facebook rss twitter

Review: CrossFire and SLI dongle-less / bridgeless performance analysis

by Steve Kerrison on 18 September 2006, 09:11

Tags: ATi Technologies (NYSE:AMD)

Quick Link: HEXUS.net/qagq3

Add to My Vault: x

System setup and notes

NVIDIA platformATI platform
CPUIntel Core 2 Extreme X6800 LGA775 (2.93GHz, 4MiB L2 cache, dual-core)
MotherboardNVIDIA NF5 590 SLI Intel Edition Reference boardAsus P5W-DH Deluxe (975X+ICH7R)
Memory2GBytes (2 x 1GByte) Corsair EPP - 4-4-4-12 @ DDR2-800
Graphics cardsNVIDIA GeForce 7900 GT 256MB
(450/1320 - single and SLI, bridged and bridgeless)
NVIDIA GeForce 7950 GX2 1GB (500/1200)HIS X1900GT IceQ3 Dual DL-DVI VIVO 256MB GDDR3
(574/594 - single and dongle-less CrossFire)
HIS Radeon X1900 CrossFire Edition 256MB
(621/1440 - Dongled CrossFire)
ATI Radeon X1950XTX 512MB (648/999)
Graphics BIOS5.
DriverForceWare 91.31Catalyst 6.8
Hard drive160GB Seagate Barracuda 7200.9 (3Gb/s mode)
MonitorDell 2405FPW
OSWindows XP Professional SP2

We've got a nice spread of cards here, including the SKUs we'll be using in CrossFire and SLI. We'll be able to see how much impact removing direct inter-card communication has on SLI and CrossFire, and whether that impact takes performance down too close to that of a single card.

To test dongle-less CrossFire we used a pair of HIS X1900 GTs, but for 'dongled' CrossFire we needed a master card. Enter the HIS X1900 CrossFire edition. This card sports 512MiB of memory and 16 pipelines, but when coupled with a HIS X1900 GT in CrossFire, the card drops its memory usage to 256MiB and turns off four of its pipelines to bring it into line with the spec of the X1900 GT. However, the card's higher clock speeds remain.

Despite our best efforts to clock the card back to GT speeds, no software would make it hold such settings. With our CrossFire edition card in place then, there's both the dongle and the higher clocks to consider, which to be fair, is what would be seen in the real world anyway.

SLI's nice and simple. Two 7900 GTs with a bridge across them, then the same 7900 GTs without a bridge. No clock speed differences... Huzzah.

Testing software

  • Far Cry v1.33
  • Quake 4 v1.4
  • Splinter Cell: Chaos Theory

If benchmark runs counted as gaming time, we'd be the finest gamers on the planet. Alas, they don't. Still, we ran our cards through the familiar trio of gaming performance tests, using 1600x1200 and 1920x1200 as our resolutions of choice. Far Cry and Splinter Cell were run with 4xAA and 8xAF, while Quake 4 was run with 4xAA and 16xAF. We also ran HDR tests in Splinter Cell with 0xAA and 8xAF.


Apart from our CrossFire Master Card clock speed issues, we also experienced less than ideal CrossFire performance in i975X, particularly in Quake 4. It's a combination of driver issues and somewhat weaker OpenGL performance, the former the most in need of fixing, we feel.