CPU Recommendations

The problem being… Supported (via a list) and running at full throttle are 2 different things. This is even an issue with current MB chip combinations made for each other. The VRM’s simply aren’t good enough.

What is “full throttle” according to you, and which combinations are you referring to specifically?

Full throttle = Max clock speed. I wasn’t referring to any specifically, it’s a general statement. It will run but you will get throttling (it’s an issue on current gen, it will be worse mixing gen). The issue is those “supported lists with Bios upgrade” usually don’t go into this.

I understood that “full throttle” = “max clock speed”, but I was wondering “according to whom”.

Really the most important thing is if the CPUs will run at guaranteed clock speeds on compatible motherboards according to what we’ve been told.

So again, which combinations aren’t running according to specs? A “general statement” is pretty much meaningless as I don’t know who you are (or vice versa) or where you got that impression from. “Throttling” is really only a serious issue if the CPU drops speeds from what is guaranteed, within reasonable conditions.

So far I haven’t seen a single report on “throttling” being a widespread issue.

So, where can I read up on this?

PS: My 1700 got up to 3.7GHz all-cores overclock on the stock cooler. Zero problem.

By ohh well I mean the 3900x

The first round of testing as noted up top I used Cubase 10 initially for the DAWBench VI testing and everything looked great right up till the final test. On the 3900X I saw a 30% performance drop at 64 buffer with only 70% of the CPU being used at maximum load and 128 gave me 80% – 90% with the full CPU being leveraged at the 256 buffer and above mirroring the low buffer latency we’ve seen in previous generations.

It’s still early days, and BIOS updates are coming, Windows10-1903 has been partly optimized for Ryzen 3 and there is more coming.

I really would like to spent my money on AMD instead of INTEL, lets see how the 16 core is going to perform.

Language is an important thing.
What does it even mean - “performance drop”?

I would say it should mean that you have a reference point, or a starting point, and then you measure the performance difference between that and the next point, using a relevant parameter… like the ones in the DSP and VI tests.

3900x @ 128 buffer = 760 voices, and 400 @ 64 buffer. So what is the ‘drop’? It should be:

(760-400) / 760 = 47%

So that’s even worse. The new performance dropped by 47%, not just 30%.

BUT;

(1860-500) / 1860 = 73%
(1400-480) / 1400 = 66%

for the top Intel CPUs.

There’s a good reason for why Pete at Scan doesn’t create a chart with that specific number, 30%, and that’s because all it is is “untapped” potential, maybe, and it isn’t something we should care about when buying CPUs for DAWs. It would be beyond idiotic to buy a CPU that utilizes 100% of its capacity yet performs 50% worse than its competitor at the same price, for example. What we should care about is the performance that matters which is what the charts show - instances / polyphony.

So rather than pull that piece of text out of the article, how about just pointing out what’s relevant???

98.8% of the performance of a 9960x at about 30% of the cost!..
110% the performance of the 9940x at about 37% of the cost!..
33% more performance than the 9900k at a 3% premium.

…on the DSP test. Same trend on VI test.

Language is an important thing.

On Mac forums lol :slight_smile:

lol

Because that is IMHO the most relevant point (for me), the performance is there, just don’t expect that kind of performance at low latency buffers (on the 3900x)
I’m not arguing that the 3000 series are bad, or that they are not priced better.
I actually agree with what You are saying, I was just a little disappointed that the two chiplet design of the 3900x still has a negative performance impact.
Although after looking over the charts again, and looking at what buffersize I realistically could use, (RME Raydat PCIe. 128 would be low enough for anything I do), I really need to lighten up :wink:

Now the only “relevant” question is: Will Cubase be able to keep up with the ever increasing core count, that is going to come over the next years ?

But how do you normally judge performance of a new CPU if you’re not using DAWbench’s actual results on DSP / VI tests?

What you’re really saying here is that despite what Pete tested in the real world the CPU will perform differently from his tests. That’s really what you’re saying, right?

As for core count improvements I’m hoping Cubendo will keep up. I think it’s on their radar for sure.

The throttling I’m referring to has nothing to do with CPU heat. There’s loads of videos on youtube about VRM’s, partly the reason why we’re seeing improvements there. One major MB manufacture (I forget which without reresearching) was know for it’s last gen to be seriously bad. They stepped up big time this gen due to the amount of coverage and bad rep it got them.

When I was researching the latest I7 chips and MB’s, it’s all the reviewers were going on about. The current gen i7/i9 is backwards compatible with Bios update but it’s not recommended. It’s pointless putting those chips in anything but the best boards. One might as well buy a cheaper chip.

My CPU will clock to 5.0. I have a mid priced current gen motherboard (a cheap z390 £120), it will get current throttling under full load. Unless you check for this in utility software you wont even notice it as a thing. In audio that could translate to asio spikes (speculation). Even if not OC and locked to max turbo, extreme utility will show a quick current occasional throttling, these show as a little spike every 30 seconds or so. There’s 3 other types of throttling displayed, they’re all fine, pushed hard.

Everyone’s been so obsessed with cooling CPU’s the VRMs have been (until now) completely ignored by the consumer. The fan on my cooler doesn’t even spin up. It’s not the weakest link in performance.

Ok, I just searched for “Ryzen throttling” on YouTube and the first video uses an 1800x as an example, and starts with an all-core overclock at 4GHz. When hot it drops to the actually rated minimum of 3.6GHz. That’s exactly what it’s supposed to do. The automatic “max” values are a) not all-core values and b) dependent on the cooling ability of your equipment.

It’s not a throttling down that’s happening as much as it is a failure to maintain an overclock. Those two things aren’t the same.

Ok, I hear you. But I still haven’t seen any cases where the rated performance that the manufacturer lists isn’t achieved on a compatible motherboard. I don’t see “loads of videos” that show this.

Can you link to at least one?

No, just that I had hoped performance would scale like Intel CPU’s.
And kind of Yes, I don’t trust that I won’t run in to problems at low latency if the CPU already starts to show performance decrease. Different motherboard, processes running that I need… etc…
It is not as clear as if I wanted a rig for video rendering or 3D modeling, then the 3900x is insane for the money.
But I see Your point.

Edited to merge posts but delete post option has gone.

I referred to it as throttling in my OP as that is what it’s called in intel extreme utility. I specifically specified it as “current” throttling to highlight that I indeed wasn’t referring to what normal people call throttling. I did this deliberately and repeated it to avoid that conception being made.

My point is, what’s the point in upgrading to a new CPU, on an old motherboard, which will have sub par VRMs, therefore won’t be able to maintain a stable TURBO for duration*. Meaning backwards compatibility “(wink intel)” wasn’t really an issue, even though current gen is. Even if they were further back compatible, it would be a very bad idea, in a waste of money sense. If you were on a budget and had a rubbish chip, buying the top end 2nd hand chip from that generation would be a better idea. If you can afford a new top end chip, you can afford a new MB too. If you already have a top end MB that would run a new chip well, you’d already have the top end chip, just older, it’s be more of a side step, than huge leap, wait a few more years.

*Lets not forget… In audio when our project is maxing out, it pretty much maxes out for duration, it’s not something it can throttle back or just take longer to compute.

You’ve quoted me “Loads of Videos” in reference to rated performance not being achieved. That’s an incorrect quote, my “loads of videos” is in reference to videos debating, comparing VRMs. There’s also some that go on about the advantage of the heatsink placements… How some that appear to be good aren’t, due to the heat build up on the back of the board.

You not knowing about this, someone knowledgeable about PC’s, is proving my other point, up until now it’s been mostly over looked by consumers. This is why I chimed in on the subject. Lots of us are wasting big money on fancy CPU coolers when it’s the VRM’s that are failing us. The more people care about this the more the mid tier lower tier board will start to see it improve. For us needing that constant turbo, I’d say it’s very important, we need solid stable and accurate power. If we don’t care about it, newer boards could get worse (in that department) as they come up with cost saving ideas. That’s just business.

Correction: I didn’t use the word throttling in my OP.

Here’s some videos i remember watching. Not sure which is which or has what. I was looking into z390 boards around Christmas time.

That linus, what ever his name is has a few too. Not sure which is the 1 where a guy test the cool down, with the heat sinks and compares the effectiveness. I’ve already spent too long typing my reply to spend even more time hunting the specific ones.

it is a thing though. I’m not making it up.

Edit: to merge posts.

AGAIN

SEARCH VRM

I know full well that the motherboard plays a role in how the CPU is able to perform. That wasn’t what I was asking about. I was asking about specific examples of how AMD CPUs no longer performed as they should because of motherboards not being able to provide what the CPUs needed.

The first video you posted which you cued up for me didn’t say anything about “throttling” during the first three minutes of watching it, and the second was over 30 minutes long and I’m obviously not going to sit through that waiting for someone to talk about throttling on AM4 boards.

So again: Where are all these videos about AM4 setups throttling the CPU below maximum guaranteed performance?

No need for all-caps. The video I was talking about I assume was about VRM overheating.

But the point was that that’s all I could find. So far you’ve shown me nothing.

https://www.youtube.com/results?search_query=ryzen+vrm+throttling
https://www.youtube.com/results?search_query=am4+vrm+throttling

Where do we see loads of videos of these issues on AM4?

Oh dear!

I’ve not watched any videos on AM4, no where have I made that claim.

I highly doubt the MB manufactures developed perfect VRMs for all AMD board but not Intel. It’s simple electronics. If your MB has poor VRM, it will not achieve super high clock speeds on all cores for duration. Do not cheap out on your MB. Just because it will work, supported, doesn’t mean you will get all it’s glory. It’s a bad idea, apart from the variations I’ve already covered. The older the motherboard the worse the VRMs*, generally, price relative.

*Edit:Some newer could/might be worse due to manufactures saving costs, cutting corners.

I didn’t want to get locked into a debate. Hence why my first few posts were short. It’s all people need to know, be aware of how good your VRM’s are, it’s important. I wasn’t here to teach, you can do that yourself. All I did was raise a finger.

But that’s the statement I’m making, directly related to the VRM, all the ‘stuff’ related to delivering power to the CPU, and that ‘stuff’ keeping cool. The question your asking me I can’t answer, I haven’t got time to look, there might not be any specific to AMD, AMD users might not have caught on to the issue yet to start making videos about it.