Printable View
INSANITY !!! :shocked:
http://img248.imageshack.us/img248/3637/cpuz32.jpg
Check out Stephen's (Xtreme Legend) thread on page 3: http://www.xtremesystems.org/forums/...d.php?t=233565
Now who wants to guess how this crunches :eek: :D
UPDATE: More MADNESS: :eek:
http://img513.imageshack.us/img513/7592/wprime29b.jpg
Perkam
holy!!!! dam 12 cores! :cool:
Holy crap! 12 threads @ 3.2ghz w/ only 1.037v?!?!?!?
How does this hold up to other chips (phenom II, i7/5, C2Q) clock for clock?
And whats up with 10mb for L3? Isn't that an error or something with the way the system is utilizing the L3 cache, so it only shows up at 10 instead of 12?
Nice stuff. ;)
10MB L3 is due to probe filter functionality that can be turned off.For 2P recommended setting is off.Quote:
Originally Posted by [XC] Synthetickiller
At 3.2GHz the system was not actually stable but it managed to finish some tests. The cooling was pretty weak(standard opty coolers) and volts were very low which could be the cause of instabilities. s7e9h3n did a great job.
Still,12 cores in one chip @ 3.2GHz with that vcore and that cooling is amazing. That's 2 Lisbon dice directly connected with a lot faster(than usual) ccHT link. The final clocks when it hits the market should be 2.2-2.3Ghz,not much but will still be done within 95W TDP envelope.
Weird thing is, it seems the info on this chip has been stuck in a little bubble in the AMD section on this forum. I don't see info anywhere else on it. I am looking forward to more results/info when he returns.:up:
Dual socket 12 core, 24 in total?
Yep! :DQuote:
Originally Posted by spicypixel
More Ram
Multiple 58xx series GPUs
some SSD's for kicks
More volts and cooling.
Then win Vantage, or any other benchmark you desire. (aslong as it scales)
Holy sh12, 12 Cores in one chip? Thats pretty cool AMD, Keep it up.
do want.
I'd like to see those on the new F@h work units that give extra points for turning in big WUs quickly... :D
Yep these new CPUs are pretty amazing :O Cant wait for the new server boards too so I can finally OC my dual Shanghai and soon to be dual Istanbul system. Its these current NForce 3600 boards thats holding us all back now. Stephen proved how great the new chipsets can be as well as the CPUs.
yeah! kudos to Perkam for creating the thread here 'cos it will most definitely reach mainstream news sites toQuote:
Originally Posted by charged3800z24
I want a six-core for my AM3 desktop! :eek:
i could imagine less than a day.:shocked: my stupid nehalem takes almost 3 days.Quote:
Originally Posted by Rise
u must more than 1/2 year waitin at Thuban, or find ES in early time...
That is a disgusting amount of information processing capacity!
Is it Prime95 stable for 24hrs and does it get 7.9 in the Windows 7 Performance Index?
Seems likely that someone with two ES 12 core CPUs and an ES mb would try to run Prime95 (12 instances) for 24h to validate stability. Or not. It's not said to be stable either. But with those volts and that cooling it's nothing less than very impressive.Quote:
Originally Posted by initialised
In the desktop environment we don't need 12 cores. Most user don't even need 4. But its still very cool to see 12 cores on a single chip.
And it does wprime @ 3Ghz
These are server chips... Some folders seem to drool over them though :p:Quote:
Originally Posted by vengance_01
No, but it's only a matter of time until we have them.Quote:
Originally Posted by vengance_01
Wow, with a bunch of magny cours one person could make a dent into FAH or WCG.
Imagine the crunching/video encoding power of MC with larrabee (if it's compatible)
We could do all the protein folding required at ten times the speed of just 5 years ago. :yepp:
I don't know where Shintai is now,but he hasn't done any WCG work for a long time. His "xtreme socialist" title was a bit misleading.Quote:
Originally Posted by Chumbucket843
On the Magny C. topic, s7e9h3n said in his thread that he will be back soon(Terminator™) ,so expect more in a week or so.
very impressive
Quote:
Originally Posted by [XC] Synthetickiller
clock for clock is 99.9% similar in single threaded apps
maybe will be a few percent faster in multi-threaded apps, thanks to HT Assist
low operation voltage thanks for the increasingly matured 45nm manufacturing process; nothing to do with the architecture
Aww you just had to post it in the news section did you? Good thing comes to those who search for it (it was in the AMD section for a looong time)
Go MCM... Began has the CPU core wars
While 3.2GHz is not impressive as a number, the specs of the CPU are damn impressive. 12 cores, 3MB L1, 6MB L2, 10MB L3. :D
Fixed.Quote:
Originally Posted by gOJDO
Or 10MB L3 depending on if probefiltering is off/on ;)Quote:
Originally Posted by AKM
Early ES and 3.2Ghz overclock is not impressive? Yeah the word overclock alone is amazing on this platform. And the voltage is really low too. I say you dont fully understand the meaning of this number here.Quote:
Originally Posted by gOJDO
Impressive stuff. Waiting for battle of giants (aka Magny Cours vs. Nehalem EX).
fixedQuote:
Originally Posted by gOJDO
[Since each core has 64KB L1 Data, 64KB L1 Instruction, 512KB L2 Unified, and each die has 6MB L3]
Is it really low voltage? as in...REALLY?
Its not as if CPU-Z voltage monitors have EVER been inaccurate..... :rolleyes:
Quote:
Originally Posted by nn_step
Again :p:Quote:
Originally Posted by Smartidiot89
If you would take time to read the thread posted by S7, you would know the answer. And that it was done on stock (1.1v). http://www.xtremesystems.org/forums/...4&postcount=61Quote:
Originally Posted by K404
All attempts to downplay this result fail.
the amount of physical L3 is not changed, only the amount that is free for allocation changedQuote:
Originally Posted by Smartidiot89
Thus to say a total of 12MB of physical L3 is accurate but if you would like we could always make the distinction of the logically available L3 [Which fluctuates significantly during software execution for several reasons such as the shrinking and growing of the probe directories]
6% off from CPU-Z. My point stands.Quote:
Originally Posted by v0dka
It wasnt a comment on the quality of the result
Checking CPU-Z validations- the way most external ppl will see this result- will give them wrong info about the work done. Id rather see CPU-Z not report voltages than do it wrong.
He did it on default vcore(3.2Ghz) and there was slight variation in cpuz readings,that's all.
so we can play Crysis at lastQuote:
Originally Posted by spicypixel
Actually just let the thing mature and get into mass production, give it a proper motherboard and cooling and I could see this thing hitting HIGH. 12 cores @ 6 GHz with LN2? ;)
What? Isn't that like saying a P4 at 4Ghz would be better than a i7 at 3.2Ghz, I don't think so.Quote:
Originally Posted by gOJDO
this is based on instanbul, which is not the same as phenom right? im not sure how different the architecture is, i just know its 2 more cores. so i havnt a clue how well we should see these OC if we look at phenom numbers
It's based on Lisbon,a slightly different revision from Istanbul. There are things that are improved/added when compared to Shanghai,but details are scarce atm. OCing should be similar to desktop(deneb) cores if not better,but we need open multiplier for serious OCing which will come next year with Thuban desktop version.
hmm, good news. now i wonder if this will be enough for me to switch away from my 940/ddr2 setup, or if i should stick and wait for 32nm 8-12 core chipsQuote:
Originally Posted by informal
Noone noticed the wPrime numbers?
1024M - 75.768s
32M - 3.539s
For those unfamiliar with wPrime numbers, I quote mr chew*:So close to 1 sec faster for 32M and more than 40 sec faster in 1GB :rofl::rofl:Quote:
Originally Posted by chew*
Speaking about a long time, i'm surprised Movieman didn't show up in the AMD section thread, hope the poor old man's heart didn't let him down after seeing that task manager screenshot :DQuote:
Originally Posted by blindbox
I think we've all forgotten something... This is an engineering sample, i don't think the official ones will have an unlocked multiplier.
Do people think this is gonna be released as a desktop processor, or less than $2,000?
AMD must be getting nervous about Beckton. Should be some extensive demos in two weeks at IDF.
Maybe AMD rents a hotel room ;) Haven't they done that before?Quote:
Originally Posted by terrace215
Doubt it, but this is an enthusiast forum. If Intel can dish out $1000 i7-975 McNothin specials people would likely pay 2k for 12 cores with a lot of fun potential. I mean you can take a $300 chip to OC to Intel Core i7-975. Having 12 cores is a niche and a damn fun one when you can OC to twice it's stock speed.Quote:
Originally Posted by Chad Boga
you think the 12 cores will come stock at 3ghz, i doubt it, more like OC to 3x its speed (with some LN2)Quote:
Originally Posted by Glow9
No they come stock at 1.8Ghz I believe the original thread said.Quote:
Originally Posted by Manicdan
I don't think anyone is expecting 12 cores at 3ghz, but realistically a 2.4 to 2.6ghz part is quite doable. Maybe 3.5 (+/- 200mhz) on a very good air cooler with a little voltage! :)Quote:
Originally Posted by Manicdan
and what does that has to do with this thread about performance and overclocking flexibility of AMD next gen K10 core(s)??Quote:
Originally Posted by Chad Boga
Quote:
Originally Posted by Glow9
i was trying to say the OC would be 6ghz if people would get double the speed of stock. i expected these to be around 2ghz for tdp reasonsQuote:
Originally Posted by freeloader
Understood! :yepp:Quote:
Originally Posted by Manicdan
Would be nice to see a stock desktop part at 3ghz under 125 watts. Maybe on a 32nm node. :D
Yes, with a 5870 integrated on-die, for less than $300, please!
It's a 24 core system to be more precise.There won't be any 1p systems on the g34.
I know that OP knows this,but seeing how people get excited
about all the cache/cores ---well we better double all this excitement up.:up:
Correct me if I am wrong but both CPU's can access each other's L3
(among other things),and the maximum memory bandwidth goes up a lot when 2 or more sockets are used.I am counting 24MB of L3 btw (not 10 or 12 :p:)
Anyhow I don't know where am going with this,but I do know that this should make a nice "entry level" system :shocked: :rofl:
Also why people think it will cost thousands for a single CPU ?
The 4P/8P will,but not the 2P ones.I assume there will be differences in the
number of HT links and the price (similar to how it is now with the 2xxx and 8xxx series Opty's))
is it really 1v vcore? :o
Probably not, CPU-Z sometimes misreads vcoreQuote:
Originally Posted by saaya
It's around 1.15V.Quote:
Originally Posted by Clairvoyant129
Go AMD, Sweet!
Someone pointed me to this thread...Thanks Perkam - haven't made the news here in quite awhile.
Yes, it really is 1.037Vcore defaultQuote:
Originally Posted by saaya
Actually, for runs @ ~3Ghz, all the cores were set to run at 1.165Vcore. I was able to figure out the value for the P-State register which allows for Vcore increases. IIRC, I even had 1.5+Vcore available but didn't even think twice about it for fear of melting the coolers.Quote:
Originally Posted by informal
I'll try to address some of the other questions/comments later - I have a bad habit of reading threads from back to front...:p:
yes, it was about 1.15V only, for SOI nice :)
sweet :DQuote:
Originally Posted by s7e9h3n
I've been looking at this and here it is compared to some new Intel I9 cpu.
AMD Magny 3Ghz 24 cores - 24 threads:
32M - 3.539 sec
1024M - 75.768 sec
Intel i9 3,8Ghz 6 cores - 12 threads:
32M - 4.9 sec
1024 - 140.814 sec
I think it's quite impressive.
If this could be pared up with better oc, some good ddr3, few ssd's, 58xx it would be overclockers dream :)
Here is the screenshot for intel cpu
http://www.xtremesystems.org/forums/...8&postcount=50
I suppose that AMD's result was obtained on dual-cpu platform, so actually there were 24 threads, not 12. Also Core i9 is not a competitor to Magny Cours since i9 is a desktop CPU and probably will at least twice cheaper then a cheapest Magny Cours (without considering matherboard prices). Nehalem-EX (8 cores, 4 channel mem, 24MB L3) is a direct competitor of Magny Cours.Quote:
Originally Posted by LaMpiR
Still new AMD platform has very good bandwidth per socket with its 4 mem channels.
I know you can't compare those to directly but still. Idea that it's dual socket crossed my mind also but didn't know that results were on 24 threads.Quote:
Originally Posted by kl0012
Tnx for clarifying ;)
Also there was some thermal throttling at OCed clocks due to heat and results in wprime 1024 are not scaling very well.With better cooling(non stock) that issue could be cleared up.
Isnīt it 24 cores / 24 threads (@ 3Ghz) ? (12 cores per processor)Quote:
Originally Posted by LaMpiR
edit: 24 cores...
http://img441.imageshack.us/img441/4831/wprimetask.jpg
:shocked: This thread should have a XXX rating. :shocked:
That's unreal. You'll never have to buy another cpu again with that much raw power.
I really like the fact that this CPU can do 3Ghz at such low voltages. It is said that AMD continually improves their process and it shows here. We can probably expect good things from C3 Phenoms and HKMG in 2010.
If this CPU can overclock at 4Ghz, I wonder how many voltages it requires. But yes, overclocked at 3Ghz with that low voltage is quite impressive.
bigjezus...4p system, with these...48 cores...:shocked:
a BOINCmonster
Thinking about how jcool has a true 4x4 system right now (based on 8347HE that was released on May 12, 2008), this would make it an almost viable option in about 2 years time...
In 2 years time, or by September 2011, you'll have 22nm shrink of Sandybridge by the name of IvyBridge from Intel with 8-cores 16-threads on desktops, and likely 22nm 12 and 16-core (24 and 32-thread) chips on the server side.Quote:
Originally Posted by anubis
From AMD, you'll have Octo-core Bulldozer chips on the desktop and 32nm 16-core Bulldozer cpus on the server side with CMT (Clustered Multi-threading) and an unknown amount of threads.
I.E. Whatever will be out 2 years from now will make Magny-Cours look like utter trash.
Perkam
Good grief. :rofl::rofl:Quote:
Originally Posted by perkam
and 90% of the programs will just begin to work with quadsQuote:
Originally Posted by perkam
the debate of many slow cores v a few fast cores begins!:argue:Quote:
Originally Posted by perkam
Desktop programs, perhaps. Workstation and Server programs, definitely not. They can scale with the no. of faster cores with newer architectures.Quote:
Originally Posted by Manicdan
Perkam
well you know I always think of it like thisQuote:
Originally Posted by Chumbucket843
two lines for one booth?
line for every booth?
lol silly I know but there always that core with set amount of data to work with. some are more efficient then others.
two lines for booth with normal person?Quote:
Originally Posted by demonkevy666
one line for booth with mentally disabled persons with a dog (HT)?
offensive, I know but it's not supposed to be meant that way. :(
I'd rather have a dual core at 5 Ghz than a quad at 3.4 Ghz. Alot of desktop apps are still single threaded and for every core you add it decreases scaling efficiency.
umm increasing frequency increases power consumption exponentially. scaling with more cores will always be more efficient. how do you think your gpu works? AMD and intel are pushing multicore hard and apps that really need the power take advantage of all available resources.Quote:
Originally Posted by BeepBeep2
dont hate on HT. its useful and i like it.