All things Linux

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33817494324
RAC: 37825107

That temp spread is totally

That temp spread is totally normal and reasonable. Your temps seem fine. 
 

don’t worry about it. 

_________________________________________________________________________

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4699
Credit: 17542325049
RAC: 6372027

Those temps are fine.  Don't

Those temps are fine.  Don't forget that the OS moves/schedules work around in the cpu.  The microcode in the cpu itself moves processes around the CCX's and cores to keep the heat load uniform across the die according to its power/heat budget.

So the heat load on the dies continuously varies and the temp sensors in the CCX complexes are sampled at different times and rates.

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 5585
Credit: 7672972902
RAC: 1745116

So what is the "simple

So what is the "simple patch"?

TechRadar: Linux running slowly? This one simple tweak could fix it for you. https://www.techradar.com/news/this-simple-amd-fix-can-make-linux-much-faster/

It is not clear but I think kernel update after it is made available will be the only way "we" will see it. Which means it is not a "hotfix" away.

Tom M

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)

Wedge009
Wedge009
Joined: 5 Mar 05
Posts: 117
Credit: 15649067190
RAC: 7418678
Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33817494324
RAC: 37825107

need to wait for Linux kernel

need to wait for Linux kernel 6.0+ for the fix.

 

the information so far seems pretty vague about which types of systems are affected. lots of talk about chipsets, but what about AMD systems without a chipset, like AMD EPYC? are those affected as well? it seems to also only affect certain workloads. I'm not really seeing any degradation on AMD performance for any BOINC projects, so I will hazard a guess that it's not affecting most people in this space.

_________________________________________________________________________

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4699
Credit: 17542325049
RAC: 6372027

The I/O die in Zen, Epyc and

The I/O die in Zen, Epyc and TR IS the chipset.

They are affected by this old code bug.

 

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33817494324
RAC: 37825107

It’s not really a chipset.

It’s not really a chipset. Ryzen zen chips have an I/O die as well + an actual chipset. Epyc lacks this “+”. 
 

still, any evidence that this even matters for any BOINC project outside of the one obscure benchmark listed in the article? The whole thing seems blown out of proportion. I haven’t seen ANY complaints about Linux AMD performance recently, and as referenced by the leaderboards of most BOINC projects, AMD is actually dominating Intel for non-AVX512 projects. So it can’t be hurting AMD to the extent suggested in real world use. 
 

just don’t get your hopes up that Linux kernel 6+ will bring any significant performance improvement s because of this for BOINC. 

_________________________________________________________________________

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4699
Credit: 17542325049
RAC: 6372027

The historical purpose of the

The historical purpose of the motherboard chipset was for handling memory handshaking predominately.

Today's cpu architecture has the memory controller inside the die itself for both Intel and AMD.

Now the main purpose of the I/O die in the Zen architecture is for I/O operations as its name suggests.  But it does have silicon dedicated to handshaking with main memory. The chipset on modern AMD boards mainly is for USB and storage control. Nothing to do with memory.

We will see shortly whether the code fix greatly improves the performance of modern AMD cpus next week as I am positive that Michael at Phoronix will have done a complete benchmark suite on his stable of cpus after he configures the tests with the latest 6.0 kernel with the bugfix included.

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 5585
Credit: 7672972902
RAC: 1745116

I just found this quote. It

I just found this quote. It seems to make it clearer that there will not be high computation gains from that AMD patch.

"...Most notable among them could be a patch that prevents a nearly two-decade slowdown for AMD chips, based on workaround code for power management in the early 2000s that hung around for far too long. Intel's Dave Hansen wrote the patch that made it into 6.0, noting in a comment on an Ars post that the issue had become an expensive drain as AMD systems gained higher CPU core counts. The average desktop user won't see huge gains, but larger systems working on intensive input/output applications should benefit....'

From ( now I can't find it) but it was something from one of my "usual suspects" in my news feed I get on my cellphone.

Tom M

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4699
Credit: 17542325049
RAC: 6372027

Well I think we then qualify

Well I think we then qualify as "high intensive input/output" systems as all my hosts have a constantly flashing hard drive access light becaus of the massive disk read/writes from all my current projects.

 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.