Blogs Which benchmarks do you want to see with 3rd Gen Ryzen?

Yes, they specifically switched from Thread Expansion (keeping active cores as far as possible from each other to prevent thermal throttling) to Thread Grouping (keep neighbouring threads on the same complexes), to minimise the problem Zen/Zen+ had with inter-CCX latency.

 

If you are willing to sacrifice your sanity for the job and dabble in Linux...

https://github.com/opsengine/cpulimit

 

It is about heat and power consumption to get an idea of the curve between idle and full load.

Some chips get really, really hot under all core full load whilst remaining cool at only half load. Some chips get hot early, some later. The curve might give some idea of this rather than the jump from idle to full load.

As more more people become aware of the fact that heat equals power usage and potentially noise, I was trying to find some way of being able to benchmark something in between idle and full load. Benching "average or non gaming" usage is beset with pitfalls but, if possible with the tools available, measuring at 50% load (or even other steps in between) may be a way to give a better idea.

At present when looking at charts of heat, power consumption and noise on any site, we have to guess what the figures would be with "average" usage - browsing, office work, multimedia, graphics work etc. As I said above, measuring this kind of usage is a minefield but "stepping" the CPU utilisation would move us in the right direction

 

According to this (very old) page, a Microsoft engineer used a Sysinternals CPU Stress tool to limit thread usage to sumulate 50% CPU load, not sure if useful?

https://blogs.msdn.microsoft.com/vijaysk/2012/10/26/tools-to-simulate-cpu-memory-disk-load/

Would limiting a process to 50% threads achieve it? Could you also limit it to 25% and 75%?

Edit: Unfortunately this tool doesn't do what I was hoping, I just tried it. Yes it can limit threads but (i) it only shows 4 threads, even on my i5 8400 and (ii) it doesn't make the CPU work enough to make it hot, resource monitor shows 25%, 50%, 75% and 100% utilisation but my CPU temp only got to 16C above idle, meaning it wasn't actually working hard.

Edit 2: I tried playing with Prime, oddly enough, if I run a small FFT torture test using 1 thread (my CPU has 6), it shows 26-27% CPU utilisation in resource manager, 49% with 2 threads, 72% with 3 threads, 92% with 4 threads, 114% with 5 threads and 135% with all 6 threads. Is this Turbo kicking in and resource monitor not understanding it?

HWMonitor does show the test only utilising the correct number of cores and it does actually achieve what I was looking for. Not sure how accurate it is though.

 

That should be very interesting then. If BT has the budget for an IR camera (back of motherboard temperature measurement) or the Zen 2 parts report per chiplet temperatures that would be very interesting to look at. What is considered "neighboring threads"? Any desktop OS spawns dozens of threads so if a program, say a web browser, spawns 2 threads will they be allocated to the same core which handles 2 threads via SMT or to two different but physically adjacent cores? I have to imagine AMD (Microsoft too) took some lessons learned from Bulldozer in regards to CMT and thread placement to assign related threads to locations where they can share resources without stalling.

 

Vulcan based games.

 

Even Glide, NVlib, ATICIF, and S3D?

 

The new Ryzens running X570 vs X370 motherboards.
Vs 1. Gen Ryzen ie. Ryzen 7 1700.

Running battlefield 5.
Preferred resolution 1440p.