E5 v3 CPU systemboard BIOS mod to increase hashrate

[Marsh]

E5 v3 CPU have a "feature" to allow max turbo with all cores.
See What controls Turbo Core in Xeons?

For the past month, I had successfully flash my Asrock , GigaByte , MSI board bios to "all cores max turbo"
There is a nice person at anandtech.com that would mod the bios with "all core turbo" and reduce cpu voltage.
The net result ( for some E5 v3 CPU ) is over 10% with the same power usage.

Example of a single CPU E5-2673 v3 , all core turbo to 3.1GHZ

Model name: Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz
Stepping: 2
CPU MHz: 3099.937
CPU max MHz: 3100.0000
CPU min MHz: 1200.0000

Hashrate went from around 590 H/s to 657 H/s without using extra power.

Next up is do the same mod to dual CPU server board, it is more risky.
I like Asrock board , it has dual bios , bios chip is removable from socket, making it fool proof , impossible to brick the motherboard.

 

[Patrick]

Does it work with v4?

Do you need a custom bios for each CPU?

 

[nthu9280]

@Marsh

Though I'm aware of that thread, I have not been following that closely. May have missed but didn't see Supermicro boards listed. Do you know if X10SRA-F can be modded?

Sent from my Nexus 6 using Tapatalk

 

[nthu9280]

Does it work with v4?

Do you need a custom bios for each CPU?

It's only possible for V3 and not on V4.

I'm oversimplifying but the concept is to remove the processor microcode update from the bios and then load it in the OS after enabling all core turbo in BIOS.



Sent from my Nexus 6 using Tapatalk

 

[Marsh]

No ready to flash bios . The nice person published a guide to mod the bios yourself.
This will be my next adventure.

There is a ASUS Z10PE WS bios ready to go in the thread. But I have a ASUS Z10PE-D6-4L board, so I need to roll my own bios, as well as supermicro boards.

I picked Asrock board to experiment first , it has dual bios chips, fool proof to recover the board.

Since XMR mining difficulty is really high, I need to squeeze more performance per watt.
I was skeptical first, now I am convince the benefit ready to roll my own BIOS mod.

 

[britinpdx]

May have missed but didn't see Supermicro boards listed.

I think that there were a couple of reports of success with both single CPU and Dual CPU Supermicro boards. I have been able to modify an X10DAi BIOS and had initial success with 2x E5-2658v3.

I've only tested one of the numerous options for microcode and voltage offsets (v3x2_50_39_vcc.efi). That has allowed Cinebench R15 results to go from 2873 to 3295

 

[maxermaxer]

I am curious, what is the downside of doing this? Will it make the system unstable?


Sent from my iPhone using Tapatalk

 

[Marsh]

I have 4 systems with bios mod doing XMR mining for few weeks now
No stability problem. The bios mod also reduce the CPU IA offset voltage, more performance , without more heat and power. If no using turbo, when system is idle , it actually save power usage.

I noticed that Windows 10 would crash sometimes after wake up from sleep.
I spent couple hours of troubleshooting the problem.
In the end, I gave up because I use Ubuntu for mining, and no sleep for the miners.

 

[Klee]

E5 v3 CPU have a "feature" to allow max turbo with all cores.
See What controls Turbo Core in Xeons?

For the past month, I had successfully flash my Asrock , GigaByte , MSI board bios to "all cores max turbo"
There is a nice person at anandtech.com that would mod the bios with "all core turbo" and reduce cpu voltage.
The net result ( for some E5 v3 CPU ) is over 10% with the same power usage.

Example of a single CPU E5-2673 v3 , all core turbo to 3.1GHZ

Model name: Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz
Stepping: 2
CPU MHz: 3099.937
CPU max MHz: 3100.0000
CPU min MHz: 1200.0000

Hashrate went from around 590 H/s to 657 H/s without using extra power.

Next up is do the same mod to dual CPU server board, it is more risky.
I like Asrock board , it has dual bios , bios chip is removable from socket, making it fool proof , impossible to brick the motherboard.

I wish I would have known about that before I bought my E5-2667 V3 ES cpu's , apparently the ES cpu's have issues doing that because mine wont.

I was deciding between E5-2683 V3, when they were much cheaper about 7 months ago, and the 2667's ES cpu's that were about $40 cheaper.

I went with the 2667's because of the higher clockspeed.........if I only knew.....

 

[Klee]

I am curious, what is the downside of doing this? Will it make the system unstable?


Sent from my iPhone using Tapatalk

Big downside is if you do anything wrong you have a dead board until you can replace/reflash it.

And that could be either a minor issue (if you have a bios chip programmer or a second working board to do a hot swap and reflash) or a major issue ( if you do not have the programmer or a way to reflash the chip).

 

[Marsh]

@britinpdx

I have SM X10DAX board, do you have any tips on modding the BIOS?
thanks

 

[Marsh]

I wish I would have known about that before I bought my E5-2667 V3 ES cpu's

The turbo mod work with E5 v3 QS CPU

Model name: Intel(R) Xeon(R) CPU E5-2650L v3 @ 1.80GHz
Stepping: 2
CPU MHz: 2499.820
CPU max MHz: 2500.0000
CPU min MHz: 1200.0000

My system have CPU E5-2650L v3 QGP8

Miner23 , Asrock X99 , CPU E5-2650L v3 QGP8 , L3 cache 30MB , 15 Threads , BIOS active 8 cores
Before E5-2650L v3 ( 8 active cores , 15 threads ) , 482 H/s 74w = 6.5 H/s per watt
Turbo BIOS with Intel-microcode package
E5-2650L v3 ( 8 active cores , 15 threads ) , 544 H/s 74w = 7.35 H/s per watt

 

[britinpdx]

I have SM X10DAX board, do you have any tips on modding the BIOS?

Unfortunately I didn't take precise notes so I can't give guaranteed instructions.

The Anandtech thread is a long read and there's a ton of information in there, it took me a while and several passes to pick out the details.

I could never get the UBU tool to work correctly and it was because of an attributes bit not being set correctly prior to attempting to remove unwanted microcode. There are 2 methods posted in the thread to overcome this, post #1067 and post #1068.

I think that was the key to success. I seem to remember I followed the instructions in part 1 and 2A of the #1067 post.

 

[Klee]

The turbo mod work with E5 v3 QS CPU

Model name: Intel(R) Xeon(R) CPU E5-2650L v3 @ 1.80GHz
Stepping: 2
CPU MHz: 2499.820
CPU max MHz: 2500.0000
CPU min MHz: 1200.0000

My system have CPU E5-2650L v3 QGP8

Miner23 , Asrock X99 , CPU E5-2650L v3 QGP8 , L3 cache 30MB , 15 Threads , BIOS active 8 cores
Before E5-2650L v3 ( 8 active cores , 15 threads ) , 482 H/s 74w = 6.5 H/s per watt
Turbo BIOS with Intel-microcode package
E5-2650L v3 ( 8 active cores , 15 threads ) , 544 H/s 74w = 7.35 H/s per watt

These are ES not QS, and even if it would work the base clock is 2.9 Ghz and turbo boost is not a whole lot higher.

 

[Marsh]

@britinpdx

With SM board , How difficult to recover a failed bios mod booboo ?
Have you had a failure and recover from a bad bios mod?

I want to read up all methods to recover failed attempt with ASUS , SM board before modding BIOS.

 

[Patrick]

@Marsh I just setup the dual E5-2650L V3 system

• Wolf's: 869H/s
• Stak (auto conf): 993H/s
• xmrig 30T: 1008H/s
• xmrig 2x 15T: 1034H/s

Those turbo numbers probably have you around 1100H/s for two. Usually one CPU mines slightly slower than the other in 2P systems from what I can tell due overhead.

I need to double-check the PDU reading, but it is showing 105W right now. Only 1x 16GB DIMM per CPU running at 2133MHz.

 

[britinpdx]

With SM board , How difficult to recover a failed bios mod booboo ?

Not at all sure, I've never had to do it ... my first attempt at a bios mod appears to have been successful. In hindsight, I probably should have thought through the possible outcomes and have been prepared !

My fallback would probably be a return to Supermicro for repair.

 

[nthu9280]

@britinpdx

With SM board , How difficult to recover a failed bios mod booboo ?
Have you had a failure and recover from a bad bios mod?

I want to read up all methods to recover failed attempt with ASUS , SM board before modding BIOS.

This was my exact thought too. I wanted to get contingency plan first before attempting this. And the thread on AnandTech, like the E5-2670 thread here, lot of useful info but sprinkled all over the place. My day job which is totally different was crazy with Fiscal Year close. Now that is behind, I can start on this.

The ch341A bios programmers are cheap (<$7) from US sources. It's my soldering skills (or lack there of) with delicate components is what I'm worried about.



Sent from my Nexus 6 using Tapatalk

 

[Klee]

@Marsh I just setup the dual E5-2650L V3 system

• Wolf's: 869H/s
• Stak (auto conf): 993H/s
• xmrig 30T: 1008H/s
• xmrig 2x 15T: 1034H/s

Those turbo numbers probably have you around 1100H/s for two. Usually one CPU mines slightly slower than the other in 2P systems from what I can tell due overhead.

I need to double-check the PDU reading, but it is showing 105W right now. Only 1x 16GB DIMM per CPU running at 2133MHz.

Bare metal or docker?

 

[Marsh]

I just checked the overclock E5-2650L v3 now.
hashrate went up to 554 h/s without making any change.
Totals: 553.7 554.0 554.0

I noticed that many times, miner hashrate would fluctuate 10 h/s due to variation of difficulty.

Running xmr-stak-cpu with Ubuntu 16.0.4