Overclock.net › Forums › AMD › AMD CPUs › [Official] AMD Ryzen DDR4 24/7 Memory Stability Thread
New Posts  All Forums:Forum Nav:

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread - Page 8

post #71 of 193
Thread Starter 
Quote:
Originally Posted by gtbtk View Post


Firestrike is an excellent tool for our purposes and was the reason that I have been so vocal that it is not the windows scheduler or CCX thread switching as the primary cause of the performance issues (yes there is a small amount of latency added when threads switch but that is symptomatic of the low memory frequency and high memory latency. You don't use it in this situation looking at absolute numbers. The total score really doesn't mean much for out purposes,  You use the balance of the individual component scores to identify trends. 

FS Graphics scores for a given GPU all fall within a reasonably small range for that particular model of card eg a GTX 1070 will run roughly about 19000 stock and about 21000 when overclocked, give or take.  That remains true on both Intel and Ryzen systems. That test Loads the GPU without stressing the CPU very much

FS Physics scores likewise tracks the CPU frequency/IPC pretty consistently for both Intel and Ryzen. At the same clocks, an R7 and a 6900K will both roughly sore about the same physics score. That has been true since release of the Ryzen CPUs. This test stresses the CPU and puts minimal load on the GPU. If it was thread switching causing the major part of the problem, you would see it impacting the physics score as well but both brands of CPU tend to track each other quite well. The 6900K only pulls ahead if it is overclocked above about 4.1Ghz. 

The FS combined score demonstrates where the performance is being impacted. That is in the part of the system between the cores and the GPU itself that has to run double duty supporting both GPU and CPU related memory traffic and traffic between the CPU and the GPU over PCIe. 6900K will do a 9000-10000 Combined score and the R7 and R5 Ryzens have, until the recent memory OC breakthroughs, was stuck at the 6000-6500 level. Since 3200 c14 single rank speeds became a reality, The best scores are now at around 8500 which is almost matching the Intel equivalent chip. Dual Rank memory with the extra interleaving also seems to help the performance as well.

If a score of 10000 pops up, we then have the chance of analyzing what is happening on that system that no-one else is doing


Being forthright, you need to let that go. Furthermore, this is a stability thread. If you want to start a thread comparing 3DMark scores you can do so yourself, however, there is already one in the graphics section. thumb.gif
post #72 of 193
@gtbtk

See this post 6 runs of 2133MHz vs 3200MHz. "We" can only control primary timings, the rest are dictated by strap.





Due to nature of this thread, happy to discuss there my results smile.gif .
XPS - R7 1700
(14 items)
 
XPS - i5 4690K
(12 items)
 
XPS - Q6600
(8 items)
 
CPUMotherboardGraphicsRAM
Ryzen 7 1700 Asus Crosshair VI Hero Sapphire Fury X@1145/545 G.Skill Trident Z 2x 8GB 3200MHz C14 
Hard DriveCoolingCoolingCooling
Samsung Evo 840 ThermalRight Archon IB-E X2 + 2x TY143 ThermalRight TY-143 2x front case intake Arctic Cooling F12 + 2x F9 as rear case exhaust 
OSMonitorKeyboardPower
Win 7 Pro x64 / Win 10 Pro x64 Asus MG279Q Cherry MX-Board 3.0 Cooler Master V850 
CaseMouse
SilverStone TJ06 Logitech G700S 
CPUMotherboardGraphicsRAM
i5 4690K@4.9GHz 1.255V Asus Maximus VII Ranger Sapphire Fury X@1145/545 HyperX Savage 2x8GB 2400MHz C11 
Hard DriveCoolingOSMonitor
Samsung Evo 840 TR Archon SB-E X2 Win 7 Pro x64 / Win 10 x64 Asus MG279Q 
KeyboardPowerCaseMouse
Cherry MX-Board 3.0 Cooler Master V850 SilverStone TJ06 Logitech G700S 
CPUMotherboardGraphicsRAM
Intel Q6600 G0 Asus P5K Premium Black Pearl Sapphire Toxic HD5850 Corsair Dominator 4GB 
Hard DriveCoolingOSPower
Crucial MX 100 256GB TR TRUE Rev.A Win 7 Pro x64 Cooler Master V650 
  hide details  
Reply
XPS - R7 1700
(14 items)
 
XPS - i5 4690K
(12 items)
 
XPS - Q6600
(8 items)
 
CPUMotherboardGraphicsRAM
Ryzen 7 1700 Asus Crosshair VI Hero Sapphire Fury X@1145/545 G.Skill Trident Z 2x 8GB 3200MHz C14 
Hard DriveCoolingCoolingCooling
Samsung Evo 840 ThermalRight Archon IB-E X2 + 2x TY143 ThermalRight TY-143 2x front case intake Arctic Cooling F12 + 2x F9 as rear case exhaust 
OSMonitorKeyboardPower
Win 7 Pro x64 / Win 10 Pro x64 Asus MG279Q Cherry MX-Board 3.0 Cooler Master V850 
CaseMouse
SilverStone TJ06 Logitech G700S 
CPUMotherboardGraphicsRAM
i5 4690K@4.9GHz 1.255V Asus Maximus VII Ranger Sapphire Fury X@1145/545 HyperX Savage 2x8GB 2400MHz C11 
Hard DriveCoolingOSMonitor
Samsung Evo 840 TR Archon SB-E X2 Win 7 Pro x64 / Win 10 x64 Asus MG279Q 
KeyboardPowerCaseMouse
Cherry MX-Board 3.0 Cooler Master V850 SilverStone TJ06 Logitech G700S 
CPUMotherboardGraphicsRAM
Intel Q6600 G0 Asus P5K Premium Black Pearl Sapphire Toxic HD5850 Corsair Dominator 4GB 
Hard DriveCoolingOSPower
Crucial MX 100 256GB TR TRUE Rev.A Win 7 Pro x64 Cooler Master V650 
  hide details  
Reply
post #73 of 193
Quote:
Originally Posted by gtbtk View Post


Firestrike is an excellent tool for our purposes and was the reason that I have been so vocal that it is not the windows scheduler or CCX thread switching as the primary cause of the performance issues (yes there is a small amount of latency added when threads switch but that is symptomatic of the low memory frequency and high memory latency. You don't use it in this situation looking at absolute numbers. The total score really doesn't mean much for out purposes,  You use the balance of the individual component scores to identify trends. 

FS Graphics scores for a given GPU all fall within a reasonably small range for that particular model of card eg a GTX 1070 will run roughly about 19000 stock and about 21000 when overclocked, give or take.  That remains true on both Intel and Ryzen systems. That test Loads the GPU without stressing the CPU very much

FS Physics scores likewise tracks the CPU frequency/IPC pretty consistently for both Intel and Ryzen. At the same clocks, an R7 and a 6900K will both roughly sore about the same physics score. That has been true since release of the Ryzen CPUs. This test stresses the CPU and puts minimal load on the GPU. If it was thread switching causing the major part of the problem, you would see it impacting the physics score as well but both brands of CPU tend to track each other quite well. The 6900K only pulls ahead if it is overclocked above about 4.1Ghz. 

The FS combined score demonstrates where the performance is being impacted. That is in the part of the system between the cores and the GPU itself that has to run double duty supporting both GPU and CPU related memory traffic and traffic between the CPU and the GPU over PCIe. 6900K will do a 9000-10000 Combined score and the R7 and R5 Ryzens have, until the recent memory OC breakthroughs, was stuck at the 6000-6500 level. Since 3200 c14 single rank speeds became a reality, The best scores are now at around 8500 which is almost matching the Intel equivalent chip. Dual Rank memory with the extra interleaving also seems to help the performance as well.

If a score of 10000 pops up, we then have the chance of analyzing what is happening on that system that no-one else is doing

Hello

That was quite a wall of text considering as stated previously it is irrelevant to this thread.
post #74 of 193
Quote:
Originally Posted by Silent Scone View Post
 
Quote:
Originally Posted by gtbtk View Post


Firestrike is an excellent tool for our purposes and was the reason that I have been so vocal that it is not the windows scheduler or CCX thread switching as the primary cause of the performance issues (yes there is a small amount of latency added when threads switch but that is symptomatic of the low memory frequency and high memory latency. You don't use it in this situation looking at absolute numbers. The total score really doesn't mean much for out purposes,  You use the balance of the individual component scores to identify trends. 

FS Graphics scores for a given GPU all fall within a reasonably small range for that particular model of card eg a GTX 1070 will run roughly about 19000 stock and about 21000 when overclocked, give or take.  That remains true on both Intel and Ryzen systems. That test Loads the GPU without stressing the CPU very much

FS Physics scores likewise tracks the CPU frequency/IPC pretty consistently for both Intel and Ryzen. At the same clocks, an R7 and a 6900K will both roughly sore about the same physics score. That has been true since release of the Ryzen CPUs. This test stresses the CPU and puts minimal load on the GPU. If it was thread switching causing the major part of the problem, you would see it impacting the physics score as well but both brands of CPU tend to track each other quite well. The 6900K only pulls ahead if it is overclocked above about 4.1Ghz. 

The FS combined score demonstrates where the performance is being impacted. That is in the part of the system between the cores and the GPU itself that has to run double duty supporting both GPU and CPU related memory traffic and traffic between the CPU and the GPU over PCIe. 6900K will do a 9000-10000 Combined score and the R7 and R5 Ryzens have, until the recent memory OC breakthroughs, was stuck at the 6000-6500 level. Since 3200 c14 single rank speeds became a reality, The best scores are now at around 8500 which is almost matching the Intel equivalent chip. Dual Rank memory with the extra interleaving also seems to help the performance as well.

If a score of 10000 pops up, we then have the chance of analyzing what is happening on that system that no-one else is doing


Being forthright, you need to let that go. Furthermore, this is a stability thread. If you want to start a thread comparing 3DMark scores you can do so yourself, however, there is already one in the graphics section. thumb.gif

 

This thread, and overclocking in general is not only about stability, it is about stability with the best performance. If you only want stability then you should be running your ram at 2133 with default timings and you don't need this thread in the first place. All I and suggesting is for you guys to list the latency results you get from the Stirling efforts that you are making. You are generally testing for it anyway.

 

If you also add a firestrike score to the work that you are already doing, YOU guys get to see an additional data point, together with super Pi numbers that will show YOU what are the best methods to use to get the best stability and performance out of this new system.  You may even find things that you don't expect like dual rank performance vs single rank performance. Does the extra interleaving give better benefits than single rank frequency and tight timings? I don't know for certain, but there is some evidence that suggest that there might be something to it if the top scores that have been posted at 2400 and 2666Mhz were not a fluke.

post #75 of 193
colorfuel--1700X@3.5---3200Mhz-C14-14-14-34-1T---1.35v---SOC 1.0v---BIOS F3---HCI---500%

edit: F4-3200C14D-16GFX

Just testing the standard XMP profile settings for this 16GB G.Skill Flare X kit on stock CPU clocks.

On my Gigabyte K7, it sometimes takes a few reboots to train the setting, but it has proven stable until now.

I have to manually set the SOC voltage to 1.0, on auto, this board gives the SOC 1.25v, which, apparently, is too much.


Edited by colorfuel - 5/4/17 at 10:31am
Hope (for AMD)
(14 items)
 
  
CPUMotherboardGraphicsRAM
AMD 1700X Gigabyte AX-370X-Gaming K7 Sapphire Tri-X 290X new edition G.SKILL Flare X Series 16GB - F4-3200C14D-16GFX 
Hard DriveHard DriveHard DriveHard Drive
SAMSUNG SSD 960 EVO 250GB Crucial M4 CT256M4 Crucial CT500BX Western Digital WD10EZEX 
CoolingOSMonitorPower
Thermalright HR-02 Macho Rev.A (BW) Windows 10 Fujitsu P24W-6 IPS be quiet! Straight Power E9 CM 80+Gold 580W 
CaseMouse
Fractal R4 Logitech Performance MX 
  hide details  
Reply
Hope (for AMD)
(14 items)
 
  
CPUMotherboardGraphicsRAM
AMD 1700X Gigabyte AX-370X-Gaming K7 Sapphire Tri-X 290X new edition G.SKILL Flare X Series 16GB - F4-3200C14D-16GFX 
Hard DriveHard DriveHard DriveHard Drive
SAMSUNG SSD 960 EVO 250GB Crucial M4 CT256M4 Crucial CT500BX Western Digital WD10EZEX 
CoolingOSMonitorPower
Thermalright HR-02 Macho Rev.A (BW) Windows 10 Fujitsu P24W-6 IPS be quiet! Straight Power E9 CM 80+Gold 580W 
CaseMouse
Fractal R4 Logitech Performance MX 
  hide details  
Reply
post #76 of 193
Thread Starter 
Quote:
Originally Posted by colorfuel View Post

colorfuel--1700X@3.5---3200Mhz-C14-14-14-34-1T---1.35v---SOC 1.0v---BIOS F3---HCI---500%


Just testing the standard XMP profile settings for this 16GB G.Skill Flare X kit on stock CPU clocks.

On my Gigabyte K7, it sometimes takes a few reboots to train the setting, but it has proven stable until now.

I have to manually set the SOC voltage to 1.0, on auto, this board gives the SOC 1.25v, which, apparently, is too much.


If stable at 1v for that frequency, then 1.25v is excessive, yes.

Great result smile.gif
post #77 of 193


Nighthog--1700@4.0---2666Mhz-C14-15-13-24-2T---1.20v---SOC 0.9v---BIOS F6---HCI---1000%

CPU VCORE: 1.500V
Gigabyte GA-AB350-Gaming 3

CL 14.15.13.(13).24.62.312 2T
CMK16GX4M2A2666C16R ver 3.21 Micron 2Rx8
    
CPUMotherboardGraphicsRAM
AMD Ryzen 7 1700 Gigabyte AB350-Gaming 3 XFX RX 480 RS 4Gb Corsair LPX red 2666Mhz C16 2x8Gb @ 14.15.13.13.24 
CoolingOSMonitorMonitor
Alphacool NexXxoS Cool Answer 360XT/DDC Windows 10 Home Samsung C27F390 LG L245WP-BN 
KeyboardPowerCaseMouse
Logitech G610 Orion Brown Corsair HX 620W CoolerMaster Centurion 5 Logitech MX Master 
Mouse Pad
FUNC surface 1030 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
AMD Ryzen 7 1700 Gigabyte AB350-Gaming 3 XFX RX 480 RS 4Gb Corsair LPX red 2666Mhz C16 2x8Gb @ 14.15.13.13.24 
CoolingOSMonitorMonitor
Alphacool NexXxoS Cool Answer 360XT/DDC Windows 10 Home Samsung C27F390 LG L245WP-BN 
KeyboardPowerCaseMouse
Logitech G610 Orion Brown Corsair HX 620W CoolerMaster Centurion 5 Logitech MX Master 
Mouse Pad
FUNC surface 1030 
  hide details  
Reply
post #78 of 193
Not sure if this would be enough, 4x8gb 3200 but on samsung e-die, tried 16-18-18 but failed : / cl18 is working so might try now to lower voltages and check again

haszek--1700@3.9---3200Mhz-C18-16-16-36-1T---1.40v---SOC 1.15v---BIOS 1107---HCI---220%

ryzen R7 build
(13 items)
 
  
CPUMotherboardGraphicsRAM
r7 1700@3.9 1.36v asus crosshair 6 hero msi gtx 1080 sea hawk ek corsair 4x8gb vengeance led 3466@3200 1.45v 
Hard DriveHard DriveCoolingOS
samsung m.2 evo 960 samsung evo 850 ekwb preformance 360+240 windows 10 
MonitorPowerCaseMouse
lg oled c6 evga p2 1000w phanteks evolv atx logitech g602 
Audio
denon x2200 
  hide details  
Reply
ryzen R7 build
(13 items)
 
  
CPUMotherboardGraphicsRAM
r7 1700@3.9 1.36v asus crosshair 6 hero msi gtx 1080 sea hawk ek corsair 4x8gb vengeance led 3466@3200 1.45v 
Hard DriveHard DriveCoolingOS
samsung m.2 evo 960 samsung evo 850 ekwb preformance 360+240 windows 10 
MonitorPowerCaseMouse
lg oled c6 evga p2 1000w phanteks evolv atx logitech g602 
Audio
denon x2200 
  hide details  
Reply
post #79 of 193
Thread Starter 
Quote:
Originally Posted by haszek View Post

Not sure if this would be enough, 4x8gb 3200 but on samsung e-die, tried 16-18-18 but failed : / cl18 is working so might try now to lower voltages and check again

haszek--1700@3.9---3200Mhz-C18-16-16-36-1T---1.40v---SOC 1.15v---BIOS 1107---HCI---220%



Thanks, it's enough - as minimal coverage for 32GB is 200%. At least enough for entry. With densities above 16GB I would recommend using Google Stress App
post #80 of 193
Quote:
Originally Posted by haszek View Post

Not sure if this would be enough, 4x8gb 3200 but on samsung e-die, tried 16-18-18 but failed : / cl18 is working so might try now to lower voltages and check again

haszek--1700@3.9---3200Mhz-C18-16-16-36-1T---1.40v---SOC 1.15v---BIOS 1107---HCI---220%


Test went through at 1.4v but every windows restart was resetting pstates due to failed boot. Only 1.45v DDR boot and 1.44v DDR voltage helped to make it stable.
ryzen R7 build
(13 items)
 
  
CPUMotherboardGraphicsRAM
r7 1700@3.9 1.36v asus crosshair 6 hero msi gtx 1080 sea hawk ek corsair 4x8gb vengeance led 3466@3200 1.45v 
Hard DriveHard DriveCoolingOS
samsung m.2 evo 960 samsung evo 850 ekwb preformance 360+240 windows 10 
MonitorPowerCaseMouse
lg oled c6 evga p2 1000w phanteks evolv atx logitech g602 
Audio
denon x2200 
  hide details  
Reply
ryzen R7 build
(13 items)
 
  
CPUMotherboardGraphicsRAM
r7 1700@3.9 1.36v asus crosshair 6 hero msi gtx 1080 sea hawk ek corsair 4x8gb vengeance led 3466@3200 1.45v 
Hard DriveHard DriveCoolingOS
samsung m.2 evo 960 samsung evo 850 ekwb preformance 360+240 windows 10 
MonitorPowerCaseMouse
lg oled c6 evga p2 1000w phanteks evolv atx logitech g602 
Audio
denon x2200 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD CPUs
Overclock.net › Forums › AMD › AMD CPUs › [Official] AMD Ryzen DDR4 24/7 Memory Stability Thread