Friday, 20 August 2010 13:14

Phenom II X6 in the lab - 3 Overclocking & Undervoltage

Written by Eliot Kucharik

amd_phenom_x6_front_small recommended08_75 
Review: Affordable Hexa-Cores


First the sad story. Our 1090T does not like overclocking and during the process we scraped three boards. That's also the reason why we could not put on this review much earlier. AMD will send us a replacement but it seems the new CPUs are in such high demand, they can't ship enough of them. So be warned, even if you do not overvoltage other components such as HT or memory controller, even the VCore limit which is reported in the BIOS may cause harm to your board. But we did not expect the CPU to die at "just" 1.5500VCore. The 1055T fared much better, with just 1.4350V which will results in about 1.48VCore reported by CPUz and Everest under load, we managed 3850MHz without a hassle with FSB 275. We did not try to go any higher, because we did not like to scrap another board. Please notice we already got a 95W 1055T, with the currently selling 125W version you may need higher voltages.





Of course it is possible to reduce the VCore to save some energy. Depending on the board you can go down from the 1.3000V of the 1095T to 1.2000V to 1.2250V. As long as Turbo-Core is active, during Turbo the VCore will raise to 1.4750V regardless of the VCore setting and the savings are somewhat limited. If you disable Turbo-Core you can manage to get even further down. We did 1.1750V in the BIOS. Everest still shows about 1.2000VCore, but that's still impressive when six cores are at work.


(Page 3 of 6)
Last modified on Friday, 20 August 2010 12:30
blog comments powered by Disqus


+7 #1 anno 2010-08-20 20:42
Excellent review. :) It was very good to see the system cost and cost per performance graphs, when so many other review site ignore that aspect all together. And that while it's one of the key factors in deciding what to buy.

The conclusion was to be expected, and is in all honesty already fairly well known.

In some cases additional cores benefit games too though:

This will likely hold more and more true in the future - at least, when the graphics card isn't a bottleneck.
+6 #2 blandead 2010-08-20 21:06
"Would we use later software version of the x264 codec AMD would have beaten an Intel i7-975, but as you know to retest every platfrom and CPU takes lots time."

I hope people didn't overlook this comment. Just goes to show you when more programs are optimized to use all 6-cores, how much better these processors can be even compared to the i7-975
+9 #3 Bl0bb3r 2010-08-20 22:17
Nope blandead, it's not just about the programs themselves but the compiler.

This just shows that when using GNU C Compiler, both AMD and Intel platforms can perform good. But when using some intel compiler, it will actually reduce the performance of the competition. And guess what, most benchmarks are compiled exactly in that compiler.
+2 #4 blandead 2010-08-21 20:28
I'm sure there are other factors, but nonetheless if a program is optimized to use all 6 cores it brings huge improvements to these chips over any quad core. what i said was not wrong, if you wanna add more info go ahead, but my logic holds true.
0 #5 Peter Ong 2010-08-22 04:36
Does database server, Java EE apps benefited from 6 cores? Thanks in advance for anyone casting some light here.
+1 #6 blandead 2010-08-22 06:22
database server would surely benefit I do not know about Java EE
0 #7 Jaberwocky 2010-08-22 10:30
Err Chaps.It's all in the cache.I run a 1055T o/clocked to 3.4 Ghz.It's the 6MB cache that is limiting.A baseline for Rendering a scene with just 2 core enabled works out at 100%, so if you enable all 6 cores it should speed the rendering to 300%.IE 3X as fast.Wrong.It actually renders at around 230%.Implying that once the 6MB cache is distributed amongst all 6 cores ,it becomes the bottleneck.If AMD could work on this and double the amount of on board cache then they would have a performer on their hands.
+5 #8 anno 2010-08-22 12:09
Jaberwocky, it could be that you're right, but that's not necessarily true. There are more resources shared between the cores, for instance the memory interface and northbridge and hypertransport switch. Even with double the cache you will never see 100% scaling in chips, that's practically impossible. You don't see it with Gulftown either, which by the way has less cache per core too, seeing as Intel uses inclusive caches.

Of course I'm sure you understand that Thuban is already quite large and increasing its size would reduce yields, and so making it more expensive than you'd expect. It's all trade offs. :)
-1 #9 Bl0bb3r 2010-08-22 15:36
Quoting blandead:
I'm sure there are other factors, but nonetheless if a program is optimized to use all 6 cores it brings huge improvements to these chips over any quad core. what i said was not wrong, if you wanna add more info go ahead, but my logic holds true.

Yes, in theory.. but in reality things are quite different. You can optimize all you want, a rigged compiler will screw with those optimizations anyway.

To be able to post comments please log-in with Disqus


Facebook activity

Latest Commented Articles

Recent Comments