AMD Introduces Dynamic Local Mode for Threadripper: up to 47% Performance Gain

[Deleted member 66701]

Just to add to the questions will it work with 1st gen?

There aren't any 1st gen TR processors that have dies not directly connected to the memory, so it's pointless for 1st gen.

 

[Vince]

There aren't any 1st gen TR processors that don't have dies not directly connected.

you make a good point, my question is though I run distributed 100% of the time. Will they lock some memory to cores to allow "local" for running applications if that makes sense. Does that even make sense? I guess what I am saying is will they allow application specific use of numa/uma depending on what I am running.

 

[Deleted member 66701]

My understanding is that dynamic local mode will only be exposed if running a wx cpu, so can't be used on x cpus.

 

[Vince]

My understanding is that dynamic local mode will only be exposed if running a wx cpu, so can't be used on x cpus.

My understanding as well from the marketing material - I would quite like the advantages of it on my 1950x though Although not quite the advantage it is for the WX series it can still make a difference on our chips... can't it? You know that extra 1 frame a second or something, half a second saved here and there or whatever. Yea I know, clutching at straws now

 

[Panos]

Just to add to the questions will it work with 1st gen?

Probably yes. On the original blog post, there is no indication is restricted 2nd gen.

https://community.amd.com/community...e-amd-ryzen-threadripper-wx-series-processors

I mean even "gaming mode" is allowed on Ryzen 3/5/7 even if is designed for TR cpus.

We shall see.

 

[Deleted member 66701]

My understanding as well from the marketing material - I would quite like the advantages of it on my 1950x though Although not quite the advantage it is for the WX series it can still make a difference on our chips... can't it? You know that extra 1 frame a second or something, half a second saved here and there or whatever. Yea I know, clutching at straws now

NUMA vs UMA used to show a difference when I built my 1950x system in Aug 2017 but since upgrading to the 2950x and running tests again, I can't see a performance difference - I guess the windows scheduler/drivers and/or games have been patched to behave better.

The reason why local mode is working better on the wx series is just because two dies don't have direct access to memory - it's not a problem shared with the x series cpus.

 

[Vince]

NUMA vs UMA used to show a difference when I built my 1950x system in Aug 2017 but since upgrading to the 2950x and running tests again, I can't see a performance difference - I guess the windows scheduler/drivers and/or games have been patched to behave better.

The reason why local mode is working better on the wx series is just because two dies don't have direct access to memory - it's not a problem shared with the x series cpus.

Is it that long ago? i got mine on release or there abouts, tbh I never even tested local, just slapped it together and used the damn thing.

 

[smogsy]

NUMA vs UMA used to show a difference when I built my 1950x system in Aug 2017 but since upgrading to the 2950x and running tests again, I can't see a performance difference - I guess the windows scheduler/drivers and/or games have been patched to behave better.

The reason why local mode is working better on the wx series is just because two dies don't have direct access to memory - it's not a problem shared with the x series cpus.

very true,

but it could add benefit. If it could put the memory /GPU on to the same CCX as where a games running in theory getting local latency on "dynamic mode"

windows could use core 0-4 on CCX 0 but the memory is on CCX1 as is GPU0

 

[DarrenM343]

Mine shows a couple of FPS difference when running benchmarks (creator vs game modes). I usually just leave it in Creator Mode as I can't be bothered to reboot for games to get a few FPS back.