System setup and notes
Here's a quick rundown of the test system should you wish to compare benchmark results with your own.AMD Athlon 64 FX-53, 2400MHz, S940, 1.5v, DDR400 ECC DC memory (2.5-3-3-5)
AMD Athlon 64 FX-51, 2200MHz, S940, 1.5v, DDR400 ECC DC memory (2.5-3-3-5)
AMD Athlon 64 Model 3400+, 2200MHz, S754, 1.5v, DDR400 SC memory (2.5-2-2-6)
Intel Pentium 4 3.4GHz Northwood Extreme Edition, S478, 1.525v, DDR400 DC memory (2-2-2-6)
Intel Pentium 4 3.4GHz Northwood, S478, 1.525v, DDR400 DC (2-2-2-6)
Intel Pentium 4 3.2GHz Prescott, S478, 1.4v, DDR400 DC (2-2-2-6)
AMD Athlon XP-M 2500+, 1833MHz, S462, 1.45v, DDR333 DC (2-2-2-6)
Other components
ASUS SK8N nForce3 Pro 150 S940 motherboard (Rev 1005 Beta 006)
DFI LANPARTY i865PE S478 motherboard with PAT-like BIOS (19/12/2003 BIOS)
EPoX 8HDA3+ VIA K8T880 S754 motherboard (02/12/03 BIOS)
EPoX 8RDA3G nForce2 Ultra 400 S462 motherboard
2 x 256MB Corsair XMS3500C2 memory used for all non-FX boards
2 x 512MB Legacy Electronics DDR400 ECC memory used for the FX-51 and FX-53
Hercules Radeon 9800XT 256MB (412/730)
IBM 40GB P-ATA hard drive
Toshiba 8x DVD
Samcheer 420w PSU
Dell P991 19" monitor
Zalman 7000Cu S478 / 754 cooler
Software
Windows XP Professional SP1 (32-bit)
Windows XP Professional 64-bit (build 1069)
DirectX9.0b
NVIDIA ForceWare 3.13 drivers
VIA Hyperion 4.51 drivers
Intel 5.01.1002 chipset drivers
ATI CATALYST 4.1 drivers and control panel
Pifast v41 to 10m places
LAME v3.92 MP3 encoding with Razor-Lame 1.15 front-end using U2's Pop album
(607MB)
Gordian Knot - XviD encoding test using first vob of Sleepy Hollow. 1433kbit/s bitrate
Kribi Bench 1.1
ScienceMark 2.0
Realstorm Raytracing benchmark 320x180x32
3DMark 2001SE v330
UT2003 Retail (Build 2225)- HEXUS Custom Low-Detail test
X2: The Threat - Rolling Demo
Comanche 4 benchmark
Call of Duty - HEXUS Custom test
Quake 3 v1.30 HQ
Notes
No problems to report during installation and strest testing. ASUS' SK8N motherboard reported a BIOS load temperature of around 50c, which seems high for an FX CPU. The sample exhibited consistent performance throughout testing, so we're adamant that results weren't coloured by an overheating processor. AMD's FX-53 shipped with 2 x 512MB (1GByte) Legacy Electronics ECC DDR400 RAM. It's debatable how much effect an extra 512MB has over the 2 x 256MB run in all other boards. It also has to be noted that the FX-51 was created as a result of changing the multiplier from 12x to 11x.
A word on upcoming performance. Performance is inextricably linked in with RAM timings. We managed to use 2.5-3-3-5 latencies for FX-5x testing. Had we been able to use our standard 2-2-2-6 settings we reckon that both FXs would have looked that much better. Tighter latencies, ceteris paribus, often afford a 2-5% performance leap. Please bear this important point in mind when comparing results.
The FX-53 will be positioned at the very highest echelons and targeted towards enthusiasts who simply want the fastest CPU going, irrespective of cost. That's why the performance fight is likely to be waged against Intel's suped-up 3.4GHz Northwood Extreme Edition processor. We've also added in an Athlon XP-M 2500+ processor so that the majority of enthusiasts can get a feel for what kind of performance increases top-of-the-line CPUs afford over a known, lesser CPU.
Overclocking
So just how far would it go?. Ryzard mentioned that a recent FX-51 would barely hit FX-53 speeds. Rasing the voltage to an unhealthy 1.75v and praying a bit, 2600MHz made it as far as Windows but fell down soon afterwards. It looks good on CPU-Z shot, but it's of no practical use. 'hammer technology is still has a significant MHz ceiling problem; that's why it's taken this long for AMD to be confident in releasing 2.4GHz parts. We'll explore overcloking in a separate article.
