|
|
PCMark Vantage on Phenom 9900: No Overclock?
... THE new OC stress benchmark?
By Nebojsa Novakovic: Monday, 31 December 2007, 10:03 PM
I HAVEN'T embraced Windoze Vista much, until a post-SP1 version rolls out - in fact, I don't even have its 32-bit version yet. Therefore, I haven't done too many runs of the brand new, Vista-only, PCMark Vantage.
The tested new Phenom 9900 config on Asus M3A32 MVP Deluxe mobo, CoolerMaster Hyper212 fan and dual Asus EAH3870 TOP cards in Crossfire was my first shot at PCMark Vantage. Including the November 2008 fix, the whole installation takes some 20 minutes.
The benchmark is quite different from the previous PCMark rounds, as it focuses on 'practical angle' of synthetic benchmarks, with heavily multithreaded "gaming", "memories" - not DRAM but family picture memories, "music", " communications" and so on scenarios, straining everything from CPU, chipset and memory to the GPU and even storage. Rather than me repeating their product features and, gotta say, unique pricing system, have a look at http://www.futuremark.com/download/pcmarkvantage/ for more specs and such.
Combined with the added system load that fully 64-bit Vista flavour imposes, it would be an interesting stress test.
As I managed to push this Phenom to run reasonably well, and complete all other 64-bit Vista benchmarks, at 2.8 GHz on multiplier 14 on default FSB200 setting, I was hoping to get the first PCMark Vantage OC Phenom benchmarks around at that speed.
The default 2.6 GHz speed Vantage benchmark completed fine in just under an hour, including all the optional suites - each suite now has its own results set, by the way. And, it completed at a 1.2875 Vcore for that matter - encouraging!
![]()
Done fine at 2.6 GHz...
Then I tried it at the same Cross-Fired 2.8 GHz 1.337 Vcore setting that completed all other 64-bit test runs fine, including CPU intensive multithreaded Cinebench and Povray runs. It went fine... until about 20 minutes into the main PCMark run. It froze at the "communication" portion, a web page rendering window. I waited another 10 minutes just in case, but no, it was done for.
A hard reset was needed to get back up again, I upped to Vcore to 1.35 volts, still the thing stuck at the same place. Pushing the voltage down to 1.325 volts had the test run crash even earlier, at the "music" portion.
Oh boy, what to do? Was it the dual-GPU setup problem? Not likely. Well, I lowered the speed to 2.7 GHz, multiplier 13.5x, and kept the voltage at 1.325 volts. Now, the PCmark portion alone completed fine, including all the web page renders. But... the overall performance index went down! No, not due to the CPU, which gave the extra 3% or so expected. The hard disk seems to have been the culprit.
Also, notice that PCMark Vantage - unlike 3DMark - doesn't seem to make use of CrossFire dual GPU configuration for the gaming tests.
Knowing that the memory and I/O were kept at identical setting all the while, this 64-bit PCMark run seemingly set the absolute stress limit on how far truly stable CPU operation can be pushed on the Phenom 9900. The answer: 2.7 GHz, not 2.8 as I thought yesterday - just 3% above its stock speed.
Take a look at the results...
![]()
As I now run a fresh Vi$ta 64-bit reinstall on the deep-frozen Intel QX9770, we'll see how much of an overclocker stability test the new PCMark Vantage is on the dominant Chipzilla platform.
Based on the Phenom run I saw, it seems to strain the CPU and the system just as much as the Linpack runs, exceeding most other benchmarks - and it shows this Phenom 9900 is pretty much at the limit of the AMD stepping speed possible right now, with hardly any overclocking possible under sustained full load.
Do we have the new ultimate overclocker's stability test run? Methinks so... I'd love Futuremark to do a 32/64-bit XP version too, just in case Vista becomes a "system crash" suspect in some cases. µ |
|
IP卡
狗仔卡
发表于 2008-1-1 00:22
QQ好友和群
收藏
分享
好贴
烂贴
提升卡
置顶卡
沉默卡
喧嚣卡
变色卡
显身卡
楼主
