we observed significant performance boosts if you raise manually your RAM frequency and lower RAM timings with Core Ultra 200s (eg 265k, 285k). Overall they are not as good as 9000series Ryzen in our test. However they might have higher performance to price ratio now considering those real stock prices.
Considering
and Core Ultra 200s have really fast E cores and there are a lot of E cores, you may get high performance in one project, but unexpected low performance in another projects.
Though Ryzens get lower RAM latency due to the design, they reach the limit of RAM IO bandwidth way earlier than intel Core Ultra 200s IN THE SAME DDR FREQUENCY also due to the design. So if your projects go large with plugins that are bandwidth hungry, these Ryzen 9000s may sub-perform. This matches our observation.
Hopefully you are all using CUBASE 14 Why are you guys buying such overclocking chips unless youāre in tho gaming but as for our beloved DAW you donāt need it just make sure you have a minimum of 64 GB Ran and some fast drives. Put your money to better use We never have a problem in the studio with simple equipment and yes we use the ultra 7 265 not K but then again we use audio modeling SWAM library which hardly uses any CPU compared to sampled VST3 instruments.
It of course depends on how you use the DAW. Some users work on large projects with several hundreds of tracks, many with insert effects. Then add sophisticated signal routing with busses and further inserts on those busses and you will find that you do indeed need a strong CPU in order for it to keep up.
I would love to hear a project that uses several hundred tracks on you tube. I would be very curious of the genre and the libraries. My biggest problem with CUBASE 14 in my studio is the score editor
Just turn on the radio and listen to some contemporary music.
In addition, film and TV composers often use very large templates where the track count can reach over 1,000.
itās not necessarily the actual audio tracks that make projects large. A lot of mixers these days use templates that might have 50 FX tracks because they split things up into instrument groups so stems can be printed without the rverbs etc being poluted with other intruments. So youād have all your various amb, short, long, FX etc for just your drums, and the same again for vocals etc etc. Then everything is grouped and bussed separately and then ontop master Bus groups for your Mix tracks then to the Master outs. All thsi routing hammers CPU and you can end up with 100 tracks on a small project.
take a look at MArc Daniel nelsons template as a good example.
5 continuous notes of polyphony with 3 instruments playing The chord (ultrasphere (heavy), fluidchords2, vps avenger)/ loopcloud on/2x chop-r.
1x futureverb on fluidchords channel DaisyZL 2x oversampled+IHNY2 on master. 256 samples/asioguard ON.
Process lasso probalance On, cubase excluded from probalance. (I will now try different ProcessLasso configurations to check which one suits the best but I took the ez way as a first trial. )
Usual daw powerscheme+bios tweak.
Nvdia latest studio drivers.
MSI pro z890 p-wifi- Xmp on 8000 MHz.
No cpu OverClock/TurboBoost on.
I tried to put a maximum of information on one picā¦.
Second pic: same situation + heavy constraint on master bus.
OAK+INK both at HIGHER oversampled mode = absolut cpu hogs beasts
256 samples/asioguard OFF
(at the limit of spikes.)
That ultimately means i can work with this configuration at 1024 samples easily and maybe still improves it. (while waiting for a better optimization from Steinberg about core/thread work distribution)
I tried different thingsā¦.but it seems thatās the core Cubase use the most in my 285k rig (msi pro890 P-wifi) are obviousely the P-Cores (0-1/10-13/22-23) but the E-cores used are a 6-7/8-9ā¦.
Waiting for Cubase 16ā¦.
ProcessLasso on.
All audio on p-cores (except 0) + effective e-cores
All different plugin downloaders and suchā¦on e-cores not the ones used by audio.
Interrupt affinity tool applied:
Nvidia Gpu on core 0
RmeUfx2 on the others p-cores and the mentionned e-cores.
I can work on heavy project but still feel a bit robbed somewhere, somehowā¦.
Since youāre posting screen shots, for LatencyMon results, you really need to run it for around 10 minutes. Running for < 1 minute like you have here can be misleading. What LatencyMon is showing is also across all cores, whereas DPC latency itself is per-core.
That all said, when you use process isolation tools, the displayed results from LatencyMon arenāt all that helpful or relevant if youāve already isolated audio from other cores.
Also keep in mind that if you use a tool like process lasso, you restrict any plugins hosted in the process from creating worker threads on other cores. Any impact here is going to vary by plugin.
The RME ASIO driver will run in the Cubase process and so will fall into whatever you have setup for that process. The RME kernel driver will run under the kernel, not restricted to any specific cores, as would the NVIDIA GPU kernel driver. Itās been a while since I last looked at PL, but does it let you pin kernel driver threads? Or are you doing just the interrupt pinning?
Iād be curious to see what your results look like if you run your validation as-is, and then try it again without the process lasso pins in place.
Otherwise, cool. Looks like you have a nice and stable setup
Thankās for your advice.
I got indeed a single major spike after 4 minutes caused by the infamous nvidia driver but in cubase there is no glitch at all.
Generally when my system is more unstable (when trying to find a balanceā¦) ā¦those spikes do appear after a few seconds of latencymon bringing severe glitches in audio.
A single one after 4 minutes means I can work without and problem. (Generally).
I will try your recommandation this afternoon and post back after trial.
I must admit that I canāt see any real obvious and significant difference in the cubase cpumeter on the actual projects Iām working on (theyāre not cpu saturatedā¦at least not yet.), with or without Process Lasso running.
I let the Interrupt affinity tool (from microsoft) with the modifications I initially made
ā> 5060Ti on P-core 0.
ā>RME on P-core 1 ,E-core6,e7,e8,e9,p10,p11,p12,p13,p22,p23
Seems ok like thisā¦(and maybe there is no big deal here after allā¦but Im not sure)
Maybe it will a bit more significant on an extra charged cpuā¦I donāt know yet.
I will repost here if I find interesting new aspects or findings because Im also curious to understand and check things by myself for my own rig.
Still investigating the numerous combinations while scrutinizing projects and plugins structures.
About latencymon: no spikes or really rare spikes unless I begin to open other apps or soā¦
(special notice: I realized that on my MSI z890Pro P-wifi motherboard, while tweaking the bios to improve dpc latencyā¦that if the PCI power management (haaaaa cant really remember the exact item name) was disabled (for audio it is recommanded to disabled it)ā¦it rendered my iGPU drivers totally nutsā¦disabling themā¦and this can be REALLY frustrating if you wanna test your igpu to escape nvidia kernel to try dpc for instanceā¦because it is not an ez culprit to find at allā¦Maybe it can help some people as advice). So I decided to let that function enabledā¦Because i need my igpu sometimes.
I also loaded the āAustin Hull Cubase 13 demo projectā into Cubase Pro 14, as did af06fr. I am running the Ultra 9 285k. I took a screenshot while playing the chorus which seemed to be the most intensive area of the song. The buffer size is set at 48 samples. Only the (P) cores are activated for Cubase.
This song is only used as a comparison standard.
Also the number of PCI lanes available to the user depends on the motherboard model. The high end models have about 24 lanes. distributed to the graphics card 16 lanes, SSD etc. If you want to expand without reducing performance, go with a motherboard with most lanes.
This is a dangerous configuration for a PC, because RAID 0 is just a stripe of discs. If one disc out of the configured set fails, all data are doomed. There is no fault tolerance in this setup.
The fastest Raid with fault tolerance is Raid 1, but that requires an even set of discs. It is striping together with mirroring, so you need 2, 4, 6, 8, etc. discs to create a set of stripe discs, that are mirrored on another identical set.
The backdraw in Raid 1 is, the size of the disk is exactly half of the total amount. As @MattiasNYC says, speed is not an issue, so the best is to take his recommendation of Raid 5.
