AMD FX.. Centurion?

[wez]

Just saw this over at hexus, rumor of a limited edition 5ghz FX chip out of the box with a $795 (!) price tag. TWKR II?

Interesting if true, and bound to make some headlines.

Source: http://hexus.net/tech/news/cpu/54057...urion-fx-5ghz/

[Tomasis]

I have thought to post this in Xtreme news.. Seems to be too good to be true.

Lets see some Ln2, helium records

[FlanK3r]

wow, I must have it!

[AMDforME]

It would be cool if AMD does sell some 5.0 GHz. chips but the $795 price tag is Intel-ish though not the truely absurd $995 price tag. Considering an FX-8350 will run 4.6-4.8 GHz. on air and only cost $200. I won't be buying any 5.0 GHz. CPU @ $500+.

BTW, the link doesn't seem to work for me but numerous other sites are reporting this "rumor" which has not been verified by AMD.

[FlanK3r]

if this CPU can stable 5 GHz at stock, what could it do with OC? 5200 MHz stable? Or more? Ln2 8GHz+ at 100%?

[informal]

Can we assume if it can run @ 5Ghz on conventional (good) air cooling that it will be capable of running +10% over that with some tweaking? As others pointed out, with some level of knowledge (and luck with the vishera specimen) one can hit near 5Ghz with OC on current 8350 parts (8320 clearly OC worse). So if it can hit 5.5Ghz on air with additional tweaking then it might be worth it since this will be noticeably faster than any OCed 8350 we have today. If not then it's not worth it

[FlanK3r]

next things is this....If the 5 GHz is limited series, specially for extreme fans and overclockers, this chip must be very good with Ln2. So with low VID (for this clocks).
TDP for consuming part could be up to 140W, so this chip could hit 5 GHz stock clocks at around 1.4V max. Im sure about this chips destroy Haswells in multithread (3.5/3.7 GHz i7-4770K)

[The Stilt]

Unless a new much improved version exists, this is purely a hoax (and not a good one either).

AMD can currently bin OR-C0 (N2) dies which clock significantly higher than the normal (average) parts.
However even within the specially binned batches, dies which can do 5000MHz are very rare.

Issue 1: A 4CU/8C OR-C0 operating at 5000MHz would have TDP & ETP of 180 - 200W+.
The maximum temperature for a FX (in high frequency & power condition) is around 65 celsius, higher than that will retard the clockability.

Let say the ambient temperature of the case is at constant 40 celsius.

TCPUMax (65) - TAmbient (40) = 25 celsius.
TCPUMax-TAmbient-Delta (25) / TDP (180) = 0.1388 C/W RTH (the required thermal resistance of the heatsink).

In case you use a relatively high performance TIM, such as Shin-Etsu X23-7783D (which has thermal conductivity of 6.0 W/m K):

AM3+ heatspreader area = 0.0016 square meters.
TIM layer thickness: 0.000025 meters (0.025mm)

0.000025 / (6.0 * 0.0016) = 0.002604166 K/W or C/W (thermal resistance of the TIM layer).

In total you would need a cooler with minimum thermal resistance of 0.136195834 C/W.
Any suggestions?

Issue 2: "A 4CU/8C OR-C0 operating at 5000MHz would have TDP & ETP of 180 - 200W+."

At 5000MHz the required Core VDD would be around 1.475V - 1.500V.
At the given TDP & ETP range it would mean 120 - 136A of current draw.
The "margin" (IDDSpike) is around 20-25% on AM3+ designs, so the motherboard should be able to deliver 144 - 170A for short periods of time. The power requirements alone leave the selection of motherboards quite narrow.

Since the VRM efficiency on a Crosshair V Formula-Z in these conditions is 77-80% (at most) it means the VRM will dissipate 36 - 46W of heat. That is a massive amount of heat for a relatively small, passively cooled aluminium heatsink to dissipate.

TVRMMax (135) - TAmbient (40) = 95 celsius.
TVRMMax-TAmbient-Delta (95) / VRMETP (36) = 2.63888 C/W (2.06521 C/W at 46W VRM ETP).

Any passive heatsinks you can fit on VRM with such of thermal resistance?
No.

At most AMD could release a special batch which they could say "being capable" for 5000MHz+.
However they cannot guarantee frequencies higher than currently available, simply for the reasons I listed above.

[FlanK3r]

Stilt, so, dou you think, is impossible binned example 1000 chips for this CPU which could give 5 GHz stable with example only 1.4-1.425V?

[Alex-Ro]

binning 1000 chips by amd for one cpu?not going to happen....

[FlanK3r]

enthusiasm is the best way for support this idea, so I want this chip

[AMDforME]

Since we know that some of the FX-8350's can run @ 5.0 GHz. without major thermal issues, then I could see AMD selling a 5.0 GHz. Vishera with their H2O cooling kit for $795. That would make a lot more sense than $795. for just the CPU.

IMO it would not make a good Biz case to try and sell a 5.0 GHz. CPU alone for $795. I would also speculate that the headroom for OC'ing a 5.0 GHz. CPU will not be 10% and is likely to be closer to 5% at best. Thus there would be no issue with the highend AM3+ mobos that offer a proper VRM design that is rated at 140W, which is a nominal operating load more than the maximum operating load.

I doubt the overclockability on LN2 would be higher than the ones AMD has tested already as I'm sure they picked the best of the best for their LN2 use.

[zir_blazer]

Stilt, so, dou you think, is impossible binned example 1000 chips for this CPU which could give 5 GHz stable with example only 1.4-1.425V?

Is not that is impossible to bin them, but you need the platform to give support for them, too. His point is that such a high end Vishera also means new Motherboards to officialy be able to claim support for those 200W TDP Processors, or at least re-validate existing ones to guarantee that they will be able to deliver that power without putting massive stress and possibily damage all the power delivery part of the Motherboard. Is exactly the same thing that when people avoid to use 125W TDP Processors on a Motherboard that claims support for only 95W.

[FlanK3r]

200W TDP is too much...Maybe with 1.5V and 5 GHz. But if your CPU could have low VID and higher clocks...Maybe 140W is real. I have one FX-6300 with VID under 1.2V, but there are FX-6300 with much higher VID ussually. And the same TDP. So....Maybe some binned 1.4V and cold chips...
there could be situation


Its power consumption all PC (without LCD). This piece has only +83W more than stock after OC! And with OC I jumped a bit with VDDA voltage CPUNB voltage too. There could be pieces, with example only +40W more. So maybe some binned form better 32nm process 140W TDP? Or it is really only sci-fi? Good news from graph is the things, Vishera is not so power hungry after OC as classic Bulldozer after OC.

[The Stilt]

Stilt, so, dou you think, is impossible binned example 1000 chips for this CPU which could give 5 GHz stable with example only 1.4-1.425V?

Even now when the 32nm process is more mature, the silicon characteristics are still highly variable.
Binning can make a huge difference as the difference in the leakage alone between two 'identical' dies can be anything between 0-30%.

I know as a fact that AMD has used plenty of resources to map the behavior of dies with different silicon characteristics in different conditions. Including scaling, thermal and power consumption measurements under conventional (and some unconventional) cooling methods. If and how the data was used I do not know.

The specially binned dies on average consume around 15% less power than an average equivalent normal part.
These are all very high quality dies with very low leakage. On average they can scale around 200MHz higher on conventional cooling methods due lower emitted thermal power (lower temperature). They still act like any other die of the same stepping, the critical frequency is still at around 4600MHz.

At this point the scaling slows down significantly and the Core VDD needs to be raised heavily to prevent the scaling stopping completely. At 4800MHz the emitted thermal power has increased to levels which no longer can be managed by conventional methods. At latest at this point the power consumption is also a way beyond the capabilities of a normal mainstream motherboard. Usually the VRM OTP/OCP kicks in way earlier on motherboards with 4-phase CPU VRM. Even motherboards with passively cooled 6 to 8-phase CPU VRM get hot at this point.

Ultimately the maximum frequency of the specially binned dies will still be limited by the thermal and the discrepancy between the compute units.

The scaling under unconventional cooling methods has nearly zero correlation with the scaling under conventional methods. Under unconventional methods the rules are completely different as the leakage is decreases dramatically. So a die with a very low leakage properties does not perform nearly as well as a die with high leakage properties under unconvetional cooling.

A good specimen can be pushed up and beyond 5.0GHz, this is not the issue.
The issue are the conditions which are required to make the chip and the supporting hardware to work RELIABLY at this frequency.
They are simply not possible for an average person to attain. AMD and an enthusiast might also have a slightly different definition for the term of "long term reliability and stability".

Yes, I do believe that it is impossible for AMD to release a 8-core FX part clocked to 5.0GHz (guaranteed).
The current die would require such an overhaul to make it happen that the required efforts would greatly exceed the advantages.
They would basically need to lower the power consumption by 30-50%. If such improvement would be possible in the first place we would already be using these parts.

They could maybe release a special edition (featuring low leak dies) clocked to 4.3 - 4.5GHz (constant & base) and state "support up to 5GHz*" (* being not guaranteed) or so.

A 2CU/4C part could be released at 4.6 - 4.8GHz with 125W TDP, as the thermal is much lower on these parts and the overclockability is much less limited by the discrepancy between the compute units (as there are just two, the strongest ones).

[demonkevy666]

question what if it's 5ghz Turbo core ?

[The Stilt]

question what if it's 5ghz Turbo core ?

It would not make much of a difference.
The CPU power plane is shared between the compute units so the effective Core VDD would still be at the level requested by 5.0GHz Turbo PState. The lower clock frequency on some of the CUs would lower the power consumption just by few watts.

In the current form the Turbo Core is retarded anyway IMO.
It could be much more efficient if it would not just rely on precalibrated values.

[AMDforME]

With some FX-8350's already capable of running at 5.0 GHz., it certainly is possible for AMD to release a guaranteed 5.0 GHz. Vishera based CPU that will be fine on current high end AM3+ mobos. These chips may run @ 5.0 GHz. at moderate Vcore such as ~1.4v-1.75v ??? These chips are unlikely to OC much higher however and they are merely for bragging rights like the $1000 Intel chips that are over-hyped and under-performers. As far as performance there is not much to be had going from 4.5 GHz. to 5.0 GHz. - or silimlar. It may sound like a nice jump and look good in benches but in actual system performance it's only a modest improvement at best. If the CPUs were $300, IMO they might be worth the cost but not at $795, even if supplied with a H2O cooler, which has not been mentioned so far in these "rumors".

Still if AMD decides to sell these I hope they sell everyone they can make and more power to them.

[Utroz]

Is it possible that they are a new silicon stepping or revision that includes the Resonant Clock Mesh tech they got off Cylos?? If not that than a shrink to 28nm is about the only other way I see them being able to make 5ghz 8 core cpu's in relatively commercial size batches with relatively normal air cooling. All this is just my 2 cents...

[FlanK3r]

I heard Resonant clock Mesh is not ideal for so high frequency (in effectivity)....Or Im in mistake?

[sequoia464]

I heard Resonant clock Mesh is not ideal for so high frequency (in effectivity)....Or Im in mistake?

I'm no expert here at all, but I have heard the same as you on the RCM at higher frequencies at several sites.

[justapost]

That headline is abit in contrast to the article itself. "This so-called Centurion is guaranteed to run at 5GHz, on air". I read that as if the cpu can be oced to 5GHz.

[FlanK3r]

So, Stilt had right...this will be some chips at unspecified clocks and working after OC 5 GHz at air stable...But still very good. Somethink as TWKT PII few years ago

[BeepBeep2]

Can we assume if it can run @ 5Ghz on conventional (good) air cooling that it will be capable of running +10% over that with some tweaking? As others pointed out, with some level of knowledge (and luck with the vishera specimen) one can hit near 5Ghz with OC on current 8350 parts (8320 clearly OC worse). So if it can hit 5.5Ghz on air with additional tweaking then it might be worth it since this will be noticeably faster than any OCed 8350 we have today. If not then it's not worth it

I would think "6400+" 140w TDP scenario...maybe 200-300 MHz OC would be safe to assume...

[AMDforME]

We have no idea if ANY of the claims are accurate as this is a RUMOR at the moment. Some current FX-8350s will run 5.0 GHz. on air but they are far and few between.

AMD could likely bin a quantity of existing Vishera CPUs and set them to run at 5.0 GHz. by default - if they so desired but it's unlikely that these CPUs would have much headroom at all with the current design, which is still very good but has limits. 5% headroom would probably be the limit, IMO unless these are "golden" CPUs beyond anything we've seen in the past.

I personally do not think a 5.0 GHz. Vishera is worth $795. but should AMD even offer a 5.0 GHz. Vishera based CPU for that price, I hope they sell every one they produce.