Overclock.net › Forums › Intel › Intel CPUs › Broadwell-E thread
New Posts  All Forums:Forum Nav:

Broadwell-E thread - Page 89

post #881 of 7252
Quote:
Originally Posted by DarkIdeals View Post

Do you think you could run the test with just 6 or 8 cores active on the 6950X at 4.4 / 3.7 speed? That would give a better idea of how it compares to an equivalent core count or higher core count Haswell-E chip.
"for_loop"? That is a completely single-thread function, I'd put money on zero impact to disabling cores for that. Or did you mean something else?
post #882 of 7252
Quote:
Originally Posted by cekim View Post

"for_loop"? That is a completely single-thread function, I'd put money on zero impact to disabling cores for that. Or did you mean something else?


Oh, sorry. I had no idea what the specific test you were running was. I thought it was a multi-threaded one. I'm not familiar with the list of numbers you posted, so i had no idea what program/test it was.

That's interesting though. That test is completely single threaded then? That's a quite nice indication of overall IPC then! Good to know...
 
CPUMotherboardGraphicsRAM
i7 5960X ASUS Rampage V Extreme MSI GTX 1080 Gaming X 16GB Corsair Dominator Platinum 2666mhz DDR4 C15 
Hard DriveHard DriveOptical DriveCooling
PNY CS2211 MLC SATA III SSD WD Blue 500GB 7200rpm Writemaster DVD/CD +/- RW EK Supremacy EVO 
CoolingCoolingCoolingCooling
Mayhem's Pastel Ice White Coolant Coollaboratory Liquid Copper TIM XSPC EX480mm Radiator Black Ice GTX 360mm Radiator 
CoolingCoolingCoolingCooling
EK XTC 420mm Radiator Swiftech MCP655-B 12v Pump EK RES X3 250 Reservoir Bitspower/EK 3/8" x 1/2" Compression fitting 
OSMonitorMonitorKeyboard
Windows 8.1 ASUS ROG Swift PG278Q 27" 1440p 144hz G-Sync ASUS VG23AH 23.5" Passive 3d 1080p 60hz  Razer Blackwidow Chroma Tournament Edition 
PowerCaseMouseMouse Pad
EVGA Supernova G2 1000 Caselabs SMA8 -XXL Window, Ventilated sides/t... Razer Ouroboros Norman Rockwell collection series print Mousepad 
AudioAudio
Sennheiser HD700 300ohm Open Back Headphones SupremeFX Hi-Fi 5.25" Bay AMP/DAC  
  hide details  
Reply
 
CPUMotherboardGraphicsRAM
i7 5960X ASUS Rampage V Extreme MSI GTX 1080 Gaming X 16GB Corsair Dominator Platinum 2666mhz DDR4 C15 
Hard DriveHard DriveOptical DriveCooling
PNY CS2211 MLC SATA III SSD WD Blue 500GB 7200rpm Writemaster DVD/CD +/- RW EK Supremacy EVO 
CoolingCoolingCoolingCooling
Mayhem's Pastel Ice White Coolant Coollaboratory Liquid Copper TIM XSPC EX480mm Radiator Black Ice GTX 360mm Radiator 
CoolingCoolingCoolingCooling
EK XTC 420mm Radiator Swiftech MCP655-B 12v Pump EK RES X3 250 Reservoir Bitspower/EK 3/8" x 1/2" Compression fitting 
OSMonitorMonitorKeyboard
Windows 8.1 ASUS ROG Swift PG278Q 27" 1440p 144hz G-Sync ASUS VG23AH 23.5" Passive 3d 1080p 60hz  Razer Blackwidow Chroma Tournament Edition 
PowerCaseMouseMouse Pad
EVGA Supernova G2 1000 Caselabs SMA8 -XXL Window, Ventilated sides/t... Razer Ouroboros Norman Rockwell collection series print Mousepad 
AudioAudio
Sennheiser HD700 300ohm Open Back Headphones SupremeFX Hi-Fi 5.25" Bay AMP/DAC  
  hide details  
Reply
post #883 of 7252
Quote:
Originally Posted by DarkIdeals View Post

Oh, sorry. I had no idea what the specific test you were running was. I thought it was a multi-threaded one. I'm not familiar with the list of numbers you posted, so i had no idea what program/test it was.

That's interesting though. That test is completely single threaded then? That's a quite nice indication of overall IPC then! Good to know...
The third number is "wall-clock" time in (minutes:seconds:milli-seconds) format for comparison sake.

Here's another contrived test with threads - feel free to critique or find flaws that might produce non-linear results in my code, I really just smashed this together right now, so I am sure its less than perfect:
Code:
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <pthread.h>

#define TOTAL 10000ull
#define INNER 10000000ull

void *thread_func(void *ptr)
{
    uint64_t *addr = (uint64_t *)ptr;
    volatile uint64_t a=1,b=1,c=1;   
    while(1) {
        for(uint64_t i=0;i<INNER;i++) {
            a = (a + b * c)*i;
            c = a + b;
            b = c + a;        
        }
        
        uint64_t prior = __sync_add_and_fetch(addr,1ull);
        if(prior >= TOTAL) {
            printf("Thread %lld complete\n",pthread_self());
            printf("a %llu b %llu c %llu\n",a,b,c);
            pthread_exit(0);
        }
    }
}

int main(int argc, char **argv) 
{
    int thread_count = strtol(argv[1],NULL,10);
    printf("Running with %d threads\n",thread_count);
    
    pthread_t *thread_pool = new pthread_t[thread_count];
    volatile uint64_t common_counter = 0;

    for(int i=0;i<thread_count;i++) {
        pthread_create(&thread_pool[i],NULL,thread_func,(void *)&common_counter);
    }

    for(int i=0;i<thread_count;i++) {
        pthread_join(thread_pool[i],NULL);
    }
    printf("All threads complete (counter = %llu)\n",common_counter);
    exit(0);    
}

5960x @ 4.4/4.0
time ./multi_thread 8
All threads complete (counter = 10007)
616.657u 0.011s 1:17.17 799.0% 0+0k 24+0io 0pf+0w

time ./multi_thread 16
Running with 16 threads
All threads complete (counter = 10015)
675.358u 0.011s 0:42.28 1597.3% 0+0k 0+0io 0pf+0w

*sanity check, this should certainly NOT be faster and likely slower than 16 on this CPU*
time ./multi_thread 20
Running with 20 threads
All threads complete (counter = 10019)
675.818u 0.005s 0:42.32 1596.9% 0+0k 0+0io 0pf+0w
*yep*

6950x @ 4.4/3.7
time ./multi_thread 8
All threads complete (counter = 10007)
549.889u 0.018s 1:08.82 799.0% 0+0k 24+0io 0pf+0w

time ./multi_thread 16
Running with 16 threads
All threads complete (counter = 10015)
603.635u 0.026s 0:37.80 1596.9% 0+0k 0+0io 0pf+0w

time ./multi_thread 10
Running with 10 threads
All threads complete (counter = 10009)
549.767u 0.010s 0:55.05 998.6% 0+0k 0+0io 0pf+0w

time ./multi_thread 20
Running with 20 threads
All threads complete (counter = 10019)
622.426u 0.059s 0:31.38 1983.6% 0+0k 0+0io 0pf+0w
Edited by cekim - 6/29/16 at 7:42pm
post #884 of 7252
Quote:
Originally Posted by cekim View Post

Indeed... Intel obviously focused on specific instructions and work-loads that it improved... So, you really have to try stuff to know.

Out of curiosity when I got it, I did things like this:
Code:
// compile with g++ -O3 for_loop.cc -o for_loop
#include<stdio.h>
#include<stdlib.h>
#include<stdint.h>

int main(void)
{
    volatile uint64_t a=1,b=1,c=1;   
    for(uint64_t i=0;i<10000000000ull;i++) {
        a = (a + b * c)*i;
        c = a + b;
        b = c + a;        
    }
    printf("a %llu b %llu c %llu\n",a,b,c);
    exit(0);
}

5960x @ 4.4/4.0 (didn't do the 4.7 machine, but this sort of work load is 1:1 linear)
time ./for_loop
a 14112393696577847627 b 9059408354752924673 c 13393758731884628662
61.686u 0.004s 1:01.75 99.8% 0+0k 0+0io 0pf+0w

6950x @ 4.4/3.7
time ./for_loop
a 14112393696577847627 b 9059408354752924673 c 13393758731884628662
54.818u 0.001s 0:54.87 99.8% 0+0k 24+0io 0pf+0w

^^Nice post, I'd be interested to see any other functions you tested and the results.

As for my build, it seems my so called "stable" 4.4Ghz overclock is not doing very well. It failed OCCT pretty much instantly. I pushed back to stock cache settings, RAM back to CR2, and 43x multi and now OCCT fails due to the cpu going over 85C.

I feel like my cooler isn't doing its job properly. I re-seated it today before I ran these tests, so that's definitely not the problem... Corsair Link reports my pump is pushing 2800rpm but my fan is only at 780-820? I have it configured in CL at fixed 100% speed. Corsair's website says these fans should go up to 2500rpm?
    
CPUMotherboardGraphicsRAM
Intel 6900k Asus x99 Deluxe II NVidia 980 Ti G-Skill F4-3200C14Q-64GTZSW 
Hard DriveCoolingPowerCase
Samsung Evo 950 PRO m.2 Corsair h100i v2 CLC  EVGA 220-G2-0750-XR NZXT s340 white 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel 6900k Asus x99 Deluxe II NVidia 980 Ti G-Skill F4-3200C14Q-64GTZSW 
Hard DriveCoolingPowerCase
Samsung Evo 950 PRO m.2 Corsair h100i v2 CLC  EVGA 220-G2-0750-XR NZXT s340 white 
  hide details  
Reply
post #885 of 7252
Quote:
Originally Posted by vibraslap View Post

^^Nice post, I'd be interested to see any other functions you tested and the results.

As for my build, it seems my so called "stable" 4.4Ghz overclock is not doing very well. It failed OCCT pretty much instantly. I pushed back to stock cache settings, RAM back to CR2, and 43x multi and now OCCT fails due to the cpu going over 85C.

I feel like my cooler isn't doing its job properly. I re-seated it today before I ran these tests, so that's definitely not the problem... Corsair Link reports my pump is pushing 2800rpm but my fan is only at 780-820? I have it configured in CL at fixed 100% speed. Corsair's website says these fans should go up to 2500rpm?
I posted a quick multi-thread contrived test, but it seems it got moderated... The other stuff I do, I am afraid I can't share.
post #886 of 7252
Quote:
Originally Posted by cekim View Post

I posted a quick multi-thread contrived test, but it seems it got moderated... The other stuff I do, I am afraid I can't share.


Oooh, secret agent stuff.....spooky... http://www.youtube.com/watch?v=hdl9Byw3ej8&t=0m13s

Tell us! poke.gif
 
CPUMotherboardGraphicsRAM
i7 5960X ASUS Rampage V Extreme MSI GTX 1080 Gaming X 16GB Corsair Dominator Platinum 2666mhz DDR4 C15 
Hard DriveHard DriveOptical DriveCooling
PNY CS2211 MLC SATA III SSD WD Blue 500GB 7200rpm Writemaster DVD/CD +/- RW EK Supremacy EVO 
CoolingCoolingCoolingCooling
Mayhem's Pastel Ice White Coolant Coollaboratory Liquid Copper TIM XSPC EX480mm Radiator Black Ice GTX 360mm Radiator 
CoolingCoolingCoolingCooling
EK XTC 420mm Radiator Swiftech MCP655-B 12v Pump EK RES X3 250 Reservoir Bitspower/EK 3/8" x 1/2" Compression fitting 
OSMonitorMonitorKeyboard
Windows 8.1 ASUS ROG Swift PG278Q 27" 1440p 144hz G-Sync ASUS VG23AH 23.5" Passive 3d 1080p 60hz  Razer Blackwidow Chroma Tournament Edition 
PowerCaseMouseMouse Pad
EVGA Supernova G2 1000 Caselabs SMA8 -XXL Window, Ventilated sides/t... Razer Ouroboros Norman Rockwell collection series print Mousepad 
AudioAudio
Sennheiser HD700 300ohm Open Back Headphones SupremeFX Hi-Fi 5.25" Bay AMP/DAC  
  hide details  
Reply
 
CPUMotherboardGraphicsRAM
i7 5960X ASUS Rampage V Extreme MSI GTX 1080 Gaming X 16GB Corsair Dominator Platinum 2666mhz DDR4 C15 
Hard DriveHard DriveOptical DriveCooling
PNY CS2211 MLC SATA III SSD WD Blue 500GB 7200rpm Writemaster DVD/CD +/- RW EK Supremacy EVO 
CoolingCoolingCoolingCooling
Mayhem's Pastel Ice White Coolant Coollaboratory Liquid Copper TIM XSPC EX480mm Radiator Black Ice GTX 360mm Radiator 
CoolingCoolingCoolingCooling
EK XTC 420mm Radiator Swiftech MCP655-B 12v Pump EK RES X3 250 Reservoir Bitspower/EK 3/8" x 1/2" Compression fitting 
OSMonitorMonitorKeyboard
Windows 8.1 ASUS ROG Swift PG278Q 27" 1440p 144hz G-Sync ASUS VG23AH 23.5" Passive 3d 1080p 60hz  Razer Blackwidow Chroma Tournament Edition 
PowerCaseMouseMouse Pad
EVGA Supernova G2 1000 Caselabs SMA8 -XXL Window, Ventilated sides/t... Razer Ouroboros Norman Rockwell collection series print Mousepad 
AudioAudio
Sennheiser HD700 300ohm Open Back Headphones SupremeFX Hi-Fi 5.25" Bay AMP/DAC  
  hide details  
Reply
post #887 of 7252
Quote:
Originally Posted by cekim View Post

Not easy to do quickly, I'm back into linux. Cooling is an EK full-board block for the Rampage 5 Extreme with a 420 + 240 rad (loop includes 2 video cards as well).

Aida64 Temps from memory and from seeing it fully loaded in linux:
peak is 70-72C
Average is 60-65C

I don't run Prime95, if Aida and RealBench pass, I move on to linux, starting with stressapp which usually requires a little more cache/dram voltage somewhere than aida and then on to my personal work-load examples.

4.4CPU 1.325v adaptive
3.7Cache 1.290v (offset)
SA=-0.06V offset (yes, minus, Auto was driving this up higher than needed)
LLC=5
BCLK=100
DDR4 128G 3000-14-14-14-34-2T 1.35v (sticker values for this ram - which Haswell could never do with these same sticks on this same board)

Would be good to see some pass results in the link below with 128gb
post #888 of 7252
Quote:
Originally Posted by cekim View Post

The third number is "wall-clock" time in (minutes:seconds:milli-seconds) format for comparison sake.

Here's another contrived test with threads - feel free to critique or find flaws that might produce non-linear results in my code, I really just smashed this together right now, so I am sure its less than perfect:
Tight loop with just simple multithreading from C++ library, all data are in L1 cache. Too much synchronization for my taste.

There on overclock net is somewhere my Java program for a testing difference between HT and normal running. But there is also Dolphin benchmark which, while it doesn't test heavy multithreading, tests well a single "peak core" on variable load.
post #889 of 7252
Guys, does the Turbo Boost Max 3.0 do anything on a CPU that is overclocked manually to 10x4.2GHz?
post #890 of 7252
Quote:
Originally Posted by Martin778 View Post

Guys, does the Turbo Boost Max 3.0 do anything on a CPU that is overclocked manually to 10x4.2GHz?

Only if you do not sync all cores to the same frequency. To use TBM 3.0, you need to apply the "by specific core" option (that's what it is called in the ASUS UEFI). The Intel driver will then lock suitable applications to the fastest cores (those you set to higher ratios). Bear in mind the voltage mode needs to be in Auto or Offset to make use of this feature properly. There are some side-effects to doing so, highlighted in the thermal control tool guide in my sig.
Edited by Raja@ASUS - 6/30/16 at 3:38am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Broadwell-E thread