PrimeGrid Challenges 2021

[Justinus]

6.6 at 1 hour 30 minutes. 4.1 ghz (linux, lscpu is all I have)

Edit: my memory is slower, but its not locking up, so I will let it run !

Interesting you're seeing 4.1, I'm only seeing 3.3. Maybe the extra memory bandwidth is giving the CPU more work to do, hitting the power limit easier?

 

[Markfw]

Interesting you're seeing 4.1, I'm only seeing 3.3. Maybe the extra memory bandwidth is giving the CPU more work to do, hitting the power limit easier?

What cooler are you using ? the temps could be slowing you.

 

[Justinus]

What cooler are you using ? the temps could be slowing you.

Quite the opposite. It's the Optimus Foundation AM4 block on an overbuilt loop (built for also having the 6900XT but there isn't a block for it yet).

My temps are around 50C

 

[Markfw]

Quite the opposite. It's the Optimus Foundation AM4 block on an overbuilt loop (built for also having the 6900XT but there isn't a block for it yet).

My temps are around 50C

Are you running windows or linux ?

I find this difference(s) very interesting. GHZ vs memory speed maybe ?

Edit: BTW, I am looking at about 9% in 2 hours, but its still estimating 24 hours total.

 

[Icecold]

Mark what are your temps like while running Primegrid? Psensor is a good utility for checking temps in Linux unless you already have another way of checking. The FMA3 work with Primegrid hits the CPU pretty hard, and before I turned on eco mode on I was seeing pretty high temps on a 3900x or 3950x even with a pretty good Noctua air cooler. I would think a 5950x would be similar.

 

[Justinus]

Are you running windows or linux ?

I find this difference(s) very interesting. GHZ vs memory speed maybe ?

Edit: BTW, I am looking at about 9% in 2 hours, but its still estimating 24 hours total.

On W10

 

[Markfw]

Mark what are your temps like while running Primegrid? Psensor is a good utility for checking temps in Linux unless you already have another way of checking. The FMA3 work with Primegrid hits the CPU pretty hard, and before I turned on eco mode on I was seeing pretty high temps on a 3900x or 3950x even with a pretty good Noctua air cooler. I would think a 5950x would be similar.

psensor only shows me GPU temps, and cpu usage, no temps. over 11% done in 2:32

and 21:45 remaining at 11% done, I think we are very close.

 

[Markfw]

a little over 20 hours to go . 17% done at 4.325 ghz

 

[Endgame124]

For sure, it looks like we started this batch of work around the same time - as of 3:47 PST I am 23 minutes elapsed and 2.040%/2.020% complete.

I'm on an Asus Dark Hero X570. I've got a kit of 2x16GB dual rank b-die 4000CL16 G.Skill Trident-Z RGB. I applied the timings from Ryzen Clock Tuner for 3800 fast b-die and then tweaked CL, secondary, and subtimings from there. It's taking 1.5Vmem so I got a memory cooler. The cooler keeps the memory at 35C at the sensors. Memtest will throw errors without it and the memory will heat up to 50C at the sensors.

Before this I was running 3800CL16 from the Ryzen Clock Tuner profile with no tweaking at 1.35 vmem and no cooler stable.

Where the heck did you get a dark hero? I gave up and got a hero viii WiFi for my 5950.

I may have to try out prime grid if it’s that much of a stress test. How long is the current competition running?

 

[Icecold]

Where the heck did you get a dark hero? I gave up and got a hero viii WiFi for my 5950.

I may have to try out prime grid if it’s that much of a stress test. How long is the current competition running?

A little under 8 days, 10 hours-

Sier"pi"nski's Birthday Challenge
Mar 14 12:00:00 to Mar 24 11:59:59 (UTC)

 

[Justinus]

Where the heck did you get a dark hero? I gave up and got a hero viii WiFi for my 5950.

I may have to try out prime grid if it’s that much of a stress test. How long is the current competition running?

I got in the original newegg preorder back when it launched. I also got a second one from Amazon in January.

 

[StefanR5R]

I got that all setup, 2x8 thread tasks and I set the affinity of each task's threads to the 8 physical cores on each CCD.

That is, you made sure that each task is running on its own dedicated CCD, did you? Spreading a task across CCDs would most certainly slow things down.

(On my dual socket computers, BDW-EP and Rome, I don't tweak affinity but simply rely on the Linux kernel's process scheduler to do the right thing.)

 

[Icecold]

My PC's are finally turning in some results. I wasn't sure why one machine that is nearly identical to another(other than a bit lower clock speed) was so much slower, and it turned out Folding@home was running on the CPU at the same time as Primegrid Now that I have that resolved, along with another issue on another machine, hopefully tasks will start processing a bit more consistently.

We're at 4th in the team stats, looking good so far.

 

[Ken g6]

Hey @crashtech, distance please!

;-)

Well, you did ask for it...

Day 2 stats:

Rank___Credits____Username
6______1550178____crashtech
9______1152592____xii5ku
11_____1070362____Icecold
24_____534735_____emoga
40_____309676_____Orange Kid
52_____232008_____Ken_g6

Rank__Credits____Team
1_____11572715___Ural Federal University
2_____6135106____SETI.Germany
3_____5555837____Czech National Team
4_____4849554____TeAm AnandTech
5_____3302665____Antarctic Crunchers
6_____2837682____Aggie The Pew
7_____2532053____BOINC@MIXI

The 4th is strong with us today.

 

[crashtech]

It's funny to me that my 8700K, formerly one of my fastest CPUs but now relegated to the HTPC, is so slow at SoB that I am not allowing it to continue after this next task completes.

 

[Justinus]

That is, you made sure that each task is running on its own dedicated CCD, did you? Spreading a task across CCDs would most certainly slow things down.

(On my dual socket computers, BDW-EP and Rome, I don't tweak affinity but simply rely on the Linux kernel's process scheduler to do the right thing.)

Yes. By default affinity they were all over the place. I put the 8 threads of each task onto the 8 physical cores of each CCD so they are each on their own CCD.

 

[Icecold]

It's funny to me that my 8700K, formerly one of my fastest CPUs but now relegated to the HTPC, is so slow at SoB that I am not allowing it to continue after this next task completes.

SoB definitely seems to perform different than most projects(probably because of the cache requirements?). Haswell Xeons perform better than I would have expected. It's nice to see them still perform well compared with newer hardware.

I know what you mean on the 8700k - have an i5 9400 that finally completed its first task but it took nearly 46 hours to do so using all 6 cores on one task. I'm guessing the 8700k is at least somewhat similar.

 

[Markfw]

They will finish in 23 hours and 15 minutes,

 

[Justinus]

They will finish in 23 hours and 15 minutes,

Mine is looking to be done at around 17.5 hours. I left the CPU on the stock config for our comparison but its going back into high gear for the next one.

 

[StefanR5R]

It's funny to me that my 8700K, formerly one of my fastest CPUs but now relegated to the HTPC, is so slow at SoB that I am not allowing it to continue after this next task completes.

Others are using single-core VMs.
example

 

[StefanR5R]

I looked up the current search range of Seventeen or Bust and checked its FFT sizes on a Haswell CPU. Haswell and all later Intel CPUs which don't implement AVX-512, and several AMD CPU architectures including Zen 2 and certainly Zen 3, will use the same FFT implementation here, which is "all-complex FMA3 FFT".

SoB-LLR candidates	current FMA3 FFT lengths	last-level cache size demand
21181*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
22699*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
24737*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
55459*2^n+1​	3M … 3200K​	24 MB … 25 MB​
67607*2^n+1​	3M … 3200K​	24 MB … 25 MB​

Meanwhile, the search space progressed enough that there are some tasks with 3,456 K FFT length out now, that is, with 27 MB cache footprint: Post. (Michael Goetz didn't specify if that's for all implementations, or AVX-512, or FMA3, or...)

 

[crashtech]

Others are using single-core VMs.
example

Well I have to wonder if one of those could even complete a main task in time to get credit for the contest.

 

[Markfw]

Mine is looking to be done at around 17.5 hours. I left the CPU on the stock config for our comparison but its going back into high gear for the next one.

It looks like memory speed impacts performance a LOT. 3200 vs 3800 takes 25% off the time.

And I think your $420 motherboard is what allows that vs my $220 motherboard.

 

[Justinus]

It looks like memory speed impacts performance a LOT. 3200 vs 3800 takes 25% off the time.

And I think your $420 motherboard is what allows that vs my $220 motherboard.

Long term (for the duration of the tasks) average effective clock was right around 3.2 GHz.

Task 1: 17:49:01
Task 2: 17:58:21

 

[Endgame124]

It looks like memory speed impacts performance a LOT. 3200 vs 3800 takes 25% off the time.

And I think your $420 motherboard is what allows that vs my $220 motherboard.

I spent about 2 days worth of work tightening the timings and adjusting the voltages for my 5950 on my Crosshair hero viii (wifi). The default settings w/ XMP enabled used way, way too much voltage on the memory controller, which was unstable at higher clockspeed. Dropping the SOC voltages allowed me to get to 3866 stable, and 4000 mostly stable. My guess is that 4000 would be stable if I had less than 64GB ram, but I don't have extra ram (or the patience) to buy a smaller kit to see if it changes the stability at higher clocks.