What's the best way to determine "long term" maximum single core frequency?

So here is the background to the problem:

I have tasks that only run with single thread/single processors.

Previously, I had inquired about what’s the fastest single core processor and people said that it would be the Intel Core i9-9990KS because it can boost up to 5 GHz. But then I was watching some videos about it (e.g. from Steve at Gamers Nexus) and he was saying that there is a time limit as to how long it can boost up to that speed (depending on the motherboard manufacturer) before it ramps down to some other lower, maximum single threaded speed.

My question is what is the best resource to find out what that lower, long term, maximum single threaded/single core speed is?

Thanks.

Sounds like you should just look into some benchmarks for your tasks, are they common?

I’m not 100% sure for a source, but I’d imagine this information is very dependent on CPU cooler (more thermal capacity will keep the CPU cooler for extended runs), motherboard (VRM, inductor, and capacitor temperatures (cooler temp = longer boost) as well as hard-coded BIOS settings), and power supply (how stable of an output voltage can the PSU output at higher currents). Therefore, I doubt you’ll find an actual source for this information.

Keep in mind that when a CPU enters the boosted frequencies, the per-core TDP is substainally higher, and that TDP value only represents the base frequencies. Most (all?) CPU’s + mobo combos that I’ve used can sustain max single-core frequencies indefenetly, as long as the other cores aren’t in use (and the other listed factors aren’t of issue).

Many overclocking motherboards also allow you to keep the CPU pegged at 5ghz, and not allow the CPU to underclock. This will obviously cause excessive heat and power waste, but may also be a solution for you.

Start benchmark, set affinity to 1 core, benchmark - seems like a solution.

Schedulers complicate single threaded performance. In Windows XP to 7, a single thread load would use one core at 100%, with the rest idling.

In Windows 10, you can see it spread between multiple cores. So the OS itself is doing some smart stuff that imo can actually reduce single thread performance.

Yeah, my concern with this method is that Intel being Intel, if it burns out the CPU over the life of warranty, they may void the warranty simply because they’re Intel.

No, my task in and of itself, isn’t common, but that’s why I am trying to look for published single core benchmark results, but looking at the “long term” maximum single core frequency rather than the absolute peak.

For example, Steve over at Gamer’s Nexus was talking about Intel’s PL1, PL2, and tau and how some motherboard manufacturers adhere to Intel’s spec/recommendation closer whilst other just completely ignore it and run, for example, the Intel Core i9-9900KS at 5 GHz indefinitely.

So what I’m looking for is if there is a site that has benchmarked single core performance all the way until AFTER tau has expired and it settles in to a lower maximum single core boost frequency than what is commonly advertised?

(Because if you can get a benchmark to complete the task before tau expires, then it would be benched at that higher “peak” boost frequency. But I have a feeling that my single core/single “threaded” (or lightly threaded) task is going to run longer than tau, so I’m trying to find out what would be the better processor option for me.

Thanks.

And I don’t want to just build/buy a system for the purposes of benchmarking it, because if it isn’t competitive, sending everything back would just be a pain.

Chips dont degrade if you are running them stock (I havent heard of any cases) Max OC sure that can drop over time. So just look at stock numbers.

Could talk with puget systems if, they do a lot of benches and custom stuff.

The CPU does not store code that allows Intel to know that you’ve overclocked the chip. It’s possible to implement, but it currently does not do that. The only way Intel would know is by honesty. Plus, degredation of the chip mainly occurs with heat, and higher voltages. If you can get your chip to maintaim 5ghz at stock voltages and temps, you should be fine. This shouldn’t be hard to do, and if you undervolt offset the chip, that’s a bonus (decreased temps, power usage, and degredation).

No, I know that.

But for the purposes of comparing between say which would be better for single threaded performance, (e.g. Core i9-9900KS vs. AMD Ryzen 9 3950X), because of the difference in the treatment of how both of those chips turbo/boost their base frequencies, I needed a little bit more of a consistent way to be able to answer the question of if I have a single threaded task that runs for about an hour (currently), how much of a benefit will I be able to expect (calculating the expected value) by using only publically available single core/single threaded benchmarking data.

No, I know that there’s no “code” that tells them that.

But I would also suspect that if they were to inspect the die with the technology that they may have at their disposal, that they would be able to ascertain that.

I have to think that this is probably part of the reason why they would write their limited warranty contract/documention in the manner that they do because if they have the ability to reject a warranty claim, I have to assume that they have the ability to put a processor that a customer has burned out from overvolting, thereby causing physical damage to the die that you can’t see with the naked eye, but that they will have the equipment necessary to make those types of assessments.

Whether or not they actually do and/or if they do, how often they might employ is quite the unknown at this point, but as a simple end user (e.g. not a CPU engineer), these are the assumptions that I am working with here.

Thanks.

Honestly, you’re basing a lot of your assumptions on fear and uncertainty, and not on reality.

Walmart and Intel are two of the richest companies in the world. WalMart does not follow around all shoppers with a security guard to prevent theft. That would be stupid because they would end up losing more money. Instead, they employ 0-2 security guards per entrance to deter people out of stealing.

Intel WILL NOT review every chip under an electron microscope, layer for layer, to find the exact location of failure. This would be INCREDIBLY costly, and they would lose so much more money than saved (by denying warranties).

Intel may review a small sample of chips to get an idea to how they may have failed, so they can reduce the number of failures in the future (decreasing warranty claims). However, again, they have no way to know whether the end-user intentionally destroyed the chip, or whether a problem with the CPU microcode killed the chip, preexisting fault (no silicon die is identical; all will have impurities within it), motherboard, power supply, or other thing killed the chip. They won’t be 100% certain. In some instances, when they do check a small sample of the RMA’d chips, they may be confident it was user-fault, and even then, they won’t hunt down the person at fault.

Amazon, Ebay, and Aliexpress get scammed every day, and they know it. People return bricks in their packages for a full refund, pretend like they never recieved an item, or that the item is defective on arrival and just do a swap when returning. When one of these companies finds out who scammed them, they aren’t going to send lawyers after the scammers at fault (if it’s a small amount of money). Reason one being because the lawyers will cost them way more money then they’ll get back. Another reason is that they may just end up losing the court battle in the end. Instead, what these companies do is just ban your account and prevent future transactions with them.

ANYWAY, don’t you worry about Intel finding out about you overclocking a chip (or running it out of spec). If you tell them you have never heard of XMP profiles, never heard of overclocking, and tell them you used Dimm modules with a rated speed that matches the CPU, you’ll be fine, and they’ll accept a warranty, and move on.

Lastly, keep in mind Intel runs all of their CPU’s way below the deathly voltages and temps to minimize RMA’s, so you have headroom before danger presents.

Good luck :wink:


I do not condone scamming or stealing from any companies for money, their products, or other reasons. They were only used as examples. I don’t condone breaking the law. I do not take responsibility for you breaking the laws. Live your life the way you want, and keep me out of it.

Thanks.

There is 2 physical quantity, which can degrade a CPU: heat and voltage.

Case 1
If you will only use a single core, it is unlikely, you will hit a temperature limit. Pick a safe* voltage, and increase the frequency, until your system becomes unstable (Intel Burn Test is a good app for that). Then drop the frequency by a 100 Mhz, and you have a stable OC.

Case 2
If you will hammer all your cores, you will likely hit a temperature limit, before you hit a voltage/frequency limit. My advice is to keep your temperatures below 60°C, drop voltage and frequency to keep your temps down. That’s fairy low temp for a modern CPU, but the heat spreads to other components as well! Around 60°, when capacitors starts to rapidly degrade.

*For maximum single core performance, get an unlocked (K variant) 14nm Intel CPU, Skylake, or newer (i5 6600K, i3 7350K etc). The highest maximum advised voltage for these CPUs I read was 1.45V, maximum voltage for longterm use was 1.35V.

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.