Threadripper first look

[Scotty]

Only recenlty have I been hitting the wall (albeit softly) and I personally can't manage much more complexity in terms of track count. I am thinking that going to either a threadripper or new intel chip and adding another quad UAD-2 card to the rig should give me the working room... It is certainly doable now but a little like crowding a bunch of kids around a small kitchen table. Everyone eats but the elbows get in the way.

Regarding running VEP... I already run satellite computers that are available on my rig connected via adat pairs... Two Oasys PCI servers and another rig running Scope DSP (gets less use these days). So running a farm isn't all that desireable to me. But if I get into heavier orchestral stuff that is where I'll go-- like the others.

Thanks for the input... appreciated. - Scotty

If you're hammering a system so hard that you need to freeze or bounce anything, and expecting a solution anytime soon, then you're in for heartache and misery. It won't be here for the foreseeable future, the tech's way off.

What you need to do is slave one or two systems running VEPro and farming resources. That's the way people get a working DAW with everything loaded and ready to the touch. No faffing with freeing resources because you want one more plug instance.

[Debutante]

...I'd also really be interested in seeing ssome multi-cpu benchmarks. Do you guys ever bother with that? ...Kaine et al? what, with AMDs new "NUMA friendly bandwidth" and all...

[Kaine]

Last multi CPU test round up I did had some V3 Xeons a few years back and the price to performance metrics were lousy enough that I largely stopped working with them at the time. The suggested "NUMA workaround" that the AMD creator's mode offers still gives little to no improvement for Threadripper when dealing with tight ASIO buffers makes me think that the situation won't have improved one iota.

End of the day it's pretty hard to add a workaround what is essentially a physical world problem without bending the rules of time and space... until they build a Tardis powered CPU I guess.

[Scotty]

Ok this is quite interesting. This guy is testing Ryzen with very low latency ram timing with 3600 mhz ram and getting a very sigificant boost in gaming performance. The video is worth watching.. Tons of information ... no nonsense style.

I am wonderng if anyone has had a chance to test DAW performance with this particular setup to see if any bottlenecks in terms of actual asio performance are alleviated using this specific configuration.

https://www.youtube.com/watch?v=S6yp7Pi39Z8

[nIGhT-SoN]

Ryzen loves higher clocked ram more than anything. Intel doesn't scale like that. There is no benefit over 2400Mhz on Intel cpu, but on Ryzen it unleashes it's power. That happens because of infinity fabric and from my understanding the two quad core dies, in case of Ryzen 7, communicates faster between them if you have higher clocked ram. It's the same story with Threadripper but I think it has 4 quad core dies so better ram it would help even more.

Also it's a quite new platform and it will take some time for developers to adapt their software for AMD too. We all know that in music production all software is optimized for Intel mostly, but now Ryzen got a nice piece of the market, there are many people buying. I bought a 1700 and haven't managed to max it out at stock clock.

[Scotty]

If you watch the video all the way through the Ryzen 7 with low latency memory timings on this specific memory kit is beating out a 5ghz intel in many gaming benchmarks. If the performance for asio is improved below 192 sample latency it will get very interesting. Can't wait to see someone test this configuration using daw bench.

[Kaine]

I've tested 3600 ram kits before after someone else pointed this out a while back, didn't make any difference at all but then RAM speed has never had an impact on buffer latency in any prior testing we've done.

[Scotty]

Thanks Kaine... and to keep things interesting the new 6 core INtel i7s Coffee Lake were released today and getting some impressive numbers with 5ghz clocking speed...

I've tested 3600 ram kits before after someone else pointed this out a while back, didn't make any difference at all but then RAM speed has never had an impact on buffer latency in any prior testing we've done.

[nIGhT-SoN]

True, but you have to dellid it, otherwise it can't go pass 4.7 and at that speed will be really hot. But still, quite impressive.

[Scotty]

He is using a very recent kit and doing extremely tight timings (he indicates earlier memory kits couldn't come close to what he is gettng here) and the benchmark improvements are stunning for gaming at least.

I've tested 3600 ram kits before after someone else pointed this out a while back, didn't make any difference at all but then RAM speed has never had an impact on buffer latency in any prior testing we've done.

[incubus]

Last multi CPU test round up I did had some V3 Xeons a few years back and the price to performance metrics were lousy enough that I largely stopped working with them at the time. The suggested "NUMA workaround" that the AMD creator's mode offers still gives little to no improvement for Threadripper when dealing with tight ASIO buffers makes me think that the situation won't have improved one iota.

End of the day it's pretty hard to add a workaround what is essentially a physical world problem without bending the rules of time and space... until they build a Tardis powered CPU I guess.

Yep, it's a bummer. Frankly, I'm staying put with ye ol' 4770k and slight OC'ing. And if I upgrade, it will be intel.