For more information see OCNG5: introduction.
If you find this project valuable please consider donating to Area 51 Development.
Donations allow continued maintenance of existing projects and support future
projects.
OCNG is a modification of BIOS for Supermicro 4P G34 boards (specifically, H8QGi, H8QG6 and H8QGL series) to allow overclocking.
Currently it supports Opteron 6100 series and has been adopted by a number of users (folders, mostly). You can learn more about it here.
Development of Opteron 6200/6300 series is planned to begin on December 2nd, 2013 and is intended to be conducted on full-time basis and for that reason it will be completely community-sponsored project.
If you'd like to support it, donate!
Several sponsorship packages are available:
- 10 USD: your nickname will be printed in the BIOS splash (unless you want to remain anonymous) -- provided in the first official release
- 20 USD: you will additionally receive weekly progress updates
- 50 USD: you will additionally receive custom BIOS ROM with splash of your choosing (subject to size/format limitations); the ROM will be prepared after first official release (first-come, first-served)
- 100 USD: you will additionally gain access to prerelease BIOS ROMs (with invitation to participate in testing -- no obligation; NOTE: you are responsible for following BIOS flash procedure)
- 150 USD: you will additionally receive 1-on-1 live assistance from development team to maximize performance of your system (first-come, first-served, after first official release)
The original overclocking bios for Opty 6100s was a huge hit, especially in the Distributed Computing community. I am sure this updated version will be just as good.
ReplyDeleteThere is one thing to say about tear Professional. His work is unsurpassed and ne pays extreme attention to detail and testing. There is no doubt this version will be just as successful as the last.
ReplyDeleteI've been successfully using the 6100 series OC BIOS for over a year now so I expect the next generation will be a hit too.
ReplyDeleteIt's just incredible bring Overclocking to stock server boards. Not knowing a thing about Ubuntu or server hardware, it was surprisingly simple to pickup and the support has been just awesome. It has been a complete, game changer on how I do my Folding@Home. Now that AMD 6200 & 6300 processors are become affordable, I can't wait to see the next version!!
ReplyDeleteJust wanted to say "fantastic work" Tear and to the HardOCP Team that helps test these tools! Thank-you for helping us Fold better across all Folding Teams, and added my donation to this great work today.
ReplyDeletetexinga over on the EVGA team
Hi all! I have a H8SGL-F motherboard and bunch of 6100- and 6200 CPUs, both retail as well as ES. Is it possible to inject the microcode for overclocking of (at least) 6100 series CPUs into this BIOS? 6200- and 6300 would also be great. Willing to cash in as soon as I hear back from you :)
ReplyDeleteUgh, I missed the Reply link -- see my post about H8SGL below.
DeleteAlso, given you own ES chips, some items here may apply to you (esp. #2 and #4).
It is possible to port existing (6100) changes to H8SGL series (as we happen to have a test unit).
ReplyDeleteI think, however, that completing 6200/6300 changes for H8QG[6iL] series and then porting them over to H8SGL would be more efficient -- we'd have a single set of modifications (that span 6100/6200/6300 range) to port.
How urgent is H8SGL support to you?
Hey tear,
ReplyDeletethis sounds great. I am not in a real hurry. I did some brief overclocking and could pull a 4.0GHz Cinebench R11.5 stable on water cooling with 6272 @ 1.47V. I reckon I will be able to go upwards of 5GHz with my 2-stage cascade refrigeration cooler (-100°C) and maybe 5.5GHz on LN2.
How long do you think you will need to port changes to H8SGL-F? The 6100 series CPUs are all multiplier locked, apart from one.
Hi tiborr, thanks for the donation!
DeleteSeems you're playing in the big league there :) Very nice.
We've recently released Alpha 1 version of the 6200/6300 OCNG for 4p boards (a blog post is coming
shortly). It is stable and pretty much feature-complete.
That said, next step is merging 6100 code back with the new BIOS. I anticipate this process to take
a few (2-3) weeks (which will make the beta code).
At that point, migrating the code to H8SGL should be viable :)
Hi tear,
Deleteno, thank you for all the hard work.
Sounds good about H8SGL, looking forward to finally test it. In the mean while I will be doing some FB voltmods on the IC. The droop under full 16-core load 4GHz+ is almost ~ 0.2V and I would not rather play with Vdroop mod in order to save VRM from being toasted due to higher switching frequency.
Hi tiborr,
DeleteNow that OCNG5 Beta is out, I'll be putting H8SGL on the bench.
Just thought you'd like to know :)
Cheers,
tear
I would also love to see overclocking available for my H8SGL-F. My 6376 always runs cool and has a lot of headroom available for single-threaded applications (we could definitely use more performance there).
ReplyDeleteI'd be willing to pay for something like this, for sure.
Hi Matt,
DeleteSee my response to tiborrr on anticipated timing of H8SGL code.
Out of curiosity, do you run many H8SGL boards?
We've generally been focusing on improving "all-core turbo" scenario
when dealing with 6200/6300 CPUs.
This is caused by two things:
1. Typical OCNG load has been multithreaded
2. Opteron's turbo (with no restrictions imposed) is unpredictable and buggy;
specifically, it can't sustain "all-core turbo" frequency _at_ _all_ when all
cores/CUs are loaded; instead you get only half (or so) of "all-core turbo"
Not sure if you had a chance to observe the latter but I'd be happy to hear
of your experiences.
Anyway, for these reasons Alpha 1 actually forces all CPUs in "all-core turbo"
as they are brought up -- this allows the user to get 100% of "all-core turbo",
all the time.
Not saying I'm against supporting "some-core turbo", it's just hasn't been on
the radar is all. I'll run few "some-core turbo" tests in a spare while.
Alpha 1 code (as is) could safely get you to 3120 MHz (13x240) sustained
(with good air cooling). Some-core turbo could (theoretically) translate to
even 3840 MHz (16x240) providing some-core turbo is capable of sustaining
the some-core turbo (I'm unsure if it can but I'll play with it a bit).
Err, providing the *CPU* is capable of sustaining some-core turbo :)
DeleteMatt,
DeleteI thought you might want to know that dev work on H8SGL has started:
http://area51dev.blogspot.com/2014/08/new-development.html
The system I ran some tests on (just the one) on an H8SGL-F is using CentOS 6.5. I don't follow your response:
ReplyDelete"Opteron's turbo (with no restrictions imposed) is unpredictable and buggy"
and then:
"Anyway, for these reasons Alpha 1 actually forces all CPUs in "all-core turbo""
As I have 16 cores available, flexibility in being able to adjust cores individually would be nice but simply boosting performance (to a lesser degree) across all cores would be just as good.
What I was getting at is: if you let the CPU manage boost (no-boost/some-core-boost/
Deleteall-core-boost), results may not line up with natural expectations.
You can monitor actual P-state (per core) using 'TurionPowerControl -CM' (http://code.google.com/p/turionpowercontrol/).
On 6200/6300 Opterons, P-state 0 is "some-core boost" and P-state 1 is "all-core boost".
P-state 2 and "higher" are non-boosted power states (with P-state 2 corresponding to
nominal CPU frequency -- 2300 MHz in case of 6376).
The experiment is loading all cores and observing P-states. While one would expect
to see P-state 1 across all-cores this is not what one gets. Cores are constantly
switching between P-state 2 (no boost) and P-state 1 ("all-core boost")...
There is, however, incidentally discovered and undocumented mechanism
of making all cores stay in "all-core boost", all the time (with no regard to actual load).
Credit goes to 'quickz' (http://hardforum.com/showthread.php?t=1708870).
Does this help?
Yes, I've seen this before. Locking the processor to P1 keeps all cores set at 2600MHz. The problem with this is that it appears as though no cores will throttle to P0 (3.2Ghz) in this state. Leaving the default range of P0-P6 (Pb0-Pb1 + P0-P4) allows some cores to reach P0, increasing single threaded performance on those cores.
ReplyDeleteI'm not inclined to test it at the moment, but if I remember correctly, the last time I tried to set all cores to P0, the system locked up.
Indeed. It is very much either/or situation (and we've historically favored "locked Pb1" of the two options).
DeleteNonetheless, I'll take a look how Pb0 behaves when OC'd.
I solved this by deleting all profiles expect Pb1 in PSCheck MS Windows OS software in order to sustain maximum frequency under full 16-core load. No matter what voltage or temperature, the CPU would just throttle and would not go over ~ 180W power draw (wall). By deleting other Pstates the performance increased significantly and power draw peaked over 380W.
Delete4.7GHz single-core under -100c cascade cooler: http://hwbot.org/submission/2565892_tiborrr_superpi___1m_opteron_6272_19sec_188ms
3.9GHz 16-core on water cooling: http://hwbot.org/submission/2561211_tiborrr_wprime___1024m_opteron_6272_131sec_579ms
I'm looking for a solution under Linux though.
DeleteWe have two super micro boards with 4 opteron 6262.
ReplyDeleteWe would love to run them on 3.2 Ghz all turbo.
In the moment they always step down to 1.6 Ghz. (win srv 2014.)
Wich package should we pay to join the Beta?
Jay,
Delete100 USD tier will get you access to beta code.
I anticipate a 6262 system to reach 2400-2500 MHz (all-core) (a 300-400 MHz bump from stock 2100 MHz).
Some-core turbo may be able to reach 3300-3400 MHz.
3200 MHz all-core is pretty much impossible with 6262 chips.
What kind of work do you do on these machines?
C4D render.
Deleteits not full load all the time.
there are several load breaks after 2-3 minutes for 20-30 sec.
which chips do you recommend ?
being low price - and benefit from magic OCNG
with serious air cooling?
thanks for all your effort - you open a new field of cost efficient
render power for classic CPU work.
100 seems fair to us.
Got it. C4D is a highly multithreaded load which is typical OCNG application.
DeleteI don't really have specific recommendations (all chips will work fine),
though best OC results can usually be obtained with low and standard
power chips (SE chips don't OC as much and require extra care
when it comes to cooling).
That said, 6380 chips would probably be my favorite.
Forgive me if this has been asked before. Running some ES chips (6200 series) in a dual socket unsupported Supermicro board. Planning to move to quad socket H8QGi. Can I adjust the max multiplier with this BIOS, since these chips are unlocked?
ReplyDeleteThere is a test BIOS for H8DGU series, so drop us an e-mail if you're interested. There has been no release of H8DGU OCNG (which I'd like to combine with 5.4) due to real life catching up.
DeleteTo answer your question -- yes, you will be able to adjust the multiplier on H8QGi as well as voltage albeit up to +87.5mV. Wider range is planned for 5.4.
This comment has been removed by the author.
ReplyDelete