Info needed for msi x99s 7 for xeon turbo hack

[er557]

much appreciate if anyone has info regarding msi bios hack for the known xeon e5 v3 turbo unlock, ; currently with e5 2686 v3 18c/36t getting about 30% more than 6950x in cpu z bench, but would love to see those turbo bins locked to x29 all cores.
The board is an x99s gaming 7 latest bios, win 10 creators with latest microcode.

 

[Bill_Bright]

I see you have asked around on this and not received any help (that I found). That would suggest to me it would be best to just leave things alone and go with the defaults already supported.

 

[dorsetknob]

Not sure if @unclewebb and his throttlestop software can help

Contact him by PM

 

[P4-630]

much appreciate if anyone has info regarding msi bios hack for the known xeon e5 v3 turbo unlock, ; currently with e5 2686 v3 18c/36t getting about 30% more than 6950x in cpu z bench, but would love to see those turbo bins locked to x29 all cores.
The board is an x99s gaming 7 latest bios, win 10 creators with latest microcode.

I would first replace that crappy "i7" intel stock cooler....

Edit: Oh, update your system specs....

 

[cdawall]

Don't you have to remove the microcode for that CPU from the BIOS?

Found it, http://www.overclock.net/t/1591590/xeon-hacking-and-overclocking-x79-x99-beyond-x58

 

[er557]

OK, sorry for the delay, back from the depths of hell with hex editing, bios modding, efi mapping, nail biting waiting for POST, etc., .I will keep the defaults and let my system run as is, the risk of not POSTing, c3/c6 not supporting, OS hanging, compatibility issues , are not worth the risk of a 15-20% boost. the microcode stripping from msi bios is risky, but I did achieve the following:
- 44gb--->>48gb ram . raise vdimm, turn off, place 8th dimm in 8th slot, boots fine with extra 4gb stick. this bit is irrelevant, but if I were to reset the bios again I may have to take it out again for POST success- not worth it.
- convert OS drive from mbr to uefi gpt boot, easy enough.
- learn about the navigation in uefi shell, create a thumb drive with efi, map drives, recognize the boot drive etc.
- encrypted msi bios prevented tampering, and may be good to leave it at that.
- learn about bios modding utilities, efi drivers injection before boot, microcode removal from windows.

All in all, I don't do much heavy computations, the gains would be e-peen mostly, as the gpu results are already excellent, no need to wake sleeping dogs.

P.S. ,specs updated. @cdawall: that link is not related, is about overclocking and standard methods, . the main discussion is over at
https://forums.anandtech.com/threads/what-controls-turbo-core-in-xeons.2496647/page-29

that info is quite new and emerged few months ago.

 

[cdawall]

Someone on here has one with the microcode blown out is the only reason I even knew about it. Sorry for the wrong link thanks for the updated one though.

 

[er557]

UPDATE: using an elaborate method of extracting microcode and manually modifiying hex for same checksum, using eufitool, hxd, uefi bios editor to verify, succeeded to remove microcode for encapsulated bios of msi, for my cpu id. flash, boot ok without ucode, place efi driver in boot partition, mod turbo, strip ucode file from windows, update to 0x39 with VMware tool to stay on safe side(no ucode may be faster but more bugs), benchmark- success. 60% more performance in single core, turbos all hover around 3ghz, 33% more performace in multi core.
before:

and after mod: 100% stable, 2x gtx 1080 sc acx overclocked

the cpu score in time spy is better than that achieved in review with a 6950x overclocked to 4.5ghz on all cores. my setup has no overclocking per say.
the cpu multiplier being unlocked may provide further tweaking if need be as the bios has no ucode, only in windows by VMware, which may be removable to allow multiplier tweaking, but tdp may rise exponentialy even to 300watt per cpu!!

 

[cdawall]

That's pretty dope. It is saddening that the broadwell-e stuff has this fixed.

 

[er557]

That's pretty dope. It is saddening that the broadwell-e stuff has this fixed.

Indeed, though if you talk about your 2683v4 it still is ES and the mod only works for QS or higher stepping- otherwise no boot. and of course irrelevant as you said for broadwell-e.

I modified my bios settings as instructed elsewhere, to disable c3/c6, also increase tdp to 165, and fsb to 100.02, disable cupid limit, I may try upping the multiplier and see where that gets me, I don't know how the 0x39 mcode loaded in windows manually will affect this, as mcode is the main lockdown method of xeons.
So that saved me about 4000$ I would have not paid for 2699v4 or similar for same performance.

 

[r9]

UPDATE: using an elaborate method of extracting microcode and manually modifiying hex for same checksum, using eufitool, hxd, uefi bios editor to verify, succeeded to remove microcode for encapsulated bios of msi, for my cpu id. flash, boot ok without ucode, place efi driver in boot partition, mod turbo, strip ucode file from windows, update to 0x39 with VMware tool to stay on safe side(no ucode may be faster but more bugs), benchmark- success. 60% more performance in single core, turbos all hover around 3ghz, 33% more performace in multi core.
before:

View attachment 88198 View attachment 88200



and after mod: 100% stable, 2x gtx 1080 sc acx overclocked

View attachment 88199 View attachment 88201


the cpu score in time spy is better than that achieved in review with a 6950x overclocked to 4.5ghz on all cores. my setup has no overclocking per say.
the cpu multiplier being unlocked may provide further tweaking if need be as the bios has no ucode, only in windows by VMware, which may be removable to allow multiplier tweaking, but tdp may rise exponentialy even to 300watt per cpu!!

You are saying that you could take it above 3GHz ?

 

[er557]

Of course, most of the time the TURBO bins for multiple cores is @x35 (3.5ghz) depending on load, when ALL cores maxed it peaks @ x29 multiplier ( 2.9ghz x 18 cores), instead of the default 2.3ghz all turbo, this is only with this tweak and no overclocking at all on any level, at idle it drops just fine, and vcore is default 0.675v as usuall.

 

[X4K4]

nice

 

[er557]

just measured temps with prime95 latest with avx2 and fma, which causes the amperage and voltage to increase and boosts temps and tdp; it stays nice and cool @65c all the time with the arctic freezer i30 cooler.
when compared the results from reviews, what I got is exactly double the scores of a 5960x cpu, around the ballpark of a 6950x overclocked to 5ghz.

 

[er557]

There she is, ripping thru cpu z with microcode 0x27, and bios tweaks, bclck 104, 18-->>12 cores @3.65ghz

 

[r9]

There she is, ripping thru cpu z with microcode 0x27, and bios tweaks, bclck 104, 18-->>12 cores @3.65ghz
View attachment 88238

But, can it run Crysis ?
Sorry had to say it .

 

[r9]

That's a monster of a machine.
18 cores @ 3.65GHz insane.
That mb/cpu combo cost just $550.

 

[iamthespy_]

Any clue if the mod would work on an engineering sample 2686 v3 (Stepping 2, Rev. m0) seeing a few of them go for around $300, worth trying?
they seem to be identical to the QS version and the stepping is the same (everything but the revision)

 

[er557]

from what you describe it should work fine, providing you can strip the ucode from your mb bios,
go to anandtech what controls turbo in xeons thread and hover there a bit.

 

[er557]

edit: still relevant and technically genius profit of performance, rig is dual xeon now and running smooth as butter, updated benches incoming, still going strong after four years.

 

[reks2004]

@er557 have you any link to faq about method what you use? I have X99A SLI and E5-2630 v3 (i cant reach stock 3,2ghz in turbo, i have max 2,7ghz on single core)

 

[er557]

Have a look at anand tech thread "what controls turbo in xeons", lengthy info with links spread out

 

[reks2004]

Thanks

 

[er557]

Just a little adendum, about how efficient and stable the efi driver hack is, system is 100% stable, crash free, about to be upgraded to ecc ram (while few reasons to do so). The driver tricks the SVID and TDP limit and unlocks them, here is an example of linpack extreme running @ 0.5 Tflops, and showing the cpus draw about 250W each from the PSU( max wattage vs. idle wattage)!, while the stock tdp is 135w!
also limit reasons show very few electrical limits, otherwise no throttling at all and clean as a whistle. Mind you the figures @ hwinfo are split into columns, middle is all time max, left in minimum, right is average if i'm not mistaken.

The performance is 30% above stock intel spec, all 36 cores are @ 3~3.5 ghz simultaneously