facebook rss twitter

Review: AMD's '4x4' Quad-FX platform unveiled and benchmarked

by James Morris on 30 November 2006, 07:16

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qahfu

Add to My Vault: x

System setup and testing notes



System AMD Quad FX system AMD Athlon 64 FX system Intel Core 2 system
Processors 2 x AMD Athlon 64 FX-74 (3.0GHz, 2MiB L2 cache, Socket F (1207) AMD Athlon 64 FX-62 (2.8GHz, 2MiB L2 cache, AM2) Intel Core 2 Extreme QX6700 (2.67GHz, 8MiB L2 cache, LGA775)
Intel Core 2 Extreme X6800 (2.93GHz, 4MiB L2 cache, LGA775)
Motherboard ASUS L1N64-SLIWS (nForce 680a SLI) Foxconn C51XEM2AA (nForce 590 SLI) Intel D975XBX2 Bad Axe 2 rev. 303 (Intel i975X)
Memory 4GiB (4 x 1024) Corsair Dominator EPP 8888 2GiB (2 x 1024) Patriot PC8000 XBLK 2GiB (2 x 1024) Corsair PC8500
Memory timings and speed 4-4-4-12 @ 753.36MHz* 4-4-4-12 @ 800MHz (PC6400)
Graphics card(s) Sapphire Radeon X1900 XTX 512MiB
Disk drive(s) 2 x WD Raptor 150GB SATAI (WD1500ADFD) in RAID-0
WD Caviar SE 500GB SATAII (WD5000KS)
Seagate 160GB SATAII (ST3160812AS)
Optical drive(s) Sony AW-Q170A Sony DW-Q30A
BIOS revision 0117 (25/10/06) 612W1P14 14/08/06
Mainboard software NVIDIA nForce Package 9.35 NVIDIA nForce Package 9.34 Intel Inf 8.0.1.1002
Graphics driver CATALYST 6.10 BETA
Operating System Windows XP Pro SP2 32-bit SP2
PSU PC Power and Cooling Turbo-Cool 1000W Dark Power PRO 470W OCZ PowerStream 520W
Monitor Dell 2405FPW


* - the memory speed is derived from a set divisor of the CPU's clockspeed. In this case, the closest to DDR2-800 without going above that frequency is /8.

We ran through a comprehensive set of benchmarks and utilities on the finest competition from Intel and AMD, including Intel's Core 2 Extreme QX6700 and Core 2 Extreme X6800, and AMD's Athlon 64 FX-62. The benchmarks included:

2D Benchmarks Sandra Lite 2k7 SP1 Build 2007.8.10.105 Float Buffered Memory Bandwidth Sciencemark 2.0 32-bit Build 21MAR05 Memory Latency
HEXUS Pifast
HEXUS Cryptography
Realstorm Raytracing 2004 - 512x384 No AA
LAME multi-threaded benchmark - 701.5MB file - encoded into 128kbps stereo
DivX 6.4 (existing DV-avi source-file, Home-theatre profile, 1700Kbps, Insane-quality video, 40Kbps, Stereo, 16KHz Audio) - encoded with and without enhanced multi-threading option
Cinebench R9.5
Pov-ray 32-bit 3.7.0 BETA 16 - Use internal benchmark mode

3D Benchmarks FarCry 1.33 (low-end script (1024x768)
Quake4 1.3 SMP Support Enabled - low-end script (1024x768) - demo001 recorded by HEXUS. Quake 4 also run at 1920x1200 4xAA 16xAF
Splinter Cell: Chaos Theory 1.05 (low-end script (1024x768)


Notes

As always, we ran each benchmark a trio of times and averaged the results for the reported score. If any of the three results looked erroneous, we threw all three away until we could collect three within a margin of statistical error. We report any major attempts needed to get three reliable results, of course. Apart from that, things are as noted on the graphs and in the graph commentary. Want to know more? Head for the HEXUS.community.

BIOS beware!

Before delving into our test setup and benchmark scores, however, we should point out that the ASUS L1N64-SLIWS mainboard sent to us by AMD in this NPRP reference system was not exactly final production. In particular, AMD told us that the current 0117 BIOS has a bug which affects synthetic memory benchmarks. The NUMA system is also foreshortened by problems with the BIOS's Static Resource Affinity Tables. This tells the operating system which memory blocks are associated with which processor, so processes load data into local memory, not memory attached to the other processor. We also found graphics performance very inconsistent with the SLI configuration supplied by AMD, and this wasn't directly comparable to our other single-card test setups anyway. So we switched to the same ATi Radeon X1900XTX as we currently use in our standard test setups.