3, The Chipsets: P67, X58, And NF200: The Best For CrossFire And SLI
The battle for GPU supremacy is a bit of a waiting game. The recent launch of AMD's Radeon HD 6990 (AMD Radeon HD 6990 4 GB Review: Antilles Makes [Too Much] Noise) and upcoming introduction of Nvidia's GeForce GTX 590 help demonstrate.
Roughly every six moths, two companies with similarly-capable products wait to see what the other one will showcase, hoping to trump that introduction a few days later by tweaking the clocks, power, and configuration of their own product. Both firms use these delays to build anticipation among followers. Though, as that very same 6990 story showed us, being a "fan" of either company is just silly. Why make excuses for one design team's poor execution and then slam the competition for the very same thing?
Yet, in spite of the hysteria, the past few GPU launches have barely budged the performance bar. For those who prefer to play at high resolutions without giving up any of a modern game’s splendor, multi-GPU configurations (and that includes the latest and upcoming dual-GPU cards) are still mandatory.
Launch articles rarely include multi-GPU tests, partly because most reviewers have only one card. The few reviewers that do get multiple cards often find SLI and CrossFire bugs persisting until two or three software revisions after a new card is launched, leading to hopeful comments from loyal enthusiasts like, "just wait until they get their driver situation worked out!"
As a result, when we as reviewers get asked about multi-GPU scaling, it's often really hard to answer definitively. The real data has to be collected once availability is more stable, and after driver problems have been more thoroughly addressed.
It's Time To Explore Multi-Card Scaling
Today’s question isn’t which single-GPU card is best, but which cards operate best in pairs and trios. Though current buyers can now save even more money with the 1 GB version, AMD's Radeon HD 6950 2 GB was its least expensive three-way CrossFire option when we set up this article.
The closest competitor is Nvidia’s GeForce GTX 570 1280 MB, a model that also represents its cheapest-current three-way SLI configuration. Anything less and you only get a single SLI link, support two-way setups.
We've already seen fantastic scaling from three-way SLI, so perhaps the biggest question is whether AMD’s scaling had really improved to the point where it can yank the trophy out of its rival’s hands. We've seen two-way numbers that suggest yes, in fact, Radeons work much better together now than in the past.
Test System Configuration | |
---|---|
CPU | Intel
Core i7-2600K (Sandy Bridge) LGA 1155: 3.40-3.80 GHz, 8 MB L3
Cache Overclocked to 4.00 GHz at 1.25 V |
Motherboard | Asus
P8P67 WS Revolution: LGA 1155, BIOS 1007 (01/24/2011) Intel P67 Express, Nvidia NF200 PCIe Bridge (2 x PCIe 2.0 x16) |
RAM | Kingston KHX2133C9D3T1K2/4GX: 2 x 2 GB, DDR3-2133 at DDR3-1600 CAS 7-7-7-21, 1.60 V |
Hard Drive | Western Digital WD1002FBYS: 1 TB, 7200 RPM, SATA 3Gb/s, 32 MB cache |
GeForce Graphics | Asus ENGTX570/2DI/1280MD5: 742 MHz GF110 GPU, 1280 MB GDDR5-3800 |
Radeon Graphics | XFX HD-695A-CNFC: 800 MHz Cayman GPU, 2 GB GDDR5-5000 |
Sound | Integrated HD Audio |
Network | Integrated Gigabit Networking |
Power | OCZ-Z1000M: 1000 W Modular, ATX12V v2.2, EPS12V, 80 PLUS Gold |
Software | |
OS | Microsoft Windows 7 Ultimate 64-bit |
GeForce Graphics | GeForce/Ion 266.58 |
Radeon Graphics | AMD Catalyst 11.1 |
Chipset | Intel INF 9.2.0.1019 |
CPU “bottlenecks” are always a problem at medium (and lower) resolutions. Today’s tests start at medium resolutions, so we clocked Intel’s Core i7-2600K to 4.0 GHz in an effort to minimize that issue.
The Achilles-heel of Intel’s fastest processor is a platform that supports only sixteen PCIe 2.0 lanes directly from the processor to the graphics card. Asus solves that solution by incorporating Nvidia’s 32-lane NF200 bridge, thereby repeating the same information at full bandwidth to two cards simultaneously. Triple-GPU mode operates in an x8/x8/x16 configuration, as it would with most X58 motherboards.
Power might be the price for performance, but we wanted to see if it also scaled with performance as GPU numbers increased. OCZ’s 80 PLUS Gold-rated Z1000M provides that clean power while adding only around 11% overhead at our watt meter.
Benchmark Configuration | |
---|---|
3D Games | |
Aliens Vs. Predator Benchmark | Alien vs
Predator Benchmark Tool Test Set 1: Highest Settings, No AA Test Set 2: Highest Settings, 4x AA |
Crysis | Patch
1.2.1, DirectX 10, 64-bit executable, benchmark tool Test Set 1: Highest Quality, No AA Test Set 2: Highest Quality, 4x AA |
F1 2010 | v1.01, Run with -benchmark
example_benchmark.xml Test Set 1: Highest Settings, No AA Test Set 2: Highest Settings, 4x AA |
Just Cause 2 | Version 1.0.0.2, Built-In Benchmark "Concrete Jungle" Test Set 1: Highest Details, No AA, 16x AF Test Set 2: Highest Details, 4x AA, 16x AF |
S.T.A.L.K.E.R.: Call Of Pripyat | Call Of Pripyat Benchmark version, all options,
HDAO Test Set 1: Ultra Preset, DX11 EFDL, Ultra SSAO, No AA Test Set 2: Ultra Preset, DX11 EFDL, Ultra SSAO, 4x MSAA |
Synthetic Benchmarks and Settings | |
3DMark 11 | Version: 1.0.1.0, Benchmark Only |
The latest 3DMark version is supposed to remove any Nvidia
bias as a result of artificially-inflated PhysX results, so we see big
problems for AMD if these results are repeated in
games.
Performance for AMD and Nvidia becomes approximate at 3DMark 11’s Extreme preset, and the multi-GPU scaling appears solid with both brands. Let’s see how this works in the real world, where it counts most!
Performance for AMD and Nvidia becomes approximate at 3DMark 11’s Extreme preset, and the multi-GPU scaling appears solid with both brands. Let’s see how this works in the real world, where it counts most!
The first strike against AMD’s three-way scaling comes in
Aliens vs. Predator, where overhead appears to make the triple-card
configuration lose a little performance.
Multi-card scaling usually improves at higher resolutions and details, yet three-way CrossFireX still falls behind CrossFire using a pair of cards. It appears as though this title uses only two of AMD’s cards, and that might be a problem with Catalyst Application Profiles. Excellent scaling realized on Nvidia's three-way SLI setup adds insult to AMD’s injury.
Multi-card scaling usually improves at higher resolutions and details, yet three-way CrossFireX still falls behind CrossFire using a pair of cards. It appears as though this title uses only two of AMD’s cards, and that might be a problem with Catalyst Application Profiles. Excellent scaling realized on Nvidia's three-way SLI setup adds insult to AMD’s injury.
Though we use it primarily as a system-wide test, Crysis
continues to show good performance scaling for different levels of
single-GPU performance. SLI and CrossFire were fixed long ago, but what
about three-way?
A third card appears to be about as appealing as an auxiliary nipple at 1680x1050, but higher resolutions prove the viability of both company's three-way solutions. AMD’s second card scales a little better than Nvidia’s, while Nvidia takes the scaling lead when a trio of boards run side-by-side.
A third card appears to be about as appealing as an auxiliary nipple at 1680x1050, but higher resolutions prove the viability of both company's three-way solutions. AMD’s second card scales a little better than Nvidia’s, while Nvidia takes the scaling lead when a trio of boards run side-by-side.
Tidak ada komentar:
Posting Komentar