AMD's Ryzen kicks serious a**

[craigtumps]

it is AMD, you will get to have bizarre problems with various software that no one will ever be able to quite troubleshoot properly.

I second that, been burned by AMD too many times. They can keep their fancy benchmarks, never again for me!

[sisarii]

Thanks for your reply! I actually just ordered a Ryzen 1700X + other parts to build a new machine later next week to run Bitwig. I will report too, how it runs. Can you elaborate more about the 40 tracks? Were they audio or virtual instruments with fx plugins, or fx/mastering plugins only at the master channel, what audio interface and buffer size used?

Very interested to hear your observations in heavy VI projects with multiple tracks. (Over 30 at least)

[F5DX]

PCs have always had problems with compatibility of parts etc. The Ryzen is a whole new processor with new chipsets and motherboards, nothing in common with previous AMD offerings, so we don't know yet how it performs in real life. Let's keep this thread about the new Ryzen processor and how it performs with the Bitwig. I will report once I get the parts and time to build the machine, could take 2 weeks or so. Based on some audio benchmarks posted so far, the Ryzen seems to offer good processing power with plugins when using reasonable buffer sizes (128 or above), exactly what I am looking for. I have been using 512 / 256 samples of buffer for many years, and find 256 samples optimal for my use. With this buffer size, the Ryzen seems to perform just fine (based on some benchmarks).

It must be emphasized that building a new Ryzen computer at the moment requires a little bit more effort than going with something that has been available for some time on the market. One must read carefully the compatibility information about ram etc. from manufacturer websites before ordering parts. Also, plenty of Bios updates + Windows updates are expected during the following weeks to get the performance to the normal level. Based on some reports, even Win 10 does not use the cores of the Ryzen properly, and needs to be updated. I went with Asus Prime X370-Pro motherboard with 2x 8 gig Corsair LXP DDR4 3000 MHz memory. According to Corsair website, that is the highest DDR4 memory speed supported by Corsair with AM4 motherboards having 2 dimms installed. The max. speed supported is only 2400 MHz with 4 dimms installed. The memory speed has a larger effect on speed with Ryzen than with Intel processors.

[arrakeen]

In all fairness, we don't even utilize our current cpu's to their full potential like video processing does.

You'll never see your cpu go 100% or even close when doing audio production, so it doesn't even matter.

[jonljacobi]

Thanks for your reply! I actually just ordered a Ryzen 1700X + other parts to build a new machine later next week to run Bitwig. I will report too, how it runs. Can you elaborate more about the 40 tracks? Were they audio or virtual instruments with fx plugins, or fx/mastering plugins only at the master channel, what audio interface and buffer size used?

All VSTs, no audio. Some plug-ins on every channel, but more on the master channel.

I'm going to do a more serious test tomorrow and will save the song for future use. the only area where we've noticed inconsistencies with Ryzen is with lower-resolution gaming where the CPU gets involved. It's as much as 20% slower than Intel's chips. This is very odd as high-resolution gaming with heavy-duty graphics cards is just fine. Everyone is trying to puzzle this out as no one has ever seen it before.

This much I can say. It doesn't draw much juice, and it performs just fine with everything I care about. I would like a few more PCIe lanes though...

[jonljacobi]

BTW, this is with the Asus X370-Pro motherboard and 32GB of memory.

[jonljacobi]

it is AMD, you will get to have bizarre problems with various software that no one will ever be able to quite troubleshoot properly.

I second that, been burned by AMD too many times. They can keep their fancy benchmarks, never again for me!

You know, the Athlon was a great CPU for its time, and most of the weird shit was from the chipsets. I used Athlons for a good six years with no issues.

And without AMD, you would probably only now be seeing 64-bit computing on the desktop. Okay, a bit of an exaggeration, but it was well, well into the future in Intel's and Microsoft's timeline when AMD forced the issue.

BTW, I would like to clarify the topic title. I never meant to imply it was faster than Intel's stuff. However, part of the kick a** is the price and low power draw.

[dkasak]

I've always bought AMD chips when possible. My home server ( AMD A8-5600K APU with Radeon(tm) HD Graphics ) runs 8 virtual machines, 12 database servers ( I'm a data engineer ), hosts 3 domains ( personal, charity and friend's small business ), runs a motion-detection security system, has the boinc distributed computing client using all spare CPU cycles, and *never* skips a beat. The Intel fanboii FUD "random compatibility issues" is total BS. AMD won the race to 1Ghz, gave us the AMD64 architecture, and now have very competitive chips at the high end. Bring it on

[malifica]

I'm going to do a more serious test tomorrow and will save the song for future use. the only area where we've noticed inconsistencies with Ryzen is with lower-resolution gaming where the CPU gets involved. It's as much as 20% slower than Intel's chips. This is very odd as high-resolution gaming with heavy-duty graphics cards is just fine. Everyone is trying to puzzle this out as no one has ever seen it before.

I already explained this and if you do not understand why this is happening, you shouldn't be posting benchmarks, or opinions of CPU performance.

I'll try to make this real clear for everyone considering a Ryzen system in the near future. The 1700X and 1800X are very poor price to performance value. The 1600X and 1700 look to be on par with Intel in this regard. The Intel 7700k is still by far and away the best performance for price CPU, PERIOD.

If you are using Bitwig/Abelton as a VST instrument and a live sequencing type host, the 1600X or 1700 Ryzen is an okay choice if you are in the situation of needing to purchase a complete system (motherboard, memory, etc). If you have an i3/i5 and only need a CPU, you may want to get a 4/8 core i7 and an SSD or more RAM with the difference in total cost.

If you are using any DAW and ASIO (it is a single thread) round-trip performance is a concern. The 7700k is the only CPU to even consider.

Windows 10 (and 8,7,XP, etc) core/thread scheduling. Beyond 4/8 becomes an issue with diminishing returns and either needs a dedicated and competent dev team to want to pursue wringing out every bit of performance in this area, or they focus in the areas that still have a reasonable ROI in the current OS marketplace.

As someone that uses other people's money to build out and coordinate support for various systems, currently AMD is never considered.

[malifica]

PCs have always had problems with compatibility of parts etc.

I have had a very long professional career that has required this to NOT be the case.

The memory speed has a larger effect on speed with Ryzen than with Intel processors.

You may not fully understand RAM frequency and timing. But yes, there is an issue with the AM4 chipset (a bios update is needed from the MB manufacturer) does remedy the RAM issue as the BIOS is not automatically recognizing the RAM.

[malifica]

I've always bought AMD chips when possible. My home server ( AMD A8-5600K APU with Radeon(tm) HD Graphics ) runs 8 virtual machines, 12 database servers ( I'm a data engineer ), hosts 3 domains ( personal, charity and friend's small business ), runs a motion-detection security system, has the boinc distributed computing client using all spare CPU cycles, and *never* skips a beat. The Intel fanboii FUD "random compatibility issues" is total BS. AMD won the race to 1Ghz, gave us the AMD64 architecture, and now have very competitive chips at the high end. Bring it on

IT Director and Database Engineer of 20+ years here. I have never heard of a "Data Engineer", but I guess you gave yourself that title. Anyway, while AMD was easily sourced via Dell as a competitive office desktop/lite-workstation solution, I purchased 1000's. Had it not been for the Dell service and "care" support, I would have likely been relieved of my duties by the powers that be.

Now this is ultimately anecdotal but over a 6 year period, for every Intel Dual Core system we had an issue with, we would have 4+ AMD systems with hardware trouble tickets opened. There are also other manufacturers of specific components I have learned to avoid by trial and error and "oh crap" moments.

Hard Drive - Avoid Seagate (But I did love those Barracuda 10k drives in a RAID0 before they would inevitably fail)
Motherboard - Avoid Biostar (and Abit/Asus/etc. while the capacitor problem was going on)
RAM - Avoid Generic and Off-brand
Video Card - Personal use I always use and recommend EVGA due to their replacement and upgrade policy
Aftermarket Cooling - Only Noctua Period
PSU - The Dell sourced PSU of various vintages seem to last for a very long time. Personal use I go with modified Cooler Master PSUs.

Anyway, everything you mention is easily able to run concurrently CPU wise on a low-tier eMachine from many years ago.

Running a Linux or BSD flavor does not equal "Server". Using boinc = Never get hired by a real IT department.

"High End" I don't think you know what this means.

Did I bring it?

[malifica]

Yes, I did yesterday. Works quite well. I had 40 tracks and about ten FX (comp, EQ, limiter, enhancer, etc.) on the master channel and it bounced between 19 and 34% CPU usage. No detectable dropouts in audio.

bitwig 40 tracks lots of master plugins.jpg

This is highly unscientific as I didn't use the same song in other DAWs and didn't spend more than 10 minutes at it, but if anyone was somehow worried that the program wouldn't run correctly on AMD, or there would be weird issues, you can rest easy.

I will be doing other tests. We might actually add a DAW to the CPU test MO which currently only includes CineBench for multi-threading. Probably Bitwig so stay tuned.

(referring to your screenshot)
This is NOT how you observe CPU usage when conducting any sort of benchmark. Are you sure you are who you say you are?

Thankfully that after 10 minutes you are able to give us the "ALL CLEAR" signal. whew.

[malifica]

Based on some audio benchmarks posted so far, the Ryzen seems to offer good processing power with plugins when using reasonable buffer sizes (128 or above), exactly what I am looking for. I have been using 512 / 256 samples of buffer for many years, and find 256 samples optimal for my use. With this buffer size, the Ryzen seems to perform just fine (based on some benchmarks).

Not sure which sites and corresponding audio workstation benchmarks you are referring to, but each published ASIO benchmark I have seen, and the few I have personally conducted with the whole currently available Ryzen range indicates that anything below 512 is haphazard.

Just to offer some additional perspective. A properly configured Win10 DAW system using a 3770k 32GB and SSD with a typical prosumer USB2 interface and manufacturer ASIO driver; You should be at a maximum 128 and sub 3 second latency in all but the most extreme situations (we typically run at 64 with what is practically 1ms). If you are not comfortably at this level, something is wrong, or your expectations are hobbyist.

Do people out here really use 10ms second+ latency settings and are comfortable working that way?

I even have a problem with ANY latency when recording live and doing overdubs, in these instances I'll use a real-time monitoring setup with an outboard placeholder sound, and recording only the DI (for re-amp or whatever later).

[F5DX]

Thanks for your views malifica, we get your point. I was referring to the Scanproaudio charts, mainly the following DAWbench chart:

I did not find any information about which Win10 and Bios cpu settings & DDR4 frequency did scanproaudio use in these tests. Other sites have shown that the DDR4 frequency has a larger effect on the performance of Ryzen than in the case of Intel processors + the Windows 10 scheduler is currently ruining the performance of Ryzen, if the virtual cores are in use. Win 7 works better and does not show the scheduler bug that tries to use the virtual cores in the same way as the real cores. AMD and MS are supposed to be working to solve the issue. I believe we will see the real performance of Ryzen around April-May, once the initial issues are ironed out.

Edit. Seems that they already offer a Ryzen version for audio work: https://www.scan.co.uk/3xs/configurator ... -studio-pc

My setup will be very similar, however I went with 2x 3000 MHz Corsair LXP DDR4 memory instead of the 2133 MHz. The higher memory clock has potential to result in higher performance with Ryzen. I am wondering, if scanproaudio used this memory in their DAW tests + had the virtual cores enabled in Win 10, this could explain some of the non-optimal results in some of the tests that they ran.

[malifica]

Thanks for your views malifica, we get your point. I was referring to the Scanproaudio charts, mainly the following DAWbench chart:

I did not find any information about which Win10 and Bios cpu settings & DDR4 frequency did scanproaudio use in these tests. Other sites have shown that the DDR4 frequency has a larger effect on the performance of Ryzen than in the case of Intel processors + the Windows 10 scheduler is currently ruining the performance of Ryzen, if the virtual cores are in use. Win 7 works better and does not show the scheduler bug that tries to use the virtual cores in the same way as the real cores. AMD and MS are supposed to be working to solve the issue. I believe we will see the real performance of Ryzen around April-May, once the initial issues are ironed out.

Edit. Seems that they already offer a Ryzen version for audio work: https://www.scan.co.uk/3xs/configurator ... -studio-pc

My setup will be very similar, however I went with 2x 3000 MHz Corsair LXP DDR4 memory instead of the 2133 MHz. The higher memory clock has potential to result in higher performance with Ryzen. I am wondering, if scanproaudio used this memory in their DAW tests + had the virtual cores enabled in Win 10, this could explain some of the non-optimal results in some of the tests that they ran.

Nice screenshot selection... the only chart where they actually state is a purposely fictional environment for giggles.

I commented earlier on the Scanproaudio "review" and the conclusions they come to are "interesting" to say the least. However, they do observe and report on the same issues that I, and others are seeing with ASIO performance with the Ryzen chips. Which in the real world is probably the most important aspect for a DAW.

The scheduling 'bug' everyone keeps referring to was a LINUX kernel bug. This same issue does not exist in the Windows 10 kernel.

If you look for scheduling related Ryzen patches to the linux kernel, all you find is two to amd.c c->x86 == 0x17 one deals with the max_number of cpu cores while the other is setting the last level cache ids to the correct core. An up-to-date core info in Windows 10 already maps out the cache hierarchy of the Ryzen chips correctly.

As the MS scheduler moves a thread from one core to another, it knows that this might be an issue performance wise as it is already aware of whether the L3 cache is shared between the cores (this can be more problematic in the Ryzen chip due to the micro-architecture). Unfortunately, well researched heuristics are probably to "blame" and this will be Bulldozer 2.0 all over again.

Window's scheduling is already working just as intended and I wouldn't make any monetary decisions based on a mythical fix that is almost quite possibly maybe somewhere down the road.