Precision 5820 i9 10980XE Current/EDP Limit Throttling

Adding to this, it seems that in Cinebench or Furmark's CPU Burner if I use Intel XTU to max out the current and power limits, there is no Current/EDP Throttling and the single core turbo gets to around 4.7Ghz where I'd expect, and the all core turbo to 3.8Ghz which is also correct.

I suspect an AVX-512 workload might trigger the current throttling but with the above said... Why doesn't Dell release a new BIOS that adds more headroom so that we can utilize these powerhouse CPU's.

I have been in touch with Dell support. They have requested some information which I have provided in the form of a video, description and diag results.

I have been told this is being escalated to their product engineering team.

I am not holding my breath but seeing as this is eseentially an issue for trading standards, if I don't get a resolution soon I will escalate this to every organisation or tech media outlet that needs to know Dell is selling workstation configurations which do not operate as the hardware was intended due to their oversight and artifical limits imposed.

And yes, it looks like AVX turbo boost is locked down since the motherboard is too weak. A feature designed by intel for thin laptops with a weak cooling. If you run a heavy math in Excel, which does not use AVX, there is no problem with full load. Yes, I believe it is a dirty trick used by the devs for better cost-down; in believe that in corp. market the response to performance issues is not as strong as in enthusiast market.

But we buy these systems for high-performance computation!

It was not fixed. They offered full replacement, including HDDs. Obviously, it is not acceptable in corp. segment; that's why we use removable HDDs with a lock. Maybe that was a trick to avoid solving the problem.

I am in discussions with a number of people from Dell trying to escalate this issue. I don't know what the outcome will be. In the ideal world a new BIOS that removes the current/power limits. However if it ends up being a hardware limitation then Dell have an issue on their hands...

This would be one for trading standards as they are selling configurations which are not able to perform as designed due to their own product engineering oversights.

Intel's TDP ratings on CPU is a strange one. 165w rating but it will exceed 250w under load in stock factory conditions. What is the point of that rating.

Dell has been supplying incomplete products for years, it's a wide-spread practice across the market (not only Dell).
If they do choose to make it right (from engineering standpoint) - then there will be significant profitability drop due to extra costs:

1) Assembly lanes effort + components (provisioning/outsourcing with partners/third-parties, warehousing/logistics/supply chain planning and lead times, etc.).

2) Higher warranty case numbers (not only via natural overload but via improper user settings causing system failures)

3) Increased number of defects (including during transportation/logistics, e.g. vibrations/G spikes)

4) Service and maintenance efforts (not just cases count but complexity too {required qualifications}).

That's the classic example when engineering perfection is slashed by commercial viability.

Their first priority in many cases is stability of the system, then performance (secondary objective).
Whenever first is at risk - second is sacrificed.

I have a feeling this might be fixable via a new BIOS.

I've also found that if the workstation's cooling system is limiting the overall temperature, why sell the K version of the CPU with a configuration that doesn't allow it to be used to its full potential?

Now the CPU is only slightly higher than the base frequency at full speed.

So Dell have referred me to this:

Intel i9 Processor Throttling Under AVX (Advanced Vector eXtensions)
This is working as designed based on the architecture of the Intel i9 processor. 
Having a long history in performance computing/overclocking and benchmarking I do understand this. However the root of this problem is probably due to Intel not being clear with their Turbo Boost technology, TDP ratings and AVX instructions.

AVX Instructions are getting common for big workloads now. So what is the point of giving a CPU a 165w rating, if under stock factory conditions, when loaded up with an AVX workload it will pull in excess of 240w.

Dell are enforcing a 165w power max with a burst up to around 187w.

These are the words from Dell:

Our product engineers have performed an investigation and the article published, is of our findings. The date of publishing was 29-Mar-2021 and there are no open investigations on this topic. There is no further discussion on making any changes, as the platform is performing well globally. Here is what our engineering team have reported:

CPU is running 100% AVX stress, causing it to reach power limits and reduce its frequency under load.
5820 is delivering adequate power at 165W steady state with a peak of 187.83W
5820 is delivering adequate cooling and the CPU never exceeds 81C (significant thermal margin)
It definitely drops to 2.4-2.5GHz but that’s because this is an AVX workload. The CPUs have an AVX flag that, whenever Advanced Vector eXtensions are used, the CPU drops to lower frequencies than could be otherwise supported without AVX instructions. This is because AVX instructions use wider registers, wider registers require higher power to switch than 64-bit registers, thus to stay within instantaneous current limits, the CPU selects a lower operating frequency for stability.
AMD and Intel CPUs do not adjust CPU frequency using the same algorithms. They are different silicon and different techniques to manage their unique balance of power, temperature, and operating frequencies.
Intel CPUs run for a period of time at high frequency and then roll off to a lower frequency at steady state. This is how they’re designed. The frequency at which the CPUs rolls off to is a function of many factors, one is the workload. AVX workloads result in a more significant drop in frequency.

While using the benchmarking software that you have, this is where you will encounter throttling most frequently, the article suggests if benchmarking is required, either disable AVX workloads in testing software, or use a stress test that relies on non-AVX workloads.

My findings so far are that these power limits can be raised to stop such aggressive throttling using Intel XTU, however there is a hard current/EDP limit when you really hit the CPU hard with AVX-512 for example.

However at this point it is also clear that the Dell thermal solution aka the rubbish little heatsink and fan on the CPU under the shroud is not adequate to sustain such loads at high frequencies for a length of time.

I suspect installing a better heatsink such as a Noctua U12-DX-i4 might help with that.

https://noctua.at/en/nh-u12dx-i4/specification

See update:

https://www.dell.com/community/Precision-Fixed-Workstations/Precision-5820-Xeon-W-2245-Current-EDP-limit-throttling/m-p/8205596/highlight/true#M10700

I have detailed this issue in a video here:

https://youtu.be/kRMS0Ka1FTQ

I rarely experience current/EDP limit throttling (about 10%) with the 3650 i9-11900K, but thermal throttling often occurs at no load (30~50%) due to space issues limiting the effectiveness of the heatsink. It operates at a high temperature of 90 degrees for a long time, and the core clock is only 39x at the highest speed.

I'm very disappointed that this tiny chassis design shouldn't sell an overclocked CPU at all, which obviously doesn't take advantage of the CPU's performance.