OCNG5: OC firmware for Supermicro AMD G34 platforms

tear

[H]ard|DCer of the Year 2011
Joined
Jul 25, 2011
Messages
1,568
 
OCNG5 (OC Next-Generation, version 5) is a firmware modification for Supermicro 4P G34
motherboards (complete list of supported motherboards below) that allows overclocking of
all current AMD Opteron processors, that is 6100, 6200 and 6300 series.

OCNG5 works by altering motherboard's reference clock early at boot and tuning
processor components for performance and stability.

Reference clock (referred to as BCLK in Intel processors) is used to derive a number of
system clocks (including core clock, integrated northbridge clock, hypertransport clock
and memory clock). Thus, in principle, increasing reference clock increases system
speed.

Special care must be taken when increasing reference clock as certain components and
interfaces don't operate well beyond their rated speeds. Specifically, DRAM modules may
not be able to handle increased memory clock and memory interface tuning may not be
possible for very high speeds (given lack of characterization data). Similar limitation
applies to hypertransport interface.

OCNG5 takes care of these risks by ensuring that rated memory and hypertransport
speeds are never exceeded.

Taking advantage of performance/enthusiast memory is, however, possible. OCNG5 fully
supports XMP profile 1 and automatically uses it (if available) whenever OCNG5 features
are enabled.

Configuring OCNG5 currently requires Linux but, once set up, any operating system may
be used.

As overclocking translates to increased power draw, one needs to pay special attention to
cooling the system (both, the motherboard and the processors). On one hand, additional
heat limits attainable speeds and affects quality of voltage regulation. On the other,
overheating power delivery circuits may lead to their destruction and render the
motherboard inoperable. These make two good reasons to make sure that heat is
transported away from your system.

Caring for power delivery circuits is especially important when using high power parts
(carrying 140W TDP, including 6176 SE, 6282 SE, 6284 SE, 6386 SE). When using high
power parts or whenever AC power (80% eff. assumed) starts to exceed 800W, providing
airflow across the motherboard (front-to-back) is highly recommended.

This is especially true for systems located in enthusiast cases or on open trays. Server
cooling systems typically accommodate this recommendation so no special action
(except, say, bumping fan speeds) is required.

OCNG5 also comes with a safety mechanism: if the board doesn't POST three times in a
row, stock settings are restored. This eliminates the need to reset CMOS (a
time-consuming process) during OC process.

As of today, OCNG5 supports following motherboards:

[QG63NG53.C16 image]
- H8QGi+-F
- H8QGi-F
- H8QG6+-F
- H8QG6-F

[QG73NG53.C09 image]
- H8QGi+-LN4F
- H8QGi-LN4F
- H8QG7+-LN4F
- H8QG7-LN4F

[QGL3NG53.C19 image]
- H8QGL-6F (*)
- H8QGL-6F+ (*)
- H8QGL-iF (*)
- H8QGL-iF+ (*)

[DG63NG53.B25 image]
- H8DG6
- H8DG6-F
- H8DGi
- H8DGi-F

[SGL3NG53.B25 image]
- H8SGL
- H8SGL-F

(*) incompatibility between OCNG5 and IPMI has been identified on H8QGL series boards
  (no boot when overclocked);
  if IPMI is not required, disabling IPMI (using jumper JPB1) is recommended;
  otherwise, the issue can be worked around by completely removing JPB1 jumper;
  doing so will still allow IPMI/KVM access but IPMI access from the OS will not be
  possible

For installation instructions and further information please visit:
http://area51dev.blogspot.com/p/ocng5-installation.html
 
 
Last edited:
FAQ

Q: My OC isn't stable but I haven't exceeded 230 MHz refclock (6200 series) or 240 MHz refclock (6300 series). What do I do?
A: Start troubleshooting with setting NB multiplier to 9x in advanced section of the configuration utility (make sure to power-cycle afterwards).
   If still no joy, you may have got poor pieces of silicon :-( Given that voltage regulation quality differs between sockets, you may have a chance
   of improving things by rotating CPUs between sockets. Please post and we'll provide you with details.


Q: What about dual-socket G34 boards? (e.g. H8DGi/6)
A: They are currently unsupported simply because I don't own any. If you can sponsor or loan a unit for about 6 weeks I can work on porting the mod to it.
   However, mind that I'm currently in the EU (I estimate one-way shipping from the US to amount to ca. $50).


Q: My board worked fine but now indefinitely repeats 6-beep sequence (as if memory is absent) but memory is there. What's wrong?
A: This is a known problem with these boards. BMC accesses to on-board I2C may sometimes race with BIOS reading SPD during POST.
   As BMC code has issues with I2C arbitration, this results in hung bus. The solution is to completely remove AC power from the system
   until BMC hearbeat LED goes off and then reconnect the AC.
   Performing BMC cold reset (from IPMIview) *may* also help but it hasn't been tested.
 
Last edited:
Well done tear for those of you that have not had a chance to use this yet be prepared for a feature rich OC environment. Tear has outdone himself on this one. :eek: :D
 
If Linux is required for initial use, it would be cool if you had a ready-made Linux image that could be dumped to a flash drive. Just a thought.
 
If Linux is required for initial use, it would be cool if you had a ready-made Linux image that could be dumped to a flash drive. Just a thought.
Yeah, that is one option.
Another is creating Windows configuration utility -- I'll look into creating simple console tool.

Alternatively, if there are takers for making something shiny, the source is available [ocng-cu.c].
 
Yeah, that is one option.
Another is creating Windows configuration utility -- I'll look into creating simple console tool.

Alternatively, if there are takers for making something shiny, the source is available [ocng-cu.c].

Without your source code I made something shiny:

2014-09-28%2014.22.35.jpg


I am not able to use OCNG within windows, even after making something shiny?:confused:
 
I will be in about 20min when I head to work :p I'll report my findings once I install a driver.
 
Installed a driver, got to work, but no joy on OCNG working. I running an HP Elitebook with windows 8.1 Enterprise...halp?
 
Installed a driver, got to work, but no joy on OCNG working. I running an HP Elitebook with windows 8.1 Enterprise...halp?

this is used on quad socket G34 server hardware ( supermicro) not single socket laptops
 
1P G34 (H8SGL) support is nearing completion (should be there within a week or so); if you
like the project please support it by donating :) (use 4P form, same rules apply)

EDIT: it seems that voltage regulation of H8SGL is not as good as in the case of 4P boards
which limits OC options; while 15 (6200) to 20% (6300) of OC on top of all-core turbo is
easily achievable with 4P setup, you may need to take 5 percentage points off of these
numbers in case of H8SGL (i.e. 10 to 15% of OC on top of all-core turbo).

These, naturally, aren't OC limits but rather numbers to expect (I'd rather see you pleasantly
surprised than bitterly disappointed).
 
Last edited:
tear, I get black screens on my 4p with 230 and 240 fsb using 6380s after a period of time. 225 FSB is running well. Do you think it is chip related since my MC chips ran 240 fsb with no probs?
 
Hard to tell.

If you use FAH on that box, can you run:
Code:
fahdiag | pastebinit
and then share the URL?

It may contain clues as to what is going on (I'll make a generic version of fahdiag and
include it in next rev of ocng-utils).

In order to isolate the problem, can you cap the NB speed to 2000 MHz (advanced section
of ocng-cu) and see if doing so improves things at, say, 230 MHz?

Once you do that can you do these as well (few minutes after the machine is loaded):
Code:
sudo clockspeed | pastebinit
sudo tpc -temp | pastebinit
sudo tpc -l | grep 0.pstate[01] | pastebinit
sudo voltcheck | pastebinit
The last one may take up to a minute -- don't be alarmed.
 
I'm running BOINC WU, tear. I'll run the other commands for you.
 
Are you using the DC appliance by any chance?

If so, fahdiag is already there and, as it reports few non-fah bits (searches system logs for
hardware error reports for instance), it still makes sense to run it
 
Ok, I'm running the appliance so I will run fahdiag. Check back tomorrow if you don't mind since I'm not going to mess with the system tonight.

TIA
 
Grats on 6200/6300 OC support.
Tear said that 15% OC on 6200 and 20% OC 6300 is "easily" achievable on 4p mobos.
But what are the "easily" OC numbers of retail 6168-6174 chips?

Also, the OC results are linear with CPU multis or i can expect that lower multi CPU reach ( sligthly) higher refclocks? using other words, i want to know if there is good reason to go for 6180s in place of 6174s...
 
Grats on 6200/6300 OC support.
Tear said that 15% OC on 6200 and 20% OC 6300 is "easily" achievable on 4p mobos.
But what are the "easily" OC numbers of retail 6168-6174 chips?

Also, the OC results are linear with CPU multis or i can expect that lower multi CPU reach ( sligthly) higher refclocks? using other words, i want to know if there is good reason to go for 6180s in place of 6174s...
6100 series aren't generally as consistent as later gens, however, generally, the lower
power the chip, the higher refclock you can shoot at.

Nevertheless --
20% should be easily achievable with low power chips (6164 HE, 6166 HE).
15% should be easily achievable with standard power chips (6168 through 6176)
High power chips (6176 SE, 6180 SE) vary in this department quite a lot (8% to 15% from
field data).

These numbers assume enthusiast cooling (120mm Noctuas, 212+, water).
Also, note that 6164 HE at refclock of 260 is a "mere" 2.21 GHz (~stock speed of 6174).

I'd generally avoid 6180 SE chips, getting them to run at 230 required hand-picked chips.
OTOH, tjmagneto's "random" set of 6176 SE chips runs at 230 with no hassle whatsoever.

I'd say standard power chips offer most value even though they aren't top performers.

So... chip choice depends on your objectives.

HTH,
tear
 
dang 20% Oc on a server chip, sheesh!

33%-40% is considered a bad OC for a X5650 on EVGA's SR2;)

what i like about G34 is the massive core count and memory capacity, coupled with retail availability of new Mobos for sale.
 
Find some ES chips, they range at stock speeds of 1.4-1.7 depending on flavor, and often hit 3.0 or more, again, depending on flavor.
 
Some questions out of pure curiosity:

Voltage tweaking? Is it there? Some PD 8350s can pull over 4.5Ghz, if one was willing to dump a ton of volts through them. If one has the right cooling (water or high-power air) shouldn't we be able to at least get close to this on their server equivalents?

Multiplier adjustment? Some server chips can apply AMD's Turbo Core, can we adjust the whole chip to run at the max Turbo Core multi with this firmware?
 
Retail, none.
ES, yes. Wide open, careful no safety rail to stop you, you can break things badly.
HE, some, you won't break anything.

With TPC you can adjust by cpu, so for example your could do:
CPU0=1.2
CPU1=1.21
CPU2=1.03
CPU3=1.2

Depending on board can determine optimal sockets to use for a cpu that might see too much vdroop.

There are also methods to test a single cpu at a time, rather than all running and chasing down errors/making adjustments, you minimize it to a single cpu at a time. Do one thing at a time, document, document, document. Should you want to use the forums or reach out to someone, it will help. I believe the fahdiag within the [H] DC image will pull hardware errors as well as enviromental information, and dumps the results to pastebin.
 
Also, I know there is a world of difference between the different levels of Mainboard, but is there any plans for dual-socket board support? I would love to tinker around with some AMD Opterons (lack if OCing is pretty much the only thing holding me back from purchasing a system) but quad-CPU boards (and if you choose: the chips to populate them) are quite a big investment...

Also, much more unlikely, but is anyone working on an equivalent for the dual socket 2011 Xeon E5/ E7s? I know their Bclk is pretty untouchable, but I've heard some have success bumping up the BCLK to 110 (10% OC) on prosumer single-CPU boards using retail Xeon E5s.
 
If you are running ES chips you don't need OCNG per say as you'd use TPC to change the multi.

'd be curious to know how they are bumping them to 110 BCLK.
 
Linus Sebastian managed a 110+ on a 12(?) core E5 on an Asus workstation board in one of his YouTube videos, it makes me optimistic that, were the option available, we could be looking at 20 and 24 intel cores in the upper 3's.
 
Is he sharing any of that info? I know on certain chips have gotten a minor OC and v2 chips responded better. Components on the board would flake out, like sata, then requiring the use of a USB> Sata adapter.

I remember seeing something once about a guy who worked for SuperMicro having one that did more than 110. Again, v2 chips, and I am sure he had some "inside" help. I think he was doing cine benches.
 
Multiplier adjustment? Some server chips can apply AMD's Turbo Core, can we adjust the whole chip to run at the max Turbo Core multi with this firmware?
No, but you can consistently run on all-core turbo (something these chips don't do well
without tweaking). Though note that, as sc0tty explained, ES chips are multi/VID unlocked
so using OCNG5 with them isn't absolutely required.

Also see this: http://hardforum.com/showthread.php?t=1765747

KazeoHin said:
Also, I know there is a world of difference between the different levels of Mainboard, but is there any plans for dual-socket board support?
If you can ship a board to the old continent, I can see what I can do.
 
Is he sharing any of that info? I know on certain chips have gotten a minor OC and v2 chips responded better. Components on the board would flake out, like sata, then requiring the use of a USB> Sata adapter.

I remember seeing something once about a guy who worked for SuperMicro having one that did more than 110. Again, v2 chips, and I am sure he had some "inside" help. I think he was doing cine benches.

https://www.youtube.com/watch?v=khDsbxa5_G0
 
any plans to put this bios together for boards like the H8DGI ?

dual G34 supermicro boards, that is.
 
Find some ES chips, they range at stock speeds of 1.4-1.7 depending on flavor, and often hit 3.0 or more, again, depending on flavor.

For soem weird reason 6180s are finally lowering the price, so i am torn between 4x6180 or 4x62xx ES. From what i learned here 6180s reach 2.75GHz with this new BIOS, while 62xx ES run around 2.8-3.1GHZ, but 62xx ES are the power hogs of Opterons, and they can easily burn a 4p mobo to ashes running at 2.8-3.1GHz 24x7.

For my application MC and IL Opterons are more or less equal ate the same frequency, IL higher core count compensating the Bulldozer architecture mishaps, so power consumption and raw frequency at 24x7 settings are more relevant. It does not help that 4x6172 costs the same of a single 6180/62xx ES.

Does anyone with 4x6180 or 4x 62xx ES can report power draw at 24x7 OC settings?
 
4 x 6180s at around 2.75GHz pulled around 800W at the wall if memory serves.
 
4 x 6180s at around 2.75GHz pulled around 800W at the wall if memory serves.

TYVM for the answer. I found one thread that answers the question of IL ES power draw, which is around 1100w

Such more or less point me towards 61xx ES, if i could locate matched sets of 4 CPUs:(
 
TYVM for the answer. I found one thread that answers the question of IL ES power draw, which is around 1100w

Such more or less point me towards 61xx ES, if i could locate matched sets of 4 CPUs:(
I sold my 12 core ES MC's to rvborgh last month.

There are a couple of guys that still have some ES 61xx over to the AMDZone but I don't know if they're willing to let them go.

Anyway if you get non-ES cpu's then the OCNG5 is for you and you'll probably save a lot of $$$ going that way but if you want ES chips then use TPC for OC.
 
Back
Top