FX9590 vs GLeeM's I7-4790K on NFS

[Rudy Toody]

GLeeM's I7 does a WU in approximately 3200 seconds.
My 9590 does the same type of WU in approximately 6800 seconds.

That's quite a difference! I am not overclocked; stock is 4700MHz. I am testing now with memory set to 2400MHz. That's a 50% speed up. I don't think it will make up much of the difference.

An suggestions for tweaking specifically for NFS. I have 32GB memory, so I could run multiple threads. The core usage is currently shown to be a steady 100%, so I wouldn't expect much improvement.

 

[TennesseeTony]

Sorry to hear that Fred. I was hoping I was wrong, but having built a FX-8350 4GHz in the last six months, I feared the 9590 would likewise be somewhat of a disappointment. My 8350 performed just a little bit better than my ancient 1st gen i7 at stock speed. :\

 

[Rudy Toody]

The memory tweak actually took longer! So, I am back with Leiden and EDGES. I'm going to research the memory bandwidth with two sticks instead of four. Sometimes when you fully populate the ram you will cut the bandwidth in half.

There is also a bug in the mobo. I get a message at boot to update the BIOS to fix a firmware bug in power now (what ever that is). That is something else to research to see how it might affect over-clocking.

 

[Fardringle]

I believe that AMD's Power Now is basically the same thing as Intel's Cool and Quiet (adjusts the CPU speed to reduce heat and power consumption when it's not needed).

 

[GLeeM]

I noticed the same thing this morning.

I think all recent WUs (on all three computers) were with only three or less NFS WUs running, the other threads had a different project running. I feel I get better use of hyperthreading if I mix up the projects running at the same time.

You might be onto something with the memory. But I don't know what.
My 4770K @ 4.1 has 8 GB of 1866 Mhz ram.
My 4790K @ 4.4 has 16 GB of 2133 Mhz ram.
But it seems the 4770K does a little better!?!

 

[TennesseeTony]

What work units are you timing? My 16e's are all over the place from 3000 to 4350 seconds.

<EDIT> I should note that this is with HT enabled. </EDIT>

 

[GLeeM]

My 16e's are all over the place from 3000 to 4350 seconds.

So are ours. We average them out. Or glance at the times, eliminate high and low, then estimate, you can get pretty close.

 

[Fardringle]

I just checked, and my i5 4590's are averaging around 2500-2600 seconds per work unit, ranging anywhere from 1400 to 3500. My guess is that the project just isn't optimized very well for AMD CPUs.

 

[Rudy Toody]

I noticed the same thing this morning.

I think all recent WUs (on all three computers) were with only three or less NFS WUs running, the other threads had a different project running. I feel I get better use of hyperthreading if I mix up the projects running at the same time.

You might be onto something with the memory. But I don't know what.
My 4770K @ 4.1 has 8 GB of 1866 Mhz ram.
My 4790K @ 4.4 has 16 GB of 2133 Mhz ram.
But it seems the 4770K does a little better!?!

Initially, I had all BOINC settings as they were for earlier boxes. This had memory settings which were too low, thus preventing use of all the ram.

Intel (I think) has multiple memory buses to handle the full population issue. That might account for the big differences.

Now I can't run Leiden because of a missing shared library which didn't get installed during my last Linux install. Right now it's just Edges.

Cool and Quiet is something I always turn off, so I'll see if I can turn off Power Now.

 

[Rudy Toody]

Oops! I just dropped a rubber band into my new build and now it's making snap decisions!

 

[Fardringle]

So now it's running BOING instead of BOINC?

 

[Rudy Toody]

So now it's running BOING instead of BOINC?

That explains the sounds I'm hearing!

 

[biodoc]

Now I can't run Leiden because of a missing shared library which didn't get installed during my last Linux install. Right now it's just Edges.

I had the same problem. Installing the freeglut3 package worked for me.

In the project directory run in terminal:

ldd classical_5.56_x86_64-pc-linux-gnu.exe

The output will tell which libraries are missing

 

[Rudy Toody]

Leiden Classical is up and running, sharing time with Edges.

 

[Bradtech519]

Glad to see you got the FX 9590 running. AMD seems to still do decent in pogs & universe@home. Abysmal compared to I7 in other projects when I compare my FX 8350 to 2/3/4/5k intels.

 

[biodoc]

Glad to see you got the FX 9590 running. AMD seems to still do decent in pogs & universe@home. Abysmal compared to I7 in other projects when I compare my FX 8350 to 2/3/4/5k intels.

Add Edges@home to that list. My 1090T is faster per core than my intel processors.

 

[GLeeM]

Add Edges@home to that list.

Looks like SubSetSum can be added to that list, too. And they have WUs!

If I remember right SubSetSum used to pay well, but now only low to average on Intel 🙁 and average on AMD.

Edit: be careful with Citizen Science ... the Wildlife project pays well but I expect the WUs download size to be hundreds of MBs (maybe?) 😱 You can unselect it in Preferences.