AMD back in gear, Centurion FX

[Phynaz]

No need of tweaks , historicaly AMD always managed
30% higher frequency without TDP increasing much
for all their nodes once they were mature

When was the last time that happened?

 

[Charles Kozierok]

No need of tweaks , historicaly AMD always managed
30% higher frequency without TDP increasing much
for all their nodes once they were mature , so they could
well bin exceptional parts that will do 25% higher frequency
with at most 40-50% higher TDP.

As someone pointed out earlier, where are they going to find motherboards capable of running such chips?

 

[frozentundra123456]

That's what I was trying to say in the second sentence you quoted, but that probably wasn't very clear..



How about vs i7-3930K instead?

That is the problem with the 5gh FX. The 3930k is what, 2 or 3 hundred dollars cheaper, and I have a hard time believing the FX even at 5ghz would consistently beat a stock 3930k, which still has a lot of overclocking headroom, not to mention the power and cooling problems of the FX.

If the FX was cheaper than the 3930k by a couple hundred dollars it might make some sense, but at the suggested price, it seems like it would be suited only for those who wanted to set some kind of overclocking record.

 

[Abwx]

When was the last time that happened?

Short memory ?..or wishfull trolling.?.

Phenom2 X4 was launched at barely 2.8Ghz and it peaked at 3.7.

Do the maths , it s easier than estimating AMD s stock value...

 

[Abwx]

As someone pointed out earlier, where are they going to find motherboards capable of running such chips?

FX8XXX AM3+ compatible boards are specified at 140 Amperes max
current , with about 1.4V this allow 196W peak CPU current assuming
the MB s VRMS are adequatly cooled , this is more than enough to reach
the 5GHZ mark.

 

[Exophase]

Short memory ?..or wishfull trolling.?.

Phenom2 X4 was launched at barely 2.8Ghz and it peaked at 3.7.

Do the maths , it s easier than estimating AMD s stock value...

Sure, but that's looking at GF's 45nm from start to finish, it's very clear that the first products were on an immature process and brought up to spec by the end. It's clear 32nm was well below spec at first too, hence why Llanos could barely clock at 3GHz. But Vishera was released around 1.5 years later. Surely much if not most of the process maturation of GF's 32nm has already happened by that point. Unless they've been completely neglecting it, in which case I doubt there'd suddenly be huge improvements now..

 

[cytg111]

Doesnt really make alot of sense given the semi-rumor-facts we know about the productline.. Why put out a chip that does ~20% higher clock if you are expecting SR to roll(pun) in this year with a 20% IPC increase and a modest bumb in clock? Seems like one of these products should cancel the other out.. So i guess that leaves some of these semi-facts more semi than fact!

 

[Abwx]

It took them three years or so to squeeze 45nm ?

Richland manage 8% higher base freq. and Vishera
has no GPU to be worried about , so there s still some headroom
left , in fact it all depend if it s base or turbo frequency , i must admit
that if it s base frequency it would be quite an achievement and a
testimony about GloFo s difficulties in mastering the process.

 

[Ajay]

FX8XXX AM3+ compatible boards are specified at 140 Amperes max
current , with about 1.4V this allow 196W peak CPU current assuming
the MB s VRMS are adequatly cooled , this is more than enough to reach
the 5GHZ mark.

If AMD pulled this off (5 GHz pulling only 200W) - it would be AMAZING. Seriously, I think 300W would be an excellent result, and is doable as well, though I think mobo makers would need special "Centurion Ready" motherboard (which would then sell at a price premium to standard AM3+ boards).

From the overclocking table provided earlier, I found one 8350 overclocked to 5.1 GHz @ 1.38v. That is an exceptional result! Suddenly, this seems more possible, and pretty easy to do using a method similar to what galego suggested.

Granted, at that price a hexacore IB-E would be my choice, but AMD still has enough fans to make it work. Plus, if it can do 5 GHz on air, it would be totally awesome on custom or chilled water loops. Enthusiasts pursuing very high clocks are not deterred by power consumption. Let's be clear here, there is not way that this product is aimed at the usual gamer who buys a 3570 or 8350 and is content running them at stock.

The real benefit for AMD, is if this 'Centurion' would create a halo effect and allow AMD to sell more CPUs (and have more wafer starts, potential allowing them to waste less money on the WSA). At this point, AMD needs to use every arrow in it's quiver to sustain itself until Rory's new direction has a chance to produce gains (or not).

Now, we just wait an see if we get reliable confirmation or not.

 

[Arkaign]

Sure, but that's looking at GF's 45nm from start to finish, it's very clear that the first products were on an immature process and brought up to spec by the end. It's clear 32nm was well below spec at first too, hence why Llanos could barely clock at 3GHz. But Vishera was released around 1.5 years later. Surely much if not most of the process maturation of GF's 32nm has already happened by that point. Unless they've been completely neglecting it, in which case I doubt there'd suddenly be huge improvements now..

Not to mention that early PhII's weren't that hard to hit the mid 3Ghz range with anyway. High 3Ghz range with good cooling and good board. Whereas the higher clocked models released much later never were good for much past 4Ghz even if you were lucky AND pushed it really hard. So the gap isn't nearly as large as it seems, AMD was just pushing out the highest clock models they could fit into the power/heat envelope.

 

[middlehead]

Need a meme for this.

Release 795 dollar 5GHz CPU

Lose to 135 dollar i3

No. Not everything deserves a meme.

 

[Abwx]

If AMD pulled this off (5 GHz pulling only 200W) - it would be AMAZING. Seriously, I think 300W would be an excellent result, and is doable as well, though I think mobo makers would need special "Centurion Ready" motherboard (which would then sell at a price premium to standard AM3+ boards).

From the overclocking table provided earlier, I found one 8350 overclocked to 5.1 GHz @ 1.38v. That is an exceptional result! Suddenly, this seems more possible, and pretty easy to do using a method similar to what galego suggested.

Vishera/server CPUs are low volume hence it takes time to get
exceptionnal samples binned but assuming that it can work at this frequency with all cores running it s likely that such high frequencies are either possible only with the most recent batches or that a revision has been made or even both along with RCM implementation.

 

[BrightCandle]

If its 5GHz on air then it will have to be lower than about 140W. Beyond that and you are really mostly talking water cooling. Past 200W its only really custom water cooling and 300W and above I would start moving to extreme cooling measures like vapour change.

So I doubt they are releasing a 5GHz 200W part, that would seem extremely unlikely. But where they have found the magic 60W saving from considering the architecture and process....well that I don't know.

 

[Ajay]

Vishera/server CPUs are low volume hence it takes time to get
exceptionnal samples binned but assuming that it can work at this frequency with all cores running it s likely that such high frequencies are either possible only with the most recent batches or that a revision has been made or even both along with RCM implementation.

Adding RCM would be far too expensive, design, masks, validation - it would be crazy. A minor revision to fulfill a very large customer order, a small outside chance, design time would likely be short, but they would need some new masks and regression testing.

If this is really happening, then it's probably due to some improvement in manufacturing process. For all we know, AMD could have started binning devices at the beginning of the year and will continue to do so until the Haswell launch - where they would likely want to release Richland and maybe a hopped up FX series CPU to grab some attention away from Intel.

 

[Abwx]

Seems that Haswell will be met from all angles by differently
segmented products launches from AMD , Kabini on the low power front ,
Richland in the value segment and , why not , a Vishera under steroids
for performance DT , all this in the waiting of Kaveri for Q4...

Tough year for AMD s engeenering dpts...

 

[galego]

I think you meant only 77% of the PRICE of the Intel chip.

FAIR WARNING galego. You are about to unleash the fury of the Intel I7-3900 series fans. WATCH OUT!

Typo corrected. Thank you!

 

[galego]

You forgot the part about the 220% increase in power consumption necessary to reach that 90% performance sweet spot.

I would obtain something more close to 190% from a rough extrapolation of an overclocked FX. But I do not know the TDP of the Centurion chip.

I know that the i7-3960x @ 3.3 GHz (TDP: 130 W) consumes about a 36% more than the FX-8350 @ 4 GHz (TDP: 130 W) at full load. I know that the i7-3960x consumes about a 85% more than the FX-8350 at idle.

I know that the i7-3970x @ 3.5 GHz has a TDP of 150 W and consumes much more power from the wall than the i7-3960x.

Finally I know that the 2011 socket mobos for the Intel extreme chips are much more expensive that the AM3+ mobos.

But as said above I do not know the TDP of the Centurion... and cannot join all the pieces

 

[Charles Kozierok]

FX8XXX AM3+ compatible boards are specified at 140 Amperes max
current , with about 1.4V this allow 196W peak CPU current assuming
the MB s VRMS are adequatly cooled , this is more than enough to reach
the 5GHZ mark.

Yeah... good luck with that.

I already have a space heater in my office.

 

[Abwx]

As someone pointed out earlier, where are they going to find motherboards capable of running such chips?

FX8XXX AM3+ compatible boards are specified at 140 Amperes max
current , with about 1.4V this allow 196W peak CPU current assuming
the MB s VRMS are adequatly cooled , this is more than enough to reach
the 5GHZ mark.

Yeah... good luck with that.

I already have a space heater in my office.

One can only conclude that you are trolling.....

 

[galego]

As someone pointed out earlier, where are they going to find motherboards capable of running such chips?

Maybe in the same place where users find mobos for overclocking their FX-8350 @ 5.1 GHz...

 

[Phynaz]

Short memory ?..or wishfull trolling.?.

Phenom2 X4 was launched at barely 2.8Ghz and it peaked at 3.7.

Do the maths , it s easier than estimating AMD s stock value...

Ok, so that was three generations ago. It hasn't happened since, why would it suddenly happen now, when it isn't happening for anybody anymore?

Logic fail on your part. Or you're just trolling.

 

[Idontcare]

FX8XXX AM3+ compatible boards are specified at 140 Amperes max
current , with about 1.4V this allow 196W peak CPU current assuming
the MB s VRMS are adequatly cooled , this is more than enough to reach
the 5GHZ mark.

I seriously doubt a 5GHz FX-8350 would consume less than 200W, I would be impressed if it consumed less than 300W and would not be surprised if it consumes closer to 350W.

Consider that a 5GHz 2600K consumes 225W (fully loaded with LinX) and a 5GHz 3770K consumes 215W...makes the likelihood of a world-renowned power-hog microarchitecture like Piledriver using even less power than SB or IB at 5GHz patently unbelievable.

 

[Abwx]

Ok, so that was three generations ago. It hasn't happened since, why would it suddenly happen now, when it isn't happening for anybody anymore?

Logic fail on your part. Or you're just trolling.

45nm is three generations ago for AMD.??

We were talking of nodes , didnt we?..

Talk about trolling , indeed , you are quite an expert
on this matter , far more than in bear stocks trading,
that s a given.

 

[Abwx]

I seriously doubt a 5GHz FX-8350 would consume less than 200W, I would be impressed if it consumed less than 300W and would not be surprised if it consumes closer to 350W.

Consider that a 5GHz 2600K consumes 225W (fully loaded with LinX) and a 5GHz 3770K consumes 215W...makes the likelihood of a world-renowned power-hog microarchitecture like Piledriver using even less power than SB or IB at 5GHz patently unbelievable.

Your estimation is based on your chip not in tests with
a sizeable statistical sampling.

Where would your chip be positioned in this list ?.

https://docs.google.com/spreadsheet/pub?key=0AvsxWM_f24kDdEQ4eGhrdzV2aU94QVBZV2R6LXdIY3c&output=html

 

[Enigmoid]

I would obtain something more close to 190% from a rough extrapolation of an overclocked FX. But I do not know the TDP of the Centurion chip.

I know that the i7-3960x @ 3.3 GHz (TDP: 130 W) consumes about a 36% more than the FX-8350 @ 4 GHz (TDP: 130 W) at full load. I know that the i7-3960x consumes about a 85% more than the FX-8350 at idle.

I know that the i7-3970x @ 3.5 GHz has a TDP of 150 W and consumes much more power from the wall.

Finally I know that the 2011 socket mobos for the Intel extreme chips are much more expensive that the AM3+ mobos.

But as said above I do not know the TDP of the Centurion... and cannot join all the pieces

That much more power? Not really.