i9-13900K vs Ryzen 9 7950X

Hello everybody,

I would like to get a new audio workstation PC soon. The lowest possible latency is important for my projects. Some of my tracks are quite complex, many virtual instruments are used.

I am faced with the choice between:

  • Intel i9 13900K on a z790 motherboard
  • AMD Ryzen 7950x on a x670 motherboard

It would use 64GB (128GB) DDR5 in any case. Budget and power consumption are not that important to me. Which one is better?

Thanks for replying.

1 Like
1 Like

The above support article raises some questions about 13900K that would also be an important factor in the OPs decision :

  1. So given that Nuendo doesnt like the e-cores, if we disable them at bios level will there be an indirect hit on cpu that will affect performance? Allow me to ellaborate:

Lets assume firstly that the PC is running a DAW only ie Nuendo and associated utilities only.
Additional essential utilities running on my system are RTPmidi, ilok, RME total mix, network for VePro, ShuttlePro driver. These all need to be running.

Were these services able to run on the e-cores, and will moving them to the p-cores create a net cpu deficit when comparing P-cores only vs P-cores+E-cores?

  1. What about if we want to have a browser open for manuals and quick searches etc. Was it able to run on the e-cores? but now, since disabling e-cores, is taxing p-cores and competing with Nunedo? Or were they always using p-cores?

  2. What services related to audio production (not including Nuendo, but typically are used simultaneously) can run on the e-cores?

  3. Is there a CPU affinity utility, such as Prio or Process Lasso, that is favoured to work with Nuendo so that I can have the e-cores still running and not have to change the affinity every time I load N12? I could force my browser and non-realtime utilities like total mix to use e-cores. And force Nuendo to use only p-cores. Could this be problematic?

Rant: I had a gut feeling these e-cores were going to be a waiste. So really for my relatively new 12900K I’m back to 8 cores. Does the AMD have an equivalent issue?

BTW. I have mildly overclocked and locked the frequencies on my system. P-core locked to 4.9ghz and E-cores locked to 3.6ghz.
Given that all E-states are disabled, my cooling is good etc are the e-cores then usable by Nuendo or other background audio services? Low latency is important as I am a heavy VSTi user.

Well before you disable your E-cores, I’d test things and see how they work. I think this may be a bit of Steinberg overreacting to something now and just saying “If you have a problem, you can’t blame us!” I’d test out something with the CPU fully enabled, see how it goes.

This assumes you are on an updated Windows 11 system, of course. Near as I know the improved scheduler hasn’t been added to Windows 10, just 11.

As for Intel vs AMD I don’t know, I imagine both should work well. One datapoint out there is the guy behind DAWBench tested both the AMD 5000 and 7000 series as well as the 12th and 13th gen Intel and the 13900k was able to support the highest loads, at all latencies, so long as it had DDR5 memory. The 7950X beat out the 12900k by a reasonable amount. This was with E-Cores turned on.

So if you have an Intel system, I’d try the E-Cores on unless you have problems with it. If you are thinking about Intel vs AMD I don’t know. Personally I’ll probably get an Intel when I upgrade in part because that’s what I’ve used more but also in part because I’ve had issues with AMD and USB ports, and my interface is USB.

1 Like

So at this point (and to help answer the OP’s question), can we assume that the AMD 7950X doesn’t have the same issues? I’m not as familiar with AMD processors so I’m just trying to verify that its architecture doesn’t cause issues.

AMD doesn’t have this architecture at the moment. I’m not familiar with their future plans, but what I know it is not yet planned.

However, even with the new P-Core/E-Core architecture from Intel you do not neccessarily get performance issues. This highly depends on a huge bunch of factors in your specific system and configuration.

The problem is that Steinberg currently does not have real control about how processing is done, because that is done down in the Windows 11 thread director. There seem to be issues with this new techology that Microsoft needs to work on, like this one for example

A friend of mine @Norbury_Brook recently built an 7950X system for Cubase and it’s performing incredibly well.

1 Like

Thanks for the numerous replies! After a lot of research, I decided on the Ryzen 7950x with an x670 motherboard. The performance of my system is excellent - much better than expected. I have no experience with latest generation Intel processors, but I can fully recommend the Ryzen 7950x.

4 Likes

I have exactly the same choise problem!!! I use the same workstation for 3d (3dsMax+vray/corona rendering) and video production (Premiere pro/After Effects) and the AMD solution is absolutely a better choise…but I keep hearing that intel solutions are better with cubase and amd causes problems?

Would be interesting to know what you are hearing here, any sources about that?

So to address both parts:

Debatable. The 13900K is extremely powerful, depending on the workload it can be faster. It is also a bit cheaper, though in both cases you are talking an expensive enough system as to not fret about $50 or so. Basically I’d say there is no clear winner for high end workloads right now. It will depend on the specific workload which is ahead, and they are close enough that you won’t really have an issue if you got the “wrong” one.

For anything not a high end workload, as in anything not completely loading them up, they are both more than enough.

This is one that is questionable. I’d want to see some current testing but I haven’t. There have been two issues, historically, that made Intel better:

  1. AMD’s chiplet design has latency moving loads in between them that Intel’s monolithic design doesn’t. OSes haven’t been aware of the chiplets as separate things, and I’m not sure how well they are doing with it now, so they didn’t do a good job keeping things on one chip. This is minor, but it is a thing particularly for low latency loads.

  2. AMD has had sub-par USB implementation in the past, and since most audio interfaces are USB these days, that leads to problems. This is a bigger deal. It can be solved with an addon USB board, of course (so long as said board uses a good chip). This was a real issue, not sure if it is anymore. As with any issues like that it could be a thing where one brand of interface is 100% fine and another has all kinds of problems.

The chiplet problem is something that is still a current issue to a degree, Windows 11 22H2 had some real issues with the scheduler and the chiplets. It has been patched though. The USB issue I have no idea if this still applies or if this is just people repeating poop that was true a long time ago, but isn’t anymore.

I wouldn’t worry about it a ton. If you are an AMD fan, it should work fine. Worst case scenario, budget another $150 for a high end USB3 card (Sonnet Allegro would is a recommended one). If you want Intel, get that it works great too.

When it was my money, I got an Intel CPU mostly because it is better for gaming, which is what I do a lot of with my computer (audio is just a hobby for me). It works great in Nuendo though, at least with Windows 11 which has an improved scheduler for the different kinds of cores.

I feel that is a largely irrelevant thing to consider. It’s a bit like talking about Intel’s use of power. In the end who cares if Intel uses more power or if AMD’s chiplets induce more latency - what matters is the actual performance.

Over generations latency between cores have improved on Ryzen and depending on what year it is and what half it is of that year Ryzen CPUs may outperform Intel or vice versa. It really doesn’t matter whether the Ryzen CPU has one or two chiplets (the 8 cores don’t).

As far as I know this was a fairly limited set of scenarios. I also think that a lot of these “issues” people end up talking about are issues they talk about because people are talking about them - more so than them actually being serious issues generally speaking. I mean, it’s not like we couldn’t dig up issues on Intel, either with USB or Thunderbolt. I’m sure they exist but it’s never a problem ‘in general’ with Intel. Quite frankly it seems like people are just inclined to find something wrong with a brand in general and hold onto anything that supports that (I’m not talking about you).

I’ve seen zero reports of USB being problematic recently. I run my MOTU 16A on USB or Thunderbolt and both work just fine (specs in sig).

I agree with the rest of what you wrote though and I don’t think people can go wrong here.

I mean… did you read what the OP wrote in the previous post back in December before you resurrected the thread?..:

So where it matters is low-latency, realtime processes. If you’ve ever watched your scheduler (not something most people would do, I’m a silly nerd) it bounces processes around core to core. There’s various reasons why it’ll shift a workload to another core, and outside of a kernel debugger you don’t know why but the point is it happens, and more often than you’d think.

No big deal… except there IS a small penalty when that happens. It isn’t an immediate switch. Well with everything on a monolith die, it is near enough as makes no odds, hence why it can do that. However, the more separation there is, the less true that is. This is something that has to be accounted for in actual multi-processor systems. There the scheduler will NOT move things CPU to CPU unless there’s a good reason. Likewise, programs can tell the OS that they aren’t to span NUMA nodes unless absolutely necessary because of penalties in memory access speeds.

A chiplet design is not the same thing, there’s no NUMA, all memory goes through one controller, and the internal bus is much faster than the QPI/Infinity Fabric link. But it still takes more time than moving to a different core on the same chiplet, enough that it could potentially cause hitching in time critical tasks.

Gamers found this out back in the day, that disabling one of the chiplets lead to better performance since the game wasn’t using 16 cores, and it stopped the threads from getting bounced across the chiplets. Low latency audio could likewise be affected, since there is little time to tolerate stalls.

I don’t think it is likely an issue today, at least with patched Windows 11, it has an advanced scheduler that should be aware of the chiplets and schedule accordingly, though I haven’t been able to get confirmation of that, same as it is aware of the P-cores and E-cores and schedules accordingly. Just saying it was one thing I was aware of that was an issue back when they first went to that design.

As you say, either would work well, you can’t really go wrong. CPUs are getting stupidly powerful these days and you’d need to be doing pretty huge templates before you started saying “man I could use more power” with a current gen high-end CPU.

The problem is that that statement is anecdotal and subjective, especially because it is limited to experience with the Ryzen choice only. You can start digging around The Interwebs and you see “testimonials” and benchmarks supporting both sides. I am dealing with the same situation and the problem is that there are so many variables; motherboard, RAM, drive, OS, patches, audio interface drivers, you would have to have the exact setup someone else has, including all other software and drivers on a machine to know for somewhat sure it may work better.

There is no problem with that statement. Felix clearly states it’s a good CPU on his system. The point I made by quoting him to the person reviving the thread was that a) people who talk about what others supposedly talk about are often just talking about things they don’t really know anything about, they’re just repeating sort of what they heard people say they heard people say, and b) in THIS thread the OP had already made his choice a while back and was really happy with it.

How does it make sense to resurrect this thread to ask about worries about the CPU brand the OP last said was performing excellently? Doesn’t make sense to me…

If anyone wants a solid system then just ask someone with a solid system what their setup is and replicate it. Or get close to it. Or just pick something at random if it’s all anecdotal and subjective and nothing therefore matters.

As usual with forums, resurrections happen when someone searches for a topic (and they should be commended for that, instead of just blindly posting their problem without checking if someone has already talked about it) and joins an existing thread, which sometimes may have been dormant for a bit. I mean I am not insulting anyone’s intelligence, that is just my understanding of these “The Interwebs” peculiar terms.

giusfire resurrected the thread because he said he kept hearing Intel solutions were “better”. Saying that AMD works in a certain configuration for someone else does not really address if Intel is possibly better or worth the money more. And it also does not mean others are not having problems with AMD.

We do not have to go to the extremes and just pick something at random, my point was that that statement was specific for OP’s setup, and also, it was not a comparison. Maybe another setup would have worked even better. That is actually how I stumbled upon this thread, so to me the various perspectives have been helpful. Considering that there are 12-year running threads on, for example GearSpace, picking up a thread after just a few weeks is not that much of a “resurrection” stretch…

Maybe someone has new perspective on the subject. I certainly will when my setup is finished, because it involves both processors in question. And in my case, the Thunderbolt wrench gets thrown in my game as well, so I am sure I will go through some learning lessons…

Cheers!

Time for another resurrection. I just built a 7950x based system and it absolutely smokes. Wow.
The AsRock mobo has built in ThunderBolt 4 now as well, so reasonably futureproof…

1 Like