Athlon 64 3200+ 90nm (S939) - The next P4 2.4 "C" ?

CPU by JNav89GT @ 2004-10-11

Combining affordable pricings, good overclockability, top level performance and the ability to keep things cool is what we enthusiasts pursue in our quests to build high performance computers. Does the new S939 3200+ 0.09? have these attributes? Find out as we compare it to an expensive FX53, a cheaper S754 3400+ and a mid-range S939 3500+ CPU.

  • prev
  • next

Test Setup & Overclocking

Test setup and comparison hardware used

In somewhat of a quiet fashion, which is rather odd given CPU manufacturer’s propensity to thump their chests whenever possible, the 0.09µ Athlon64 has come to market and is now available in limited quantities. Rumor mills were busy predicting heat and yield problems for AMD’s 0.09µ processors, much like the problems of initial Pentium 4 Prescott. To get a better understanding of how these new chips performed without basing my opinion on conjecture and hearsay, I preordered an Athlon64 3200+ socket 939 0.09µ CPU from Monarch Computer. I did so weeks prior to their launch and was afforded the opportunity to get one of the first samples as they were received into retail channels.

The cpu arrived last week, so I thought I’d torture this cpu and run it through a series of benchmarks to see how it stood up to other Athlon64 processors I had in my lab. Also, I wanted to see how the CPU overclocked and if heat output was a concern. Certainly, given the price point of this CPU at $225-$250, the monetary barrier for socket 939 has been lowered into the realm of affordability for many. If this CPU performs on par with its contemporaries we can rejoice! If this processor fails to provide equivalent performance to comparable CPUs in the same price points of other AMD and/or Intel platforms, we would need to consider this CPU somewhat of a disappointment.

Tests were performed on following system configurations.

JNav89GT's Test Setup
CPU 1) Athlon FX53 (S939 - 2.4ghz – 1Mb L2 cache - 0.13µ)
2) Athlon64 3500+ (S939 – 2.2ghz – 512k L2 cache – 0.13µ)
3) Athlon64 3400+ (S754 – 2.2ghz – 1Mb L2 cache – 0.13µ)
4) Athlon64 3200+ (S939 – 2ghz – 512k L2 cache - 0.09µ)
Mainboard S939: MSI K8N Neo2 Platinum
S754: Chaintech ZNF3-250
Memory S939: 2 * OCZ PC3700 Gold 512Mb (CL2.5 3-3-8)
S754: 2 * Winbond BH-5 based PC3500 512Mb (CL2 2-2-10)
Video PNY Geforce 6800GT 256Mb running at 410/1105 (just above Ultra specifications!)
Cooling S939: Stock A64 HSF and Alpha PAL8150 + 40cfm Panaflo fan
S754: ThermalRight XP-120 + Sunon 120mm fan

Extra remark:
  • Windows XP Prof SP2 DX9c (S939) & Windows XP Prof SP1 DX9b (S754) – No OS tweaks.
  • S754 Memory: Testing at 200 MHz HTT used ram speed of DDR400 for PC3200 Speeds. Testing at 250 MHz HTT used ram speed in bios of DDR333, which after HTT is overclocked produces ram speed of 208mhz or DDR416.

    Testing details, Temperatures & Overclocking

    System testing was tailored to provide comparative data in relation to stock and overclocked speeds of the 3200+ 0.09µ processor. CPU multipliers for faster running chips were lowered to 10x to bring operating clock speeds in line for each chip and provide better contrast to performance differences. Benchmarks were performed at 2 GHz and 200 MHz HTT, default clock speed for 3200+ A64 and again at 2.5 GHz and 250 MHz HTT for each of the CPUs.

    The 2.5 GHz level was decided upon as this was about the max stable speed of the 3200+ on stock air-cooling with a retail A64 heatsink. I could go higher, but it wasn’t prime stable with this cooling and I don’t want to mislead people into thinking they will get 2.6+ GHz stable with what I consider to be fairly poor air-cooling.

    Despite this fairly poor air-cooling, temperatures were 37C idle and high 40’s to low 50’s Celsius at load during overclock. My FX-53 CPU with same heatsink and fan ran a steady 5-8°C warmer during all testing.

    Another observation I made was when I physically touched the heatsink while running the 3200+@2.5 GHz and 1.6v vcore actual, it was barely warmer than ambient. The same heatsink with my FX-53 was very warm to touch. This tells me that not only is the CPU running slightly cooler, but is putting off less watts of heat as dissipated by the heatsink.

    The Thermalright XP-120 I have on my Chaintech board would not fit due to clearance issues on my MSI K8N Neo2 board, so I could not use that cooler for the Socket 939 side of testing. Furthermore, I did not come into possession of the Alpha PAL8150T until the last evening of testing, and while my max stable air speeds did improve with the 3200+, my data collection was too far along to change at that point. I will say that 2.55-2.6GHz is about max stable air-cooled speed for my 3200+ at 1.65-1.7v actual core voltages with better air-cooling. The MSI board (and I suspect others) will give 0.1v less than that set in bios, due to a lower Hammer VID (default voltage) of the 0.09µ processor which is 1.4v default. So when I set voltage in bios I had to keep in mind that actual voltages would be 0.1v lower.

    That said, I am very impressed with the overclockability of this processor on air. Sure, 2.55-2.6GHz doesn’t sound that impressive when people talk of 4 GHz overclocked Pentium 4 chips. The discerning enthusiasts though will be quick to point out that the same 4 GHz P4 will in all likelihood get beat down like a crackhead by the LAPD, when it comes to 3D gaming performance. I found out first hand how this could be true when I was running a 4.2ghz P4 Northwood in my Prometia phase change cooler, and my brand new at the time socket 754 3400+@2.5 GHz (early CO stepping chip, not same cpu as one tested on 754 for this review) on the same Prometia phase change could beat it readily in many 3D benchmarks. I will say that while 2.5 GHz on stock air and 2.55-2.6 GHz on good air-cooling are good, I would be hesitant to say it’s mind-blowing. What I do appreciate is that this 3200+ 0.09µ Socket 939 CPU is much cheaper than previous socket 939 chips, seems to overclock to similar levels given equal cooling, and runs cool in the process.


    Benchmarks were centered primarily on gaming. I’ll admit I do not perform much else with my computers, such as video encoding or running 3D CAD programming. I would feel remiss in trying to benchmark those programs with my relative ignorance in those areas. Futuremark's PCMark04 does provide a tool though to measure overall PC performance, although I do find it to favor P4 processors over AMD more so that I believe real world applications might. I will provide SiSoft Sandra benchmarks for Arithmetic, Multimedia, and Memory bandwidth, as well as SuperPi for comparison to other popular cpus. That said, I find gaming to be a commonly agreeable area on which to focus in regards to a processor’s ability.

  • SiSoft Sandra Standard 2004.SP2b
    - CPU Arithmetic Benchmark (MP/MT support)
    - CPU Multi-Media Benchmark (including MMX, MMX Enh, 3DNow!, 3DNow! Enh, SSE(2)) (MP/MT support)
    - Memory Bandwidth Benchmark (MP/MT support)
  • Futuremark PCMark04
  • SuperPi 1M calculation
  • Futuremark 3DMark2001SE Build 330
  • Doom 3 Timedemo demo1 (run at 640x480 LQ and 1280x1024 HQ)
  • UT2003 v2225 benchmarked with [H]ardoOCP's tool (run at 1024x768 HQ)
  • UT2004 benchmarked with Primeval demo (run at 1024x768 HQ)
  • FarCry1.1 benchmarked with HardwareOC Utility (run at 640x480, 1024x768, and 1280x1024 at HQ) Ubisoft Reactor demo and PC Games Hardware demo were chosen.

    Onto the results ->
    • prev
    • next