![]() |
Need Bulldozer owner volunteers
Since Bulldozer is having threading issues that are apparently are fixed on windows 8 I was wondering if anyone that owns or has access to A bulldozer chip could test something out.
Try these tweaks in windows 7!!! First of all... Disable core parking. This was mentioned by Rick last year? There is a non-destructive way of doing it now.. since windows does this (even with high performance set in power themes) it makes sense that sleeping half a module would cause more issues. Quote:
The other thing is a windows service. Note that enabling this feature will prevent windows from going into a sleep state. This one I am not as sure of I played with it a bit but never accurately tested it. By the definition it could improve or impede performance. Go to services, set "Thread Ordering Service" to automatic. AFAIK Only Bz has access to one ATM, I cant afford one so no testing here, and will probably skip it as it will require LN2 to bench. Now for a desktop Llano is my next experiment :) Thanks guys I will post this over at OCF as well. |
Will do mate as soon as possible.
I plan to test it on Linux, Win7 & Win8 for a few apps & games, and then push it hard on LN2. I did a little test with a SB in Win7 & Win8 in various apps & games, and I did notice some improvements in some tests, a lot of draws, and a few decreases. Win8 developers preview is still in a very very alpha stage, very unstable and highly... incompatible, tried like 80 apps & games, with only 30 or so working properly, a few working.... kind of, and a lot of them causing crashes/freezes or not opening at all. I shall be back with more stuff during the weekend |
Oh sory , I meant try the tweaks in Windows 7
I will update OP Thanks man. I think that hte core parking is going to be the big one (no one does it) |
Will do everything ;)
You know how I am, it's all or nothin' :D |
I spent a few hours testing core parking to relise that core parking only works on Intel I3/I5/I7 CPUs, nothing else. The CPU has to be capable of it.
I still tried to do the changes in regedit to test, and there was no difference either way. I did figure out a couple of things while testing that though. With 4M 4C active (by setting afinity, so all 8 "cores" were on), turbo still works on the other cores. With 2M 4C active (same as before, by setting afinity with all 8 "cores" on), turbo does not work on the other cores. 2M 2C share 1 thread 80%/20%, while 1M 2C shares 1 thread 98%/2%. |
Thanks Cecil for testing and the info
Core parking should be enabled on the BD as it works on the Llano/Lynx platforms. At least according to Anand. http://www.anandtech.com/show/2933 Also http://techreport.com/articles.x/19514/3 Although this states power gate is on a per module not a per core basis, which I figured was part of the problem. |
Didnt know Llano added it, havent really looked at Llano.
I did this method to disable core parking, Quote:
And OT a bit, dont try to use an 8 series BD on the 990FXA-UD3. I was running Prime at 3.8 with 1.28V and the NB and VRMs got over 70C, and in a few minutes shut down and killed both the board and the CPU. The heatsink on the vrm melted skin off my knuckles when I touched it. |
LN2 dewar filling pending ( approx 7 hrs of waiting :p ).
Shall have some tests done by tomorrow morning. |
BenchZowner benching for GHz.gr - FX-8150 CPU-z @... LN2
Numbers with the FX-8150 @ Stock ( 3.6GHz, Turbo Core Disabled ) - 4GB DC DDR3-1866 9-9-9-24 - GTX 480 Stock - Ubuntu Linux 11.10 x64 PTS ( Phoronix Test Suite ) test # - result 39 = 858C/s 119 (x264) = 100.12fps 89 = 30.84s 58 = 59.18S/s 75 = 1542.55Mflops 8 = 35.11s 86 = 112.2 115 = 68.2s 65 = 2.962s 50 = 322.43s 12 = 24981.3MB/s 79 = 14658.34MB/s 2 = 17535MB/s 10 = 50.66s CineBench 11.5 x64 = 4.91pts multi /// 7.57pts @ 5619MHz WinRAR = 3421kb/s 3D Mark Vantage CPU Score = 16329 3D Mark11 Physics Score = 5734 http://www.benchzone.com/as_tragoudisoume/screen072.jpg |
http://www.benchzone.com/as_tragoudisoume/ptsbd.png
The Stream benchmark is suffering on the Sandy Bridge because of the low memory frequency ( upping the RAM to 1866 7-7-7-22 nets 19964MB/s ) and the 7-zip test also suffers a performance loss due to the low RAM clock ( 1333 ) against BD's 1866 |
| All times are GMT -10. The time now is 09:19 AM. |
Copyright ©2009 Overclockaholics.com