New tool for undervolt/overclock AMD K8L and K10 processors

[blackshard]

Thank you for the script!!!

I am running Linux Base-1 2.6.35-22-generic #33-Ubuntu SMP Sun Sep 19 20:32:27 UTC 2010 x86_64 GNU/Linux

Turion Power States Optimization and Control - by blackshard - v0.29.1 (alpha)
Detected CPU:
Family: 0xf Model: 0x5 Stepping: 0x2
Extended Family: 0x10 Extended Model: 0x5
Package Type: 0x1 BrandId: 0x3a86
Detected Physical Cores: 4
Detected processor: AMD Family 10h Processor
Processor has 4 cores
Processor has 5 p-states

Processor Tcl Temperature register: 0
Processor temperature slew rate:slewing disabled

Any suggestions for the 0 Temp?

Nope for now. But next week I will probably get my hands on a Phenom II 1055T and will debug the code (and probably release a newer version with bugfixes).

[blackshard]

Sorry for the ignorance, but it does support the Athlon Neo MV-40 preocessor? In case if not, it will suppor it?

Thanks in advance.

Actually it is not supported because it is a K8 processor, and those are not supported by my program. I don't exclude in future some sort of support, but if I don't have access to such hardware I can't do develop anything

[kbios]

Most probably it is reported correctly, but there is an issue with AM2+ CPUs on AM2 boards: since there is just one single power plane, the processor and the northbridge (i.e. the memory controller) share the power source. So the processor is feed with northbridge voltage, which is usually higher than processor voltage, so those VID numbers means just nothing in such situation. That's the reason why AMD raise the TDP of AM2+ CPUs when put in an AM2 boards.

You may try to change those VIDs but shouldn't happen anything at all. To be precise, I guess that because I never had the chance to experiment it by myself, but just read it on AMD documentation.

At the moment the program can't change the Northbridge VID on K10 processors because I have to test and experiment by myself since the matter is a bit delicate.

You're right: changing the VID doesn't actually affect processor performance. The only way I have found to change it is to use a tool that changes northbridge vid, so the two voltages seem correlated indeed.
Thanks for your clarification

[blackshard]

About the temperature problem, I tried on a new and shining 1055T on Windows x64 and it has been reported correctly.
I wanted to boot from a live cd with fedora to double check and suddenly the motherboard died... arghhh!!!

[ntt2010]

Hi Blackshard

There is this program for setting the voltage of AMD K8 processors. It works well on my Athlon 64 X2 5200+ CPU, but the project seems idle for
nearly 2 years already. And it's a bit user unfriendly.

Could you take a look at it, and possibly and combine the two? it'd be great.

http://cpupowerd.sourceforge.net/

[blackshard]

Hi Blackshard

There is this program for setting the voltage of AMD K8 processors. It works well on my Athlon 64 X2 5200+ CPU, but the project seems idle for
nearly 2 years already. And it's a bit user unfriendly.

Could you take a look at it, and possibly and combine the two? it'd be great.

http://cpupowerd.sourceforge.net/

Hi ntt2010. cpupowerd looks like a program that takes care about cpu frequency and voltage scaling, not about frequency and voltage tweaking like tpc do. Actually this job is already done by cpufreq utilities, afaik.
Anyway it could be interesting to take a look to the code to learn K8 usage, but at the moment, since I have no K8 processor to develop on, I can't produce anything valuable.

[charliehorse55]

Hi,

I am trying to use your software but I run the compile script I get this:

Griffin.cpp: In member function ‘virtual void Griffin::perfCounterGetValue(int, int)’:
Griffin.cpp:1122: warning: format ‘%lld’ expects type ‘long long int’, but argument 4 has type ‘uint64_t’
Griffin.cpp: In member function ‘virtual void Griffin::perfMonitorCPUUsage()’:
Griffin.cpp:1149: warning: spurious trailing ‘%’ in format
Griffin.cpp: In member function ‘virtual void Griffin::perfCounterMonitor(int, int)’:
Griffin.cpp:1168: warning: format ‘%d’ expects type ‘int’, but argument 2 has type ‘long unsigned int’
Griffin.cpp:1192: warning: format ‘%llu’ expects type ‘long long unsigned int’, but argument 4 has type ‘uint64_t’
Griffin.cpp:1192: warning: format ‘%llu’ expects type ‘long long unsigned int’, but argument 5 has type ‘uint64_t’
K10Processor.cpp: In member function ‘virtual void K10Processor::perfCounterGetValue(int, int)’:
K10Processor.cpp:1237: warning: format ‘%lld’ expects type ‘long long int’, but argument 4 has type ‘uint64_t’
K10Processor.cpp: In member function ‘virtual void K10Processor::perfMonitorCPUUsage()’:
K10Processor.cpp:1263: warning: spurious trailing ‘%’ in format
K10Processor.cpp: In member function ‘virtual void K10Processor::perfCounterMonitor(int, int)’:
K10Processor.cpp:1282: warning: format ‘%d’ expects type ‘int’, but argument 2 has type ‘long unsigned int’
K10Processor.cpp:1306: warning: format ‘%llu’ expects type ‘long long unsigned int’, but argument 4 has type ‘uint64_t’
K10Processor.cpp:1306: warning: format ‘%llu’ expects type ‘long long unsigned int’, but argument 5 has type ‘uint64_t’

I downloaded version 0.29.1 from this thread.

I'm trying to control the voltage/frequency for my 7750 which is in an AM2+/AM2 Board. Kernel is 2.6.32-25-generic

Ubuntu 10.04 64 bit

[charliehorse55]

These silly forums don't let you edit your posts

anyways here is what I would have added to post above:

EDIT: I have OC'd my CPU to 3.1 GHz using a multi bump, and Cool N' Quiet is working as the CPU will scale back to 1.35 GHz when idle. However I'd like to scale back that frequency even more, as well as reducing the voltage further.

[blackshard]

Hi. Look that those are just warnings. Most probably the program has been compiled. Look into bin folder and probably you will find the compiled program.

Are you using ubuntu 9.04?

[charliehorse55]

I did manage to get the app working. I have a new question, how can I force the processor to stay in a lower p-state even when under load? I want to stress test my lower p-states with prime95 to make sure they are stable, but when I run prime95 it just switches to p-0. I also want to demonstrate how much power power the CPU uses when idling at 3.1 GHz (1.425v) vs. 0.6 GHz (0.900v).