Question Help me understand base clock, boost clock & undervolting for specific use case (DSP software development)

[MvKeinen]

Hi, first post, great forum! I'll try to be as brief as possible.

My goal is to know what CPU I need.
I have difficulties understanding the impact of base clock and boost clock in my use case which is:
- DSP software development (Digital signal processing, I'm developing software synthesizers): This is an application which is highly reliant on real time computing. You hit a key on the musical keyboard or use a microphone to input sound and everything that comes after that can only be calculated in a serial manner. The DAW (Digital audio workstation, software to produce and mix music) takes it in and sends it inside a track (or channel) through different plug ins that each have to "wait" for the result of the calculations of the earlier stages. Everything that happens inside said track can not be distributed on different cores.

those latency figures can be halved in a modern computer provided it has excellent single core performance. One sample here means one tick of the sampling clock (I'm using 96k)
BUT: Near silent operation is also very important because the PC will be located in a music studio.
And MT performance is also important because a normal project in my case takes 12-15 tracks and those tracks can be placed on different cores by the DAW.

priorities are as follows:
Stabillity: 100%
near silent op: 100%
ST perf.: 99%
MT perf.: 80%
GPU power: 10% (no gaming, iGPU is perfectly fine but will add to CPU temp)

While prototyping my DSP stuff I sometimes have to scale up number of voices and/or sampling rates to ridiculous numbers in order to be able to estimate efficiency of the code. So I'll always hit the ceiling of ST performance no matter what. In these moments I'll need the boost clock I guess. The problem here is that I don't know how long that boost clock can last. And if it can last forever provided the CPU temps are ok, what is the significance of the base clock figure?

thanks a lot for reading and thanks in advance for any tip, pointer, knowledge or reference you can provide me. I'll leave it to that for now but there are a lot of further questions.
Cheers.

 

[coercitiv]

There was time when boost clock was easier to explain, because it had rules. Nowadays, the best definition I can give is that boost clocks last as long as the motherboard was configured to keep them, provided you're not hitting power, current or temperature limits. Base clocks are easier to understand, they are the minimum frequency that the CPU will hit under the most intense workloads, "real-world" clocks will usually be higher than that.

Here's the good news for you, you can configure everything starting from your noise target and going back to your performance target. I'll walk you though a possible scenario where the aim is to get the most out of a CPU.

Pick your cooling solution and a case with good airflow​

Here you can consider the Noctua NH-P1 for example, though you can also opt for any other high performance tower cooler that can work well with low RPM fans. I have the Noctua NH D14 fitted with just 1 of the stock fans, and it can cool 125W worth of power while the fan stays under 1000RPM. How much you're willing to let the fans spin, if at all, is your choice here. You can also consider water cooling, though that means the minimum noise threshold will be higher, while max threshold should be lower.

The cooling should be configured in such a way that noise is always under control. For example, should you opt for NH-P1 which is built as a passive cooler, you can still add a fan of your choice that will work at a very specific low RPM, or add 1-2 case fans that are still creating a directional airflow and helping the cooler. This will create your baseline for max performance, a silent enclosure that can probably sustain 65-90W of continuous power while also being able to absorb 100W+ for short bursts of 10-30s.

​

Set a temperature limit and a power limit​

The temperature limit will be your best option to maximize the ST - MT - Noise triangle. You will configure a temperature limit and then allow the CPU to hit that limit consistently, since you'll be using a CPU that is capable of pulling well over 100W in a setup cooled for 60-90W. The power limit will also be customized, since you want to let the CPU work in higher power bursts, but not so high power that all the thermal capacity of your cooler is exhausted in just a few seconds. For example you can set a 120W power limit. I won't go into details on what that means for Intel and AMD, all you need to know now is that both platforms allow you to customize max temps and max power.

For temperature limit my suggestion would be something like 75-85C, the idea is to offer some added protection for long-term operation while still letting the CPU run at high temps, because high temps also improve your cooler's performance. (heat transfer is directly proportional with the temp delta between two materials)

If you opt for Intel's Alder Lake / Raptor Lake, I strongly suggest buying and installing a contact frame. (the one from Thermal Grizzly or the one from Thermaright)

Set max clocks​

I assume your aim will be to buy one of the higher end CPU from either Intel or AMD. In the case of Intel my advice would be to shave 200-300Mhz from the max boost clock intended for 1-core operation. Like you already said, you'll always hit the ST performance limit with your workload, the frequency dip will limit heat waste though, and keep some of that thermal capacity for other bursts. Max clocks for MT workloads won't matter as much, as they will be dictated by your power and temperature limits instead (which is a good thing).

Undervolting​

Not something you should consider when planning your new system. All you need to know is that both platforms (Intel/AMD) offer modern ways to undervolt, and you'll be able to optimize in this regard after your setup will be complete and stable. Undervolting will buy you more perfomance in MT scenarios and better temps in ST scenarios, but spend your focus and energy on the cooling solution, that will matter more.

Build example​

A few years ago I built a system using an R5 1600X and a Scythe Ninja 2, a cooler from a different era. That cooler was designed as a hybrid, meaning it's fins were spaced apart to work well in passive operation. The system only had 1 fan, placed at the back of the case, in close proximity to the cooler, with low fixed RPM. The cooler therefore worked in semi-passive config.

The R5 1600X is a six-core CPU with 95W TDP. I did not change any setting for the CPU other than the target temperature, which I think I configured for 75C. The result was a system that boosted to the stock 4-4.1Ghz for light bursts, and gradually dropped clocks towards base clocks as it ran under heavy MT loads. IIRC it was able to sustain 3.5-3.6Ghz, whereas stock would have been 3.7Ghz. This could have been tweaked further by adjusting the speed of the fan (or changing the fan with a better one). Undervolting would have helped as well.

All I did was set a temperature limit for the CPU and chose the appropriate cooler for the job. Power management in modern PCs did the rest.

 

[eek2121]

I did some testing with my 7950X a while back and IIRC the best way to limit the CPU was to simply limit the fan curve. Doing it any other way would significantly hamper overall performance. I may revisit this in the future and post some real numbers.

 

[MvKeinen]

Awesome! Just quickly wanted to say thanks for the great info. Will have more time in the next days to answer properly as there are some follow up questions. As I'm not native English speaking I need more time to write.