![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
2p G34 overclocking
- Thread starter Patriot
- Start date
/dev/null
[H]F Junkie
- Joined
- Mar 31, 2001
- Messages
- 15,182
That is sweet. What board are you using??
???robstar said:
Yeah that's pretty fail..
everuse CPU-Z before?
My guess is an ASUS KGP(M)E-D16
http://www.newegg.com/Product/Product.aspx?Item=N82E16813131643
its on water...
one rasa one ek amd ultimate
its the xspc 220 kit + a 220 swiftech rad + ek amd ultimate block
I can cool just about anything it throws out...
I honestly was not paying close attention to temps...
I stopped at 3.6ghz because I was not comfortable at raising the voltage anymore... dice would probably lower the vcore I would need for a given clock so... that would be the next step if I wanted to continue.
http://s1003.photobucket.com/albums/af155/mffpatriot/Dual Dodeca build/
/dev/null
[H]F Junkie
- Joined
- Mar 31, 2001
- Messages
- 15,182
Didn't notice the board model...I did see the clockspeed though....I'm not really a Windows person either...
It... Its beautiful!(I just came from the awesomeness of that overclock)
Hey not being duck weed here or anything but... not so sure about the awesomeness level of the overclock.
3.6Ghz on a 1100T at 1.56V would not be that cool?
On a coolness factor its an 11 out of 10. Since it's basically 2 1100T's in the same socket and he has 2 sockets on that board.
Have you checked temps, I think your probably hitting 60C on those processors and thats why you can't get more. You could always do a quick temporary test... try sucking out of a bucket of ice water returning to the top etc. That would tell you if you need better cooling.
It would be crazy to hit 4's on that rig.
I have a 1055t that will hit 4 on water ~55c... (single 220 rad)
as far as the dual dodecas... each are dual hexa packages... yes...
they don't have turbo as they were made before the retail chips...
heatwise...
these guys are fine up to 65-68c the rads are not pouring out fire air so I really am not too concerned... these benches take less than a minute to complete so it really has no time to get warm...
heat really is not an issue... though if they were running 20c cooler I could use less voltage to get where I am... such is the nature of overclocking
as far as the awesomeness of the OC...meh...
All I know is ... there is only 1 recorded screen shot prior of someone breaking the 3ghz barrier on these chips....
the chips I am using run at 1.7ghz stock...the fastest ones amd sells is 2.5ghz...
Also... I have not yet made a top windows stable SS run... will do that tonight...
these volts are enough to render stably on...
but for all the doubters I will grab coretemp tonight...
as far as going higher... I just don't really feel like shoving more volts down it...
I have not hit a hard wall yet... I just got tired and switched back to folding before I went to bed..
Last edited:
That is certainly a totally cool setup, wonder why they don't go higher.
I understand that this is a serious overclock for those CPU's but the cores are pretty much the same/// (maybe I'm wrong here).
Getting ones hands on a set of ES 12 cores and pushing them to 1.56V's that's [H]ard|core.
I understand that this is a serious overclock for those CPU's but the cores are pretty much the same/// (maybe I'm wrong here).
Getting ones hands on a set of ES 12 cores and pushing them to 1.56V's that's [H]ard|core.
I imagine there is some pretty serious vdroop with only 6 phases per socket...
I only have an 850w psu... @ 2.9 use 450w load ... 3.2 620 load... 3.6 820load
I can get mid 900w wall draw before I would start worrying about popping my psu...but I would rather not get that close...
Most G34 boards are designed to handle a max ACP of 120W, which in terms of TDP, is around 175W. You can be assured that board won't be able to handle that kind of o/c for very long.
C'mon, Asus, make us a Dual G34 overclocker's board, call it the ROG Assault Formula/Extreme!
C'mon, Asus, make us a Dual G34 overclocker's board, call it the ROG Assault Formula/Extreme!
right... lol...its been taking the 3.1 quite nicely though....
I do have ram sinks on all the chokes and vrms and fans keeping them nice and cool...
plus note... hwbot approved one of last nights run...
http://hwbot.org/community/submission/2158678_patriot_wprime_1024m_2x_opteron_6164_he_59sec_436ms
when I finish rebooting... after taking SS... im gonna take a run at 4 lol
considering I can't increase the v more than 1.55
http://valid.canardpc.com/show_oc.php?id=1780855
is about as good as it gets
http://valid.canardpc.com/show_oc.php?id=1780855
is about as good as it gets
Actually, my previous post wasn't really directed at you, it was directed at tlc4vtm. I was typing it on my phone, and the delay between what I type and when it appears in the message box was starting to drive me nuts.
Actually that's very interesting!!! I've found the 60C mark to be very representative of if something is going to be long term stable. I do alot of prime95 style calculations and if I can't keep the processor below 60 I get errors. This is was true for every PII I had, 550BE etc...
How far can you OC/boot without load???
always boot at 1.7....server board has not ability to change multi...
via an earlier SS you can see I have phenommrstweaker running...
in linux I use turionPowerControl....
a few post back there is another cpu-z validation link for 4.1ghz...
All the benches fail above 3.6 one runs at 3.7 sometimes...
sandra says I have like [email protected]
it uses 860w from wall peak so I may be hitting psu ceiling at higher clocks as well... though if I dice it down it very well may require much less volts to get higher...
validation @ 4.1
http://valid.canardpc.com/show_oc.php?id=1780855
wow 4.1 really...
Then I would really say that if you can get enough current on rail and keep temps below say 55C then you could probably do 3.8ish plus if not 3.9.
Very nice set oif chips you have there considering you have 12 per chip...
Think of it this way if you have a 5series BE PII the chances of uncloking and overclocking to a quad chip are really not that bad. Especially if you reduce the speed from the stock 2.8Ghz down to say 2.2 Ghz etc.
Now take 12 cores rated at 2-2.4Ghz and overclock each of those 12 or 24 for that matter and each of those overclock to 3.6 if not 3.8Ghz.
Wow that is a good overclock.
Then I would really say that if you can get enough current on rail and keep temps below say 55C then you could probably do 3.8ish plus if not 3.9.
Very nice set oif chips you have there considering you have 12 per chip...
Think of it this way if you have a 5series BE PII the chances of uncloking and overclocking to a quad chip are really not that bad. Especially if you reduce the speed from the stock 2.8Ghz down to say 2.2 Ghz etc.
Now take 12 cores rated at 2-2.4Ghz and overclock each of those 12 or 24 for that matter and each of those overclock to 3.6 if not 3.8Ghz.
Wow that is a good overclock.
Most ES chips are low clocked... only ES can be overclocked
90% do not go over 2.6ghz
99.99% dont breach 3...only one other person I have found/read about who has breached 3... SS @ 3.2
these chips are one of a kind...4.1 has no semblance of stability... I recon I would need a 1200w psu and phase change cooling for stabilizing anything round >3.4
this weekend I am going to try and see what volts/temps I need to fold > 3.1
I am really on the steep part of the exponential curve as far as ppd is concerned... 100mhz = ~40s faster on bigadv = 20k ppd
does the Engineer Sample make any difference? would a Opteron picked up retail be able to do this?
ahhh, that info's right above. i take it engineering samples arent easy to come by. unfortunate that no one else will get the joy of oc'ing these beasties.
ahhh, that info's right above. i take it engineering samples arent easy to come by. unfortunate that no one else will get the joy of oc'ing these beasties.
QuiteSufficient
Sufficiently [H]ard
- Joined
- Mar 2, 2001
- Messages
- 5,152
Patriot do you use this setup for anything other than benching?
Seems like it would be pretty bad ass for video/number crunching work
Awesome job btw
Seems like it would be pretty bad ass for video/number crunching work
Awesome job btw
sirmonkey1985
[H]ard|DCer of the Month - July 2010
- Joined
- Sep 13, 2008
- Messages
- 22,414
you can get your hands on them but it ain't easy. AMD doesn't flood the market with ES processors like intel does. there's probably 15-20 ES xeon processors for every 1 AMD opteron ES processor out in the wild.
all of the AMD opteron ES processors are always 1.7Ghz or 1.9Ghz.
the magny cour's are technically pre-phenom II. they are built using 2 x6 Istanbul processors under a single IHS which is what phenom II is built from. and the 5 series unlock just doesn't fit in this comparison. as far as overclocking you have to remember that the power draw is double everything due to the fact that you are increasing the voltage on 2 processors within the same IHS and not 1 and last but not least do not forget that G34 has quad channel DDR3.
Last edited:
Actually it really does you just don't understand the reference. Think about how it applies in this OC instance and post back.
Right I thought that AMD was making 24CPU cores on one chip
And it's not quad channel its dual dual channel.
Also if you really look at the schematic its dual dual dual channel ram for this instance.
except for r-types 2ghz dodeca
R-Type
[H]ard|DCer of the Month - October 2011
- Joined
- Mar 6, 2006
- Messages
- 2,801
Gah, I really need to know if my board will post with two different speed chips. I found another ES...
R-Type
[H]ard|DCer of the Month - October 2011
- Joined
- Mar 6, 2006
- Messages
- 2,801
Eh, I don't think I did. Its actually an SE which is the high power variant.
The guys at AMD zone confirmed that G34 will boot with mismatched cpu's and align the P-states to the lowest common denominator. What I still don't know is if I can clock a retail chip back up to its default P0 state once in OS using the same tools I do to OC the ES chips.
I.E. If I combine my 2ghz ES with a 2.5ghz retail they will boot at 2ghz but I know I can clock the ES back up. I just need to know if I could clock the retail chip back up to its default speed of 2.5ghz.
The guys at AMD zone confirmed that G34 will boot with mismatched cpu's and align the P-states to the lowest common denominator. What I still don't know is if I can clock a retail chip back up to its default P0 state once in OS using the same tools I do to OC the ES chips.
I.E. If I combine my 2ghz ES with a 2.5ghz retail they will boot at 2ghz but I know I can clock the ES back up. I just need to know if I could clock the retail chip back up to its default speed of 2.5ghz.
A. TDP is 140W
B. Don't expect G34 enthusiast boards. We've been through this dozens of times. There is no market for it that would make it profitable and we don't support overclocking on server CPUs.
No, you cannot overclock a production part, everything is locked.
Just a thought on this as someone with 14 dodecas now, pre-Interlagos it would have been interesting to have a feature that would raise the FSB by 1MHz to 3MHz. Some components are very sensitive to >3MHz in a server setting so I couldn't imagine more than that. Multiplier based-"turbo" is the superior path which is what newer generation CPUs are or will be using making this feature moot. Plus, 99%+ of users would likely pass on this feature so while a nice to have, it is probably something that AMD would need to market to make it used even slightly in practice.
With that being said, the HE chips run so cool that it is awesome to see 48 cores in a system running at 430w at load. A thought for AMD is to look at cores/ watts in terms of running systems versus static (e.g.( ACP or TDP) / core.) That is a much more compelling story. I have been seeing ~217w at 100% load with 2P systems making both well under 10w/ core.
Sorry just wanted to correct your last statement.
Sure there is no profit in it for AMD to make G34 enthusiast boards and processors.
But what AMD is affraid of is it cutting into other market share, or people using BE server cores in actual servers. Nobody would buy a BE 12 core server core for $2500+, it would have to be in the 600's and that means no profit for AMD.
What AMD would have to do is release these CPU's and disable some things that make them undesireable in the server market.
They could enable multiplier unlocking while
Disabling all virtualization
Disable the use of ECC
Disable power now and turbo have it run at a fixed voltage of 1.35V or higher
I'm sure people that deal with servers would know more things to disable
The Opteron 6180 is around $1,600 at retail not $2,500. Pricing is one of AMD's big competitive advantages.
Frankly, server folks are less likely to overclock because if a machine freezes due to overclocking, the cost to the business is likely many times more than the more expensive CPU.
True enough a 12 core BE wouldn't be 900 more than a 6180 but even a 10% premium.
At $1800 is where AMD would have to price it to make sence to them a little out of reach for most.
Here is how I came up with my price that I'd pay.
They would have to be in the same range as twice an 1100T with a little extra to make sence to hardcore overclocker geeks.
Typical price of a X6 1100T BE is around $230
So $230 X 2 ( for 12 as opposed to 6 cores ) X 120% (a premium) = $530
I'd probably pay that easy, I'd even go into the 600's but not more.
It's nice to dream of the compatibility of the socket a days.
Where you could use laptop bartons in a dual socket server board... sure you neede to pin mod but that was the fun.
At $1800 is where AMD would have to price it to make sence to them a little out of reach for most.
Here is how I came up with my price that I'd pay.
They would have to be in the same range as twice an 1100T with a little extra to make sence to hardcore overclocker geeks.
Typical price of a X6 1100T BE is around $230
So $230 X 2 ( for 12 as opposed to 6 cores ) X 120% (a premium) = $530
I'd probably pay that easy, I'd even go into the 600's but not more.
It's nice to dream of the compatibility of the socket a days.
Where you could use laptop bartons in a dual socket server board... sure you neede to pin mod but that was the fun.
That pricing does not really make sense in the server market. Looking at the avoided costs of another few U of rack space that is not present in a desktop setting helps support higher pricing. Also, one needs to look at the fact that in large systems, consolidating physical machines or completing work faster is worth money. For an example, look at the F@H results Patriot mentioned where one gets more bonus points the faster a work unit is completed. For example I have two DP Opteron 6166 HE systems that get a maximum of around 104K ppd (208K combined) while taking the chips and putting them into a 4P platform yielded maximums over 270K ppd.
When one is doing heavy calculations for things like automated trading and financial market analysis applications, 1ms faster transactional or analytic speeds mean big money. Server CPUs are priced like this and Intel charges much more for 4P CPUs rather than 2P counterparts.