AMD Magny-Cours Multicore Magnificence

[fusion238]

AMD early engineering sample that someone from Germany demo'd pretty amazing!



AMD Magny Cours Stock Clock

AMD Magny Cours Sample Overclock At 1.037 Volts

 

[alyarb]

it would've been believable if you hadn't shown us the overclock. until we see more specimens, a 45nm 12-core from AMD at 3.2 GHz with 1 volt is preposterous (45nm dual cores can't even manage it).

 

[Idontcare]

Thing is CPUz validate doesn't capture/document/report the Vcore.

So I'm a little curious why the Vcore is shown in those CPUz validate shots...

edit: and yet there it is, direct from the database, so I guess they changed that sometime in the last 3 months or so?

alyarb - 3.2GHz for dual-Istanbul's does seem a stretch...only because I am wondering what kind of TDP that would represent. What's the TDP for a single Phenom II X4 at 3.2GHz? 95W? x3 then!

 

[Yukmouth]

CPU'z is not always reliable when it comes new release CPU's (always seems they need an update).

Maybe it's just not capable of reading correct V-Core on a 12 cpu processor yet?

 

[alyarb]

no. CPU-z never reports the vcore in a validation. the fact that it was added arbitrarily for this image implies that it is fake.

 

[formulav8]

I don't think cpuz would have had "Magny-Cours" in the database yet. So this does appear to be fake to me.



Jason

 

[taltamir]

Originally posted by: Idontcare
Thing is CPUz validate doesn't capture/document/report the Vcore.

So I'm a little curious why the Vcore is shown in those CPUz validate shots...

edit: and yet there it is, direct from the database, so I guess they changed that sometime in the last 3 months or so?

alyarb - 3.2GHz for dual-Istanbul's does seem a stretch...only because I am wondering what kind of TDP that would represent. What's the TDP for a single Phenom II X4 at 3.2GHz? 95W? x3 then!

well, we need SOMETHING for the 1.2 KW PSUs to power

 

[Idontcare]

Originally posted by: alyarb
no. CPU-z never reports the vcore in a validation. the fact that it was added arbitrarily for this image implies that it is fake.

That's what I thought too, but it looks like that changed with CPUz 1.52. Randomly flipping thru the validate database anything validated with 1.52 has Vcore logged, anything 1.51 and older doesn't log Vcore.

Still though, 1.037V for 3GHz is obviously in error, so even if the validation is correct we have no idea how practical that clockspeed was or how anything about cooling (LN2 for all we know).

 

[taltamir]

why is it obviously an error?

 

[Idontcare]

Originally posted by: taltamir
why is it obviously an error?

Know of any AMD 45nm-based products that operate at 3.2Ghz on <1.1V?

Let alone two of their largest such chips (higher probability of within-chip variation necessitating even higher Vcc/GHz) operating on such a shmoo curve?

Now does this rationale rule out the possibility of some cherry-picked combination of two of the world's best Istanbul chips which for some reason can operate at 3.2GHz with a scant 1.037V? No, not at all, if you are willing to accept anything as a possibility then I suppose you can't rule out a CPUz screenshot of MagnyCours running on a mere 0.3V either.

But if the voltage isn't indicative of an actual SKU that is going to be sold, meaning 1.037V for 3.2GHz is a typical Vcc at that GHz for Magny-Cours, then the specific Vcc is irrelevant anyways as it holds no meaning to anyone who is going to buy one when they are finally sold.

Given what we have seen for Vcc vs. GHz for AMD's 45nm, 1.037V is an idle voltage with CnQ enabled, not a full-load voltage for 12 cores operating at 3.2GHz.

 

[taltamir]

i was thinking cherry picked engineering sample with monster cooling. and as of yet unrevealed potential improvements to process while at it... rather then fake picture or misreported voltage...

i learned not to care about anything other then non engineering sample tests done by trusted sources.

 

[Idontcare]

Originally posted by: taltamir
i was thinking cherry picked engineering sample with monster cooling. and as of yet unrevealed potential improvements to process while at it... rather then fake picture or misreported voltage...

i learned not to care about anything other then non engineering sample tests done by trusted sources.

Yeah I'm not into including those possibilities when I apply my occam's razor. Doesn't make me right, just means I acknowledge those possibilities have non-zero probabilities but they fail to merit consideration when I personally contemplate the situation.

We've experienced countless occasions in the past where new chips had their voltage get completely misreported by cpuz in the screenshots.

Wasn't there a thread just a couple weeks about regarding westmere and cpuz screenshot showing under-reported voltage? Something equally silly like 1.01V at 5GHz or some such?

 

[formulav8]

Maybe AMD's new optimized 45nm process with high-k, ect.. that AMD is supposed to bring out this year allows such a clockspeed at that voltage. But I still think the pics are a fake.



Jason

 

[Idontcare]

Thread here started by the guy who cpuz validated the chip(s):
http://www.xtremesystems.org/f...howthread.php?t=233565

edit: cliffs after reading the XS thread, that 3.2GHZ cpuz was bogused by accident, voltage is correct but that wasn't 3.2GHz for all cores. Doing all cores he get's 2.6GHz to be Wprime stable but is thinking it is thermal throttling. At 3GHz he is convinced he is thermal throttling because the wprime scores don't scale well with the clockspeeds at that point. ETA for retail is Q1 2010.

 

[drizek]

Originally posted by: IdontcareWasn't there a thread just a couple weeks about regarding westmere and cpuz screenshot showing under-reported voltage? Something equally silly like 1.01V at 5GHz or some such?

Are you talking about the clarkdale one? Wasn't that something to do with the CPU undervolting at idle? THe voltage at load on that one seemed about right.

Anyway, if this is true that it can run 3.2@1.1, AMD is in a really really good place right now. Hell, 3.2ghz with 12 cores at any voltage is really impressive, especially when you consider that it is only 2 years since the launch of the original Phenom.

 

[alyarb]

wait, magny-cours is MCM? here i thought all along it was a 12-core 45nm SOI monolith. so you can understand my skepticism.

are these (istanbul) the first of globalfoundry's 45nm HKMG? that would explain why the duals and quads can't undervolt as well above 3 ghz. if they are going so far in 1.52 to report vcc, then cpu-z should also run some FFT's during a validation so that we actually get the correct voltage.

if it's throttling at 2.6 he's got to be on air cooling, or he has an impressively custom made wide-aspect waterblock for socket G34 or whatever it is.

 

[VirtualLarry]

Originally posted by: Idontcare
edit: cliffs after reading the XS thread, that 3.2GHZ cpuz was bogused by accident, voltage is correct but that wasn't 3.2GHz for all cores. Doing all cores he get's 2.6GHz to be Wprime stable but is thinking it is thermal throttling. At 3GHz he is convinced he is thermal throttling because the wprime scores don't scale well with the clockspeeds at that point. ETA for retail is Q1 2010.

AMD CPUs thermal throttle now, like Intel chips? I remember their 65nm chips, they would just shut down if they got too hot.

 

[eternalone]

Behold the magni core!!! WOW

 

[taltamir]

Originally posted by: Idontcare
Thread here started by the guy who cpuz validated the chip(s):
http://www.xtremesystems.org/f...howthread.php?t=233565

edit: cliffs after reading the XS thread, that 3.2GHZ cpuz was bogused by accident, voltage is correct but that wasn't 3.2GHz for all cores. Doing all cores he get's 2.6GHz to be Wprime stable but is thinking it is thermal throttling. At 3GHz he is convinced he is thermal throttling because the wprime scores don't scale well with the clockspeeds at that point. ETA for retail is Q1 2010.

so @ 3.2 @ 1.01v... it hits 100 degrees C and auto downclocks?

 

[Idontcare]

Originally posted by: taltamir

Originally posted by: Idontcare
Thread here started by the guy who cpuz validated the chip(s):
http://www.xtremesystems.org/f...howthread.php?t=233565

edit: cliffs after reading the XS thread, that 3.2GHZ cpuz was bogused by accident, voltage is correct but that wasn't 3.2GHz for all cores. Doing all cores he get's 2.6GHz to be Wprime stable but is thinking it is thermal throttling. At 3GHz he is convinced he is thermal throttling because the wprime scores don't scale well with the clockspeeds at that point. ETA for retail is Q1 2010.

so @ 3.2 @ 1.01v... it hits 100 degrees C and auto downclocks?

He never hit 3.2GHz for all cores or for any benches, he fux0red the overclock software by accident and only got one core to that speed.

When he got the overclocking software squared away so it was boosting all cores he only went to 3GHz and even then he was quite convinced the cores were self-throttling based on the rather poor wprime scaling results from 2.6GHz -> 3GHz.

Not sure on the exact temps, but that is the guy's conclusion (the guy with the magny-cours rig at the heart of all this, its 2S too by the way).

Actually he says somewhere in that thread that he thinks even 2.6GHz was overheating somewhere near the end of the Wprime run. No idea what his cooling is actually designed for, if it is stock coolers then this might not be a problem come volume sales time with beefier heatpipe HSFs and so on.

Originally posted by: alyarb
are these (istanbul) the first of globalfoundry's 45nm HKMG? that would explain why the duals and quads can't undervolt as well above 3 ghz. if they are going so far in 1.52 to report vcc, then cpu-z should also run some FFT's during a validation so that we actually get the correct voltage.

AMD and GF stopped talking about HKMG being a 45nm deployment option over a year ago, its no longer on the roadmaps. As far as I know HKMG is pushed out to 32nm.

Not altogether surprising, a lot of industry folks were left scratching their heads when AMD first talked about re-releasing 45nm on HKMG just because of the enormous task and expense involved in qualifying a major major process change like that.

It can be done, there is nothing impossible about it, but to justify the resources and timeline to do it is what never really made sense in light of the other more pressing things they could be doing with those same limited resources (like getting 32nm release pulled in by 3-6 months).

Have you heard or seen any recent discussion/roadmaps from AMD or GF's on the topic of 45nm HKMG?

As far as I know its a dead topic for AMD. That's not a fact, that's just my impression of the matter given the utter lack of discussion on it from AMD and the fact it was dropped from the process technology roadmaps.

 

[alyarb]

that's what i thought, but formulav8 had me thinking otherwise and i just went with it. i knew there were plans to retool mid-cycle, just didn't know they'd done/not done it. anyway who cares about 45 anymore. 45nm SOI does suck though. not exactly the boon it was on 130nm.

 

[Idontcare]

Yeah its one of those things that AMD expended a lot of effort to hype in the very early days of their 45nm release as a means to appease the shareholders and justify why Intel has HKMG but AMD had missed the ball on that one. A diversionary tactic, bait and switch.

Now that no one is clamoring for the heads of the decision makers given how well 45nm w/o HKMG has done for PhII and Shanghai/Istanbul there's really no need go back and dwell on the subject.

We saw this same tactic in the run-up to releasing the original quad-opteron with those "40% better than clovertown" comments which they never ever went back and substantiated after the release and everyone was left going "huh!?".

Don't get me wrong, I think it would be absolutely fantastic if AMD figured out a way to re-release 45nm with HKMG (requires redesign of any chips to be produced on it, plus reverification of the chip and requalification of the node) without negatively impacting the 32nm node and product's timeline and capabilities...I just don't see it happening given AMD's resource situation.

 

[alyarb]

not everyone sees 140 W Phenoms to be the shortcoming they are, but at least they are certainly getting notoriety for their 45nm lineup, with or without high-k. i think 32nm will really get them back on track with intel (efficiency-wise), even if still a few months behind. look at what high-k has done for POWER7 compared to POWER6. You go from 3 ghz @100W to 5 GHz at 160W. that is a complete departure from the normal power-frequency relationship, all on a chip twice the size of power6. is IBM's high-k better than intel's in some way or can you attribute this entirely to pipelines of differing lengths? perhaps the frequencies, but the tdp?
One would expect the improvement in efficiency to be universal over SiO.

 

[Fox5]

Originally posted by: alyarb
not everyone sees 140 W Phenoms to be the shortcoming they are, but at least they are certainly getting notoriety for their 45nm lineup, with or without high-k. i think 32nm will really get them back on track with intel (efficiency-wise), even if still a few months behind. look at what high-k has done for POWER7 compared to POWER6. You go from 3 ghz @100W to 5 GHz at 160W. that is a complete departure from the normal power-frequency relationship, all on a chip twice the size of power6. is IBM's high-k better than intel's in some way or can you attribute this entirely to pipelines of differing lengths? perhaps the frequencies, but the tdp?
One would expect the improvement in efficiency to be universal over SiO.

Someone posted about this before. I think IBM's 45nm high-k was about equal to their 32nm high-k and intel's 32nm high-k. It seems like performance improvements from shrinking may be coming to an end.

 

[taltamir]

more like the nm label is an arbitrary label that doesn't really mean anything. a 45nm chip has its smallest component at 4nm and many at over 100nm, and there is no specific component that is at 45nm from which the name comes.
So sometimes the arbitrary label implies more of an improvement than what really happened.