Hyperthreading and Task number Impact Observations

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,537
Credit: 286,150,706
RAC: 106,228

RE: just stumbled over that


It is indeed! Thanks for digging that up. :-)

Those bandwidth curves look eerily familiar. I'll look closer at the article and comment further if appropriate.

FWIW : I have stopped the GPU work within and have tooled up that machine I mentioned earlier with Process Lasso. I am currently benchmarking the virtual cores when used alone for bucket WU's ie. only one WU at a time on the entire machine and I am proceeding through each core ( 5 of 8 done ). I thought I would at first examine the ( entirely reasonable ) assumption that all cores are equivalent one to the next, so that if there is some asymmetry in the hardware it'll come out and I can account for that when I move to testing the ideas I mentioned earlier. They are doing fine thus far with average times overlapping well within each others' one-standard-deviation widths. I'll publish the full spread sheet of data when complete. I'm doing 12 WU's per core, tossing out the highest and the lowest, and using the remaining 10 as 'typical' for statistics. I'm also after the idea of a 'fiducial occasion' or 'single virtual core run-time' for this machine as it is presently configured on bucket WU's, and thus have already collected very many to aggregate to form that. I've also had to admonish my offspring for daring to touch it meantime .... :-)

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

FrankHagen
FrankHagen
Joined: 13 Feb 08
Posts: 102
Credit: 272,200
RAC: 0

RE: It is indeed! Thanks

Quote:


It is indeed! Thanks for digging that up. :-)

Those bandwidth curves look eerily familiar. I'll look closer at the article and comment further if appropriate.

well - actually it's a pretty much ancient thing. P4-area you know, and things might have changed a lot since then.

another thing which comes in mind here: how are cores mapped?

if i look at my i5m i see this:

Coreinfo v2.11 - Dump information on system CPU and memory topology
Copyright (C) 2008-2010 Mark Russinovich
Sysinternals - www.sysinternals.com

Logical to Physical Processor Map:
*-*- Physical Processor 0 (Hyperthreaded)
-*-* Physical Processor 1 (Hyperthreaded)

Logical Processor to Socket Map:
**** Socket 0

Logical Processor to NUMA Node Map:
**** NUMA Node 0

Logical Processor to Cache Map:
*-*- Data Cache 0, Level 1, 32 KB, Assoc 8, LineSize 64
*-*- Instruction Cache 0, Level 1, 32 KB, Assoc 4, LineSize 64
*-*- Unified Cache 0, Level 2, 256 KB, Assoc 8, LineSize 64
-*-* Data Cache 1, Level 1, 32 KB, Assoc 8, LineSize 64
-*-* Instruction Cache 1, Level 1, 32 KB, Assoc 4, LineSize 64
-*-* Unified Cache 1, Level 2, 256 KB, Assoc 8, LineSize 64
**** Unified Cache 2, Level 3, 3 MB, Assoc 12, LineSize 64

so all this testing will need to pick the right cores.

probably the real freak-out will come, if someone shows up with a quad-socket Xeon E7-4800. ;)

archae86
archae86
Joined: 6 Dec 05
Posts: 3,145
Credit: 7,050,044,931
RAC: 1,643,320

RE: I am currently

Quote:
I am currently benchmarking the virtual cores when used alone for bucket WU's ie. only one WU at a time on the entire machine and I am proceeding through each core ( 5 of 8 done ).

While I think it rather likely that the virtual CPUs are in fact equivalent, I was surprised to find on my own host that at least one of the many background programs had an affinity.

If you're ambitious you might check, for example by using Process Explorer and right clicking one process at a time and checking under "set affinity…". Remembering I had seen something in the past I did a little of this just now and noticed that Speedfan is set to run only on CPU number five (of eight on this Westmere host).

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,537
Credit: 286,150,706
RAC: 106,228

RE: RE: I am currently

Quote:
Quote:
I am currently benchmarking the virtual cores when used alone for bucket WU's ie. only one WU at a time on the entire machine and I am proceeding through each core ( 5 of 8 done ).

While I think it rather likely that the virtual CPUs are in fact equivalent, I was surprised to find on my own host that at least one of the many background programs had an affinity.

If you're ambitious you might check, for example by using Process Explorer and right clicking one process at a time and checking under "set affinity…". Remembering I had seen something in the past I did a little of this just now and noticed that Speedfan is set to run only on CPU number five (of eight on this Westmere host).


Thank you indeed! I was basically thinking of possible mild hardware disparity, but yes there may well be OS bindings. This had not occurred to me. An excellent idea and I will check that. :-)

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

FrankHagen
FrankHagen
Joined: 13 Feb 08
Posts: 102
Credit: 272,200
RAC: 0

RE: Thank you indeed! I was

Quote:
Thank you indeed! I was basically thinking of possible mild hardware disparity, but yes there may well be OS bindings. This had not occurred to me. An excellent idea and I will check that. :-)

now that you get rolling.. ;)

get the whole sysinternals suite - things to check here: process-explorer, PSSTART (because it can use the pretty ancient API), COREinfo to tell you what's really under the hood...

i am using marks tools for a long time before the battleship of lawyers showed up and forced them to sign for redmond.

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,537
Credit: 286,150,706
RAC: 106,228

RE: RE: Thank you indeed!

Quote:
Quote:
Thank you indeed! I was basically thinking of possible mild hardware disparity, but yes there may well be OS bindings. This had not occurred to me. An excellent idea and I will check that. :-)

now that you get rolling.. ;)

get the whole sysinternals suite - things to check here: process-explorer, PSSTART (because it can use the pretty ancient API), COREinfo to tell you what's really under the hood...

i am using marks tools for a long time before the battleship of lawyers showed up and forced them to sign for redmond.


Terrific ideas. I will indeed get that to drill down and more cleanly separate the 'pure HT' aspect I seek from the rest. :-)

Aside : I can use LogMeIn - pro version with LogMeIn Ignition - which basically acts as a neat layer over Windows Remote Desktop. It creates a secure VPN connection ( long & very random key ) and thence allows full remote control, file sharing, FTP, clandestine monitoring ( so any user on the target won't obviously know ), chat even, plus other stuff. So here I am in Germany fiddling/tweaking my HT profiling experiments on the DownUnda machine - with my only trouble being the laptop screen here doesn't match the bigger desktop there. So I have to pop my spectacles on. :-)

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

Comment viewing options

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