You may have seen my "CPU App performance" thread comparing Linux app performance vs Windows.

Now I hope to find out what the Ryzen vs Intel performance is on this application on both Linux and Windows. Is it simply the difference in IPC between them or is there something greater that could affect the performance?

you may find some of the stats here
https://boincstats.com/en/stats/14/host/breakdown/cpu/

has anyone seen any i7 8700k stats running all 6 cores 12 threads?

You can test yourself the EPYC AMD processor
24 cores / 48 threads

https://www.zdnet.com/article/amd-and-packet-kick-off-epyc-processor-challenge/

There's a $250 credit with Packet which gives you 10 days @ $1.00 an hour ($240 = 240 hours or ten days).

Make sure to delete your "bare metal" server so you don't incur server holding costs.

You may have seen my "CPU App performance" thread comparing Linux app performance vs Windows.

Now I hope to find out what the Ryzen vs Intel performance is on this application on both Linux and Windows. Is it simply the difference in IPC between them or is there something greater that could affect the performance?

IPC is one high level metric that allows you to compare whether one particular machine is executing instructions of one particular program faster than another particular machine.
Another thing you want to know is what is "happening" to stall the execution.

One very large application that I worked on spent "on average" 60% of EACH CYCLE waiting on data values to be returned from memory to complete a stalled instruction.
My time was spent analyzing and reorganizing the code and data memory footprint to reduce that 60% stall.

When you measure IPC on different machines, you really have to have similar instruction sequences to compare. With the Rosetta design and the way WU are run, that is not very practical.

One very large application that I worked on spent "on average" 60% of EACH CYCLE waiting on data values to be returned from memory to complete a stalled instruction.
My time was spent analyzing and reorganizing the code and data memory footprint to reduce that 60% stall.

Have you had more contacts with developers?
Don't you know if they have implemented your suggestions?

---

One very large application that I worked on spent "on average" 60% of EACH CYCLE waiting on data values to be returned from memory to complete a stalled instruction.
My time was spent analyzing and reorganizing the code and data memory footprint to reduce that 60% stall.

Have you had more contacts with developers?
Don't you know if they have implemented your suggestions?

I communicated with David K. several years ago and I think he ran some experiments with compilers and options. As I remember, I expected a 40% improvement and David said he saw a 100% improvement on his large machines with the Intel compiler and my options. It has been awhile, but I think I just added SSE2 (which would require them to handle directing WU to SSE2 and not 386 machines, like they steer WU Android today) ... and I eased up on their aggressive inline options.

The developers did not believe my estimates and had no time/interest in adding work to their list and making changes. The source code that I have access to still seems to have no changes.

The developers did not believe my estimates and had no time/interest in adding work to their list and making changes. The source code that I have access to still seems to have no changes.

It's a pity....

---

What can we do to encourage the project team to take a closer look at these suggestions? I have 2 of the top 10 machines on the project right now & both are powered by AMD Opteron. I assume these changes will work with the Opteron processors as well as they support the extended instruction set. It would be great to get more work out of these systems and get to treatments faster.

It is shocking how much compute power we can buy for $500. Now we need to figure out how to efficiently use all of it.

---

Thx!

Paul

It is shocking how much compute power we can buy for $500. Now we need to figure out how to efficiently use all of it.

Yes, we far outperform an ILLIAC IV, if you remember what that is.
https://en.wikipedia.org/wiki/ILLIAC_IV

But I would like it used efficiently too. If they need more performance, all they have to do is go get it.

---

There haven't been many posts in here. Sorry for the dead thread post and all...

The Ryzen 2600X is producing about 10,500 credits per day.

What seems to matter a lot are RAM speed and timings. I'm testing those now on the AMD systems. Going from 2133Mhz CL 16 to 2666Mhz CL16 produced almost a 5% gain in average credits over a 9-day period. I'm testing 3000Mhz at CL 15 now.

The Intel 6700k with 3000Mhz 15CL RAM is doing around 8,750 per day.

There haven't been many posts in here. Sorry for the dead thread post and all...

The Ryzen 2600X is producing about 10,500 credits per day.

What seems to matter a lot are RAM speed and timings. I'm testing those now on the AMD systems. Going from 2133Mhz CL 16 to 2666Mhz CL16 produced almost a 5% gain in average credits over a 9-day period. I'm testing 3000Mhz at CL 15 now.

The Intel 6700k with 3000Mhz 15CL RAM is doing around 8,750 per day.

You are increasing speed AND tightening the timings at the same time. Ugh, I really want to build my own setup with premium RAM and play like this :(

I want to wait until the 3rd generation Ryzen tho.

---

You are increasing speed AND tightening the timings at the same time. Ugh, I really want to build my own setup with premium RAM and play like this :(
I want to wait until the 3rd generation Ryzen tho.

Black Friday and eBay sales :) . I'm only running 8GB in each of my Ryzen boxes. There was a 4x4GB kit for about $100 on eBay when they had a 15% off sale. That's when I went 3000Mhz at CL15. The set was for older Ripjaws 4 memory. Not sure if it matters as the timings and MHz of newer memory hasn't improved much.

The shift to that setup happened on the 23rd. It doesn't seem to have had much of an effect over 2666Mhz at CL16... but it is too early to tell.