Ryzen 3900X, 3700X, or Intel i9-9900K for DAW machine?

djburton

New Member
I’ve been researching this to the point of blindness, but wanted to reach out before making a final decision.

Current setup: 3 Intel PCs:

1) DAW (1st machine - “master”) i7-4790 (not K); 32 GB 3200 MHz RAM. Samsung 970 EVO NvME 500GB boot drive, 2 1 TB SSDs for samples, 1 500GB spinner for data. Scarlett 6i6 USB. Arturia Keylab 61 Mkii; Novation Launchpad X. Cubase 10, VE Pro 7, Kontakt 6, a smattering of synths (e.g. Pigments 2), Ableton Live 10 (standard), Bitwig 3. Some other stuff.

2) 2d Machine “Slave” - i-8700K (not overclocked at this time), 64GB of 3200 MHz memory, VEPro 7, Kontakt 6; EW Play (latest version) – dedicated to strings (EWSO, EWHO, 8dio agitato, SF Chamber strings & solo strings) choirs and percussion. 1 boot/data drive, 2 sample drives (all SSD).

3) - 3rd Machine “Slave” - i-8700K (not OC’d at this time), 64GB of 2400 MHz memory, VEPro 7, Kontakt 6, EW Play (latest version) – dedicated and brass and woodwinds (EWSO, EWHO, Chris Hein). 1 boot drive, 2 sample drives (all SSD).

Performance is OK. Some audio issues (pops/clicks) on heavy passages but a good for small to medium-sized sample orchestra arrangements.

I run the Scarlett with a 512-bit buffer.

I want to upgrade the DAW machine to a Ryzen 3700X, Ryzen 3900X, or Intel i9-9900K (with X570 or Z390 motherboard, as required) for more mixing headroom and future-proofing.

I know the Ryzens are relatively new and I’m looking for actual experience with single- and multi-core applications with these processors.

I’m leaning toward the Intel configuration. Interested in experiences from Ryzen users with either of those CPUs, especially the 3700X which still appears to be an advantage in terms of pure speed and threads at modest power requirements, but the Scan Pro data posted so far appears to show a big gap between that and the other two processors.

Thanks.
 
Last edited:
OP
D

djburton

New Member
Well that’s embarrassing. I didn’t buy the 3600X. Got the 3900X (as now edited). Downloading Bitwig content now. Build was not quite uneventful as the Asus TUF MB wireless driver thwarted Windows installation. Disabling the driver did the trick. Re-installing programs this weekend. I’ll assemble informal observations on DAW performance over the next few weeks.

DJB
 
Looking forward to your shared experience of using the AMD 3900x. I'm curious in particular about performance with a lower buffer like 128 or 256 and how that behaves running virtual instruments!
 

John Longley

Surly and Charitable
I’ve been researching this to the point of blindness, but wanted to reach out before making a final decision.

I run the Scarlett with a 512-bit buffer.


Thanks.

As for dropouts or pops and clicks, I would say the weak link in real-time ASIO are these drivers. While the Scarlett line is just fine in terms of Audio, I have noticed a LOT of people with issues with these drivers once under load. If you can borrow, rent etc an RME interface or pick up an old MOTU 424 PCIE rig for peanuts (and do some testing), I think you would be able to drop your buffer an order of magnitude and see your latency improve and your buffering issues cease or radically improve. I used to use a ZOOM UAC-2 for my travel rig and it was pretty great too, if you just need something to experiment with very cheaply.

I run a 3900X, and while my orchestral stuff is as a hobbiest, I can easily run a buffer of 64 with RME drivers with hundreds of plugins when mixing professionally and 100 plus VSTi is no issue. I could probably go sub 64 with a surprising number. I don't work with a massive template, so I have not tested a very large template-- but I would be very optimistic. I upgraded from an 8700K and I expected a drop in real time performance, but I have not seen that AT ALL. I have not tested your use-case, but I think the 3900X is a better option than the 9900K in most cases, and the ones it is not are academic. The greater cache and IPC make up for the marginal deficit in latency IMO.

Lastly, if you have a sub optimal integrated Wifi chipset, that could also be causing you some issues in ASIO performance.
 
Last edited:
OP
D

djburton

New Member
I promised an update of my experience in installing/using the Ryzen 3900X on my DAW machine.

I’ve had it up for the last couple of weeks. I have run no precise measurements or benchmarks. These are just subjective results.

New on the “Master” are the following:

1. ASUS TUF X570 Plus w/ wifi (I don’t use the wifi so I have it disabled – this was the best deal in MB’s in combination with the processor).

2. Ryzen 3900X with a Noctua NH-U14s cooler.

3. AMD RX550 graphics card (Replaced a simple nVidia card)

4. MOTU M2 Audio interface (replaced the Focusrite)

All memory and drives are the same as before. I got the X570 as a value proposition. I got one of the most basic AMD graphics cards I could get since I do not game on this machine.

I also made sure the 2 i7-8700K “slaves” are tuned up to take advantage of higher speed. They are not truly overclocked – they don’t run permanently at top speed, but with VE Pro running stay up around 4.5 - 4.6 GHz. The Ryzen is configured similarly.

The net result of all this is that I can run a moderately busy orchestral score (about 20-30 tracks with a reasonable level of CC automation) at a 256-bit buffer level with very few dropouts or pops. 512 is totally stable under the heaviest load I typically put on these. This is a substantial improvement over the 4790, and plenty good for me, but not quite up to the performance some on this forum report high-end processors, albeit there are no Xeons or Threadrippers in my rig.

Here’s the order in which I experienced improvement.

1) The old nVidia graphics card presented problems, causing periodic drop-outs and occasionally not even providing a signal at boot-up. The RX550 seems at home. This was the first obstacle to overcome.

2) I followed several You-Tube summaries of system optimization for Windows 10. This speeds the whole system up a bit but not so noticeably for audio performance. I already had my power profile on the equivalent to performance mode.

3) The largest single improvement in audio performance came from getting rid of the Scarlett. This was the last thing I replaced and before that I still had the same buffer requirement at 44.1K sampling – 1024 for stable use and 512 for small undemanding arrangements. Thanks to John Longley for prompting me to think about this. I did not want to get into the more esoteric RME or Apogee stuff, but it seems for consumer-level interfaces, the M2 sits at a very high level and so far it works perfectly for me. Note that I do no live recording or anything other than MIDI/Sample-based orchestral arrangements.

Miscellaneous notes -

-VE Pro 7 run on my DAW machine using Cubase or Studio One exhibits odd behavior with both Play and Kontakt. I can play notes from my keyboard with no issues, but if I use the GUI keyboard in either plug-in, there is a very long delay (20-30 seconds) before a sound is produced. I frankly can’t recall if this happened in my old Intel setup but I don’t think so. Direct Play and Kontakt VSTs have no such problem.

-At idle, the 3900X runs a bit hot (35-36 degrees C) vs the 8700K’s (27-28C). I don’t think this is unexpected. Voltages are below 1.4 in all my configurations.

On balance, I’m glad I did this, but there were a few pebbles in the road. Being able to hear my stuff with more headroom without audio glitches makes it much easier. Thanks to everyone who gave feedback.

DJB
 
Thanks for reporting your findings, DJB. It sounds like 20-30 tracks at 256 Buffer is a little disappointing from what I would expect. I can get much more than that on an Intel 5960x 8 core overclocked to 4.2 GHZ and this machine is 5 or 6 years old. I am using an RME Babyface Pro though so maybe performance is also based on those drivers?

Glad you are having better performance than your previous machine though!
 

John Longley

Surly and Charitable
This performance is really bad, what interface/ drivers are you running. On RME my 3900x pounds my 8700K into the dust. I can stay at 64 so long as I don't load the master down or have hundreds of plugins on with the template.
 
OP
D

djburton

New Member
Well, the interface, as stated above, is the MOTU M2, and its driver is current standard issue (1.01 I think). I haven't yet been willing to take the leap into the price zone of RME interfaces.

Sample-based music reproduction is purely a hobby for me (the most fun I've had in years). While I can live with the current performance level, I'm just obsessive enough to keep hacking away at improving it further, as time permits. Your (John's) and composingkeys' comments strongly suggest that drivers in consumer-class audio interfaces are a hard limiting factor.

I'll be back if and when I know more, but in the mean time I will keep monitoring this forum and will play around with latencymon or similar tools to see if I can narrow down what my particular wall is (while I keep working on Mozart's Serenade in Eb :)) .

DJB
 
Last edited:

novaburst

Senior Member
Glad to see some are finally using the AMD, but performance does not appear to be shinning when it comes to the things that matter.

Seem Intel has one or two secrets that AMD just cant quite work out as yet, the speed is there but the intellect out right performance seems to be lacking.
 

John Longley

Surly and Charitable
Glad to see some are finally using the AMD, but performance does not appear to be shinning when it comes to the things that matter.
I saw a thread on the NI forum where a brand new mac pro owner with 32 cores and 192 GB ram could only run 5 instances of massive x.

We have to evaluate feedback carefully.
 

josejherring

Senior Member
I use AMD and have for years as slave machines. Performance is great.

If OP is reporting ridiculously high latency numbers my guess is that it's in the samples he's using and perhaps using standard harddrives I wouldn't suspect that AMD is the culprit. In my experience AMD works just as well as Intel.
@djburton

Can you tell us a little bit more about your setup. Why libraries are you using? What is your VEPro buffer size? Are you using SSD for your most demanding libraries?
 
OP
D

djburton

New Member
Thought I'd posted this a couple of hours ago, but seems it didn't take - so sorry if this shows up twice:

OK, everybody -

1) A few reminders and extra notes:

First, I have not offered up any definitive empirical conclusion about the Ryzen 3900X. To quote myself above from last Tuesday: "I have run no precise measurements or benchmarks. These are just subjective results."

Second, basic drive specs are in two of my posts in this thread. The only spinner in any of these machines is a data drive storing my Cubase project files. All samples reside on SSDs.

As to samples, all the orchestral stuff is on my i7-8700K slaves, with Strings and Percussion on one and Brass/Woodwinds on the other. I regularly use Hollywood Orchestra, Chris Hein Woodwinds complete, Chris Hein Orchestral Brass extended, Spitfire Chamber Strings, and Spitfire Solo Strings, occasionally pulling out SF’s Sacconi String Quartet, EW Symphonic Orchestra Platinum Plus (an Oldie but really Goodie), and a few 8Dio string items.

I keep any Kontakt instruments in separate instances from the Play instruments. VE Pro is set to use 2 threads per instance, as I typically use from 2-5 (as many as 6 sometimes) instances on either slave machine, and the 8700Ks have 12 threads.

Audio returns back to the DAW machine are usually just the main outs from each VE Pro instance, feeding a common stock Cubase reverb send if required.

2) A declaration of Accidental Triumph

I say “accidental” because, while I’m an experienced PC user, my knowledge of the Mysteries of Overclocking and the relationships between frequency, voltage, temperature, etc. is even thinner than my moron-level understanding of music theory.

Yesterday I dug in on a mission of hope, focusing solely on the Ryzen DAW machine. I’d already updated the ASUS X-570 BIOS to the latest version. and performed basic Windows cleanup (performance-biased power settings, turned off the Realtek Audio, etc.). I went back and restored “Optimized Defaults” in the BIOS, being sure to turn Wi-Fi back off, turned on DOCP, and made sure the “turbo-boost” setting was on.

Worked like magic. The machine now performs much better. LatencyMon, running for a hour, now shows low peak latency and low DPC execution time (before, DPC time would creep into the red zone after 5-10 minutes). The old (2017) version of DAWBench I found allows all 240 pre-fader and 80 post-fader inserts with absolutely no drops with a 64-bit buffer level (haven’t tried 32, which is probably where the threshold will be).

Most importantly, I can play back a 42-track setting of Copland’s Our Town at a 64-bit buffer level with no dropouts, cracks or pops. This is not a particularly challenging arrangement, but with my old processor and audio interface, I had to use a 1024-bit buffer level to be completely glitch-free. Way more headroom than I used to experience.

In my world this counts as a cure, and arguably a great endorsement for the 3900X. I wish I could tell you what weird nonsense I eliminated with the resetting of the BIOS, but I can’t. I know for a fact that the MOTU ASIO driver shows an improvement over the Scarlett 6i6 in my case.

My next task is tackling the 8700K sample machines, on which I have tried to perform basic automatic overclocking, but which still show inferior results in LatencyMon. Real-world, they don’t interfere with playback of my projects (as reported above), so I have a lower sense of urgency about them, but I’m totally convinced they can be improved short of my springing for more new hardware.

This has been a seemingly interminable post. I will let your heads rest now. Thanks for your generous feedback. Not sure this helps anybody contemplating purchasing one of these, but if you know what you are doing (I did not) here’s at least one success story to encourage you.

DJB
 
Last edited:

chimuelo

Star Of Stage & Screen
I’ve been reading 10 Pro Enterprise was the recommended OS by AMD, then shortly there after I read how this wasn’t the case.


Anand Shimpli would have never made such a mistake.

Modern media sometimes races to get the story out first, and makes mistakes.
Being 1st for clicks seems to be the priority.