Sign in with
Sign up | Sign in
Your question
Closed

Chestnuts Roasting on an Open Phenom Fire - 140 Watts?

Last response: in CPUs
Share
December 3, 2007 9:47:19 AM

http://www.vr-zone.com/articles/Less_Than_Rosy_140W_TDP...

Quote:
Phenom 9900, clocked at just 2.6GHz could have a TDP of 140W. It is said that this arose from AMD's 65nm SOI process, which has comparatively poorer power dissipation.


Let's hope this is not true!



http://www.dvhardware.net/article23520.html
http://www.reghardware.co.uk/2007/11/22/amd_phenom_9700...
http://blogs.pcmag.com/miller/2007/11/amds_new_phenom_c...
December 3, 2007 10:58:59 AM

Quote:
the 9700 was nonetheless absent from a Phenom demo event hosted by AMD last week ahead of this Monday's product launch.


AMD hosting more events lol.... shouldnt they be pouring funds into R&D.

Quote:
The upcoming 2.6GHz Phenom 9900 is said to consume 140W. It's due in Q1 2008 as are AMD's already announced series of three-core Phenoms.


Hehe the heat will melt the electrical tape holding together the K8's....

Nothing like winding up the AMD fanboys lol....
December 3, 2007 11:37:06 AM

I'm concerned if this is true because it would seem to limit Phenom until they can resolve the thermal issues.

This is why I really don't understand when a dual-core Phenom did not launch. The dual cores would have much better yields, clock higher, and use less energy. For many people a fast dual is better than a slower quad.
Related resources
December 3, 2007 11:46:29 AM

Will BTX return?
December 3, 2007 11:51:39 AM

TechnologyCoordinator said:
...
This is why I really don't understand when a dual-core Phenom did not launch. The dual cores would have much better yields, clock higher, and use less energy. For many people a fast dual is better than a slower quad.


Yeh looks like AMD are blindly following their road-map for parallelism which is great in theory except they don't have Intel's R&D muscle to deliver the goods!! At this rate all the talk about 45nm processors sorting out AMD's problems (i.e. more cache, better yields per wafer) will be hot air as they will be in Receivership by then!!

Any company releasing and benchmarking vapourware is definately in big trouble. Now we know that AMD is repeating the Prescott core experience but with no hope of digging themselves out of a very big hole (no SOI process up their sleeves as far as I am aware)...

The only hope for a hopelessly undiversived like AMD is for a very big company with Silicon fabs experience to swallow them up or for AMD to license their processor architecture to Intel (like ARM in the UK do).

Bob

P.S. I should say I am an AMD fanboy with a couple of dual-Opteron based rigs (that is DDR systems before their slide into oblivion with DDR2)...

December 3, 2007 12:10:13 PM

Well this is it, spending too too much time/money now trying to match intel (and atm theres no comparison) while it should be spent on R&D (the lads having too many champagne lunches). Ive seen a few posts about buying mobos for phenoms - makes no sense at all.... At the end of the day loyality means nothing in the technology game - all about performance/price (my E6850 suprises me all the time).

December 3, 2007 12:20:43 PM

Again, I'm thinking that AMD is trying to mimic Intel's development vision without realizing that Intel also sucks a huge amount of money on projects that go nowhere in order to get their projects that do incredibly well... AMD doesn't have that sort of capital, or room for error, for that matter.
December 3, 2007 12:31:22 PM

Slobogob said:
Will BTX return?

Oh the irony... the whole idea of BTX was created by Intel to save its overheating NetBurst CPUs... once they started making C2Ds they effectively dropped the technology... and now AMD might benefit from it. Life is a circle.
December 3, 2007 3:53:47 PM

I belive dell uses a lot of BTX, at least on its Optiplex systems...
December 3, 2007 4:07:18 PM

Glad those aren't my chestnuts roasting... :oops: 
December 3, 2007 4:49:03 PM

TechnologyCoordinator said:
I belive dell uses a lot of BTX, at least on its Optiplex systems...


I believe HP and Gateway are also converting or have already converted over to BTX.
December 3, 2007 6:39:40 PM

Quote:
Phenom 9900, clocked at just 2.6GHz could have a TDP of 140W. It is said that this arose from AMD's 65nm SOI process, which has comparatively poorer power dissipation.
Actually, if this is true, it seems to have excellent power dissipation - it's dissipating lots of power in the form of heat. :lol:  (Sorry, with all the crap Prescott...aka Presshott...got back when, I just couldn't resist.)

On a more serious note - I seem to recall from a few years back that one of the issues with SOI is that it becomes less effective at reducing leakage the smaller you go. AMD has some serious problems coming their way unless they (or IBM) really do have some sort of process fix ready for prime time. Bottom line with thermals - it doesn't matter if you can cool the chip or not - once it starts to heat up the user's room you've reached a thermal limit, and that seems to be about 140W to 150W.
December 3, 2007 7:09:53 PM

sonoran said:
Quote:
Phenom 9900, clocked at just 2.6GHz could have a TDP of 140W. It is said that this arose from AMD's 65nm SOI process, which has comparatively poorer power dissipation.
Actually, if this is true, it seems to have excellent power dissipation - it's dissipating lots of power in the form of heat. :lol:  (Sorry, with all the crap Prescott...aka Presshott...got back when, I just couldn't resist.)

On a more serious note - I seem to recall from a few years back that one of the issues with SOI is that it becomes less effective at reducing leakage the smaller you go. AMD has some serious problems coming their way unless they (or IBM) really do have some sort of process fix ready for prime time. Bottom line with thermals - it doesn't matter if you can cool the chip or not - once it starts to heat up the user's room you've reached a thermal limit, and that seems to be about 140W to 150W.




IBM already has their 65nm SOI process running at around 5GHz, so I believe AMDs process is similar enough that those numbers will come down like they always do.
December 3, 2007 7:33:25 PM

BaronMatrix said:
IBM already has their 65nm SOI process running at around 5GHz, so I believe AMDs process is similar enough that those numbers will come down like they always do.


:lol:  :lol:  :lol: 

You have forgotten that unlike IBM's Power series, AMD's chips are geared towards mass market.

Sure, AMD can get a 3~4Ghz golden chip from the wafer, but how often does that happen?
December 3, 2007 8:46:12 PM

Baron is being rather optimistic. Intel released Core 2 with TONS OF OVERHEAD! Massive overclocks. That same cannot be said of K10. K10 was released much closer to its thermal and frequency limits.
December 3, 2007 8:48:23 PM

Now that you mentioned it, why didn't AMD released their duo core cpus first? That would probably be a lot better. Anyways, 140W. This isn't good at all. A Q6600 is only 95W and it can take the 9600 on. Even the B3 stepping was 105W.

Catchy title as always eh...
December 3, 2007 9:12:10 PM

BaronMatrix said:
IBM already has their 65nm SOI process running at around 5GHz, so I believe AMDs process is similar enough that those numbers will come down like they always do.


So please tell me how a 5ghz chip with a long, streamlined "simple" chip (ie cell) can compare with thermals of a 2ghz with lots more transistors / IPC?

Here's a hint -- Ghz has little to do with it... though some.

Now, if you show me that data that, normalized for transistor count AND clockspeed, the IBM process is significantly better than AMDs, I will grant your point.
a c 123 à CPUs
a b À AMD
December 3, 2007 9:32:13 PM

Too many times have someone brought up the "cell" processor which is in no way comparable to a normal CPU. In fact a QX9650 is much superior in terms that it has many more transistors than a "cell" processor and can do more than one thing at a time.

Either way, BTX was a great idea not only meant for the Prescott chips but for future use. Move the harddrives back and put the CPU in front where the cool air comes in. Its a great idea but its adoption wasn't taken at first just like ATX wasn't adopted at first.

Also, and I have said this many times, SOI has reached its limits and AMD has run into the same stumbling block Intel was at when they first went to 65nm. AMD needs a Hafnium based process(although they swear by SOI) in order to go to 45nm without having a mini nuke in any AMD based PC.

Personally that $622 million infusion should go towards fixing Phenom. Hell I think if AMD wants to make any money they need to leave ATI alone to do what they do instead of pushing them to make chipsets. Such a waste of a once great GPU company.
December 3, 2007 9:34:34 PM

This was to be expected, right? AMD doesn't want you to look at TDP anymore, remember? They want you to look at the average thingee. :-D
December 3, 2007 9:40:40 PM

Quote:
So please tell me how a 5ghz chip with a long, streamlined "simple" chip (ie cell) can compare with thermals of a 2ghz with lots more transistors / IPC?
If you're implying he's talking about Cell, I don't believe he is:

http://www-03.ibm.com/press/us/en/presskit/21546.wss

I don't know a whole lot about cpus, but I'm not sure IBM's Power6 chip qualifies as a "simple" chip.
a c 123 à CPUs
a b À AMD
December 3, 2007 9:48:29 PM

The Power6 huh. Not bad but still nowhere as impressive as Intels 80core CPU. It used 63watts and could perform the same job as 130 CPUs faster and better.

As for AMD the only way they will get this technology is if they pay IBM to make some for them.

Also this Power6 chip looks to be like the "Cell" as it is made for specific processes and wont compete over the entire server market. I really doubt IBM will be getting into the CPU market again as last time Intel pushed them out.
December 3, 2007 10:14:29 PM

The POWER6 has approximately 790 million transistors and 341 mm² large fabricated on an 65 nm process. It was released on the 8th June 2007, at speeds of 3.5 Ghz, 4.2 Ghz and 4.7 Ghz[2], but the company has noted prototypes have reached 6 GHz.[3] POWER6 reached first silicon in the middle of 2005[4].

Dr Frank Soltis, an IBM chief scientist, said IBM had solved power leakage problems associated with high frequency by using a combination of 90nm and 65nm parts in the POWER6 design.[5]

The processor is a dual core design and has 128 KiB of L1 cache (64 KiB data + 64 KiB instruction), an eight-way set-associative design with a two-stage pipeline supporting two independent 32-bit reads or one 64-bit write per cycle.[6] Each core will have a 4 MiB "semi shared" L2 cache, where the cache is assigned a specific core, but the other has a fast access to it. The two cores share a 32 MiB large L3 cache which is off die, using an 80 GB/s bus.[7]

Each core has two integer units, two binary floating-point units, and a decimal floating-point unit, and is capable of two way SMT. The binary floating-point unit incorporates “many microarchitectures, logic, circuit, latch and integration techniques to achieve [a] 6-cycle, 13-FO4 pipeline,” according to a company paper.[6] Unlike the servers from IBM's competitors, the POWER6 has hardware support for decimal arithmetic and will include the first decimal floating-point unit integrated in silicon. More than 50 new floating point instructions handle the decimal math and conversions between binary and decimal.[7] This is a feature being added to the processors powering IBM's System z.[8]
a c 123 à CPUs
a b À AMD
December 3, 2007 10:19:07 PM

As I said BM that is impressive but IBM wont just hand it to AMD. AMD will have to pay for the licsensing. IBM is not in the business to give everything away.

But still nothing compared to the 80 core CPU from Intel especially considering that 80 core was just a base design and they can have anything in them from CPUs, PPUs all the way to GPUs. I would love to have one with 25GPUs, 25PPUs, 25APUs, and 25CPUs. That would be a killer.
December 3, 2007 10:58:05 PM

jimmysmitty said:
As I said BM that is impressive but IBM wont just hand it to AMD. AMD will have to pay for the licsensing. IBM is not in the business to give everything away.

But still nothing compared to the 80 core CPU from Intel especially considering that 80 core was just a base design and they can have anything in them from CPUs, PPUs all the way to GPUs. I would love to have one with 25GPUs, 25PPUs, 25APUs, and 25CPUs. That would be a killer.
js, I work at Intel and even I will say that you're now being about as far out optimistic on the Intel side as certain people usually are on the AMD side. I do expect some...shall we say interesting...developments to come from that Larrabee test chip, but nothing quite as wild as you envision - not for a few more process generations anyhow.

BM - granted, the Power 6 is an impressive accomplishment. But given that IBM can produce them running at such high frequencies, that makes it even more curious that AMD has such serious issues with Barcelona. I could understand the GHz maxing out at a low speed due to chip design issues (LOTS of signal timing inside that has to be right to hit high clock speeds - and that's no trivial feat for any design team). But one also has to wonder about the thermals. How much wattage does that Power 6 chip consume? Is it significantly better (on a per transistor basis) than AMD's 65nm chips, and if so - why? I did see the part about IBM combining 65nm and 90nm in the chip - that was interesting - but what does it say about thermals on their 65nm SOI process?

* Not speaking for Intel Corp *
December 3, 2007 11:16:56 PM

BaronMatrix said:
The POWER6 has approximately 790 million transistors and 341 mm² large fabricated on an 65 nm process. It was released on the 8th June 2007, at speeds of 3.5 Ghz, 4.2 Ghz and 4.7 Ghz[2], but the company has noted prototypes have reached 6 GHz.[3] POWER6 reached first silicon in the middle of 2005[4].

Dr Frank Soltis, an IBM chief scientist, said IBM had solved power leakage problems associated with high frequency by using a combination of 90nm and 65nm parts in the POWER6 design.[5]

The processor is a dual core design and has 128 KiB of L1 cache (64 KiB data + 64 KiB instruction), an eight-way set-associative design with a two-stage pipeline supporting two independent 32-bit reads or one 64-bit write per cycle.[6] Each core will have a 4 MiB "semi shared" L2 cache, where the cache is assigned a specific core, but the other has a fast access to it. The two cores share a 32 MiB large L3 cache which is off die, using an 80 GB/s bus.[7]

Each core has two integer units, two binary floating-point units, and a decimal floating-point unit, and is capable of two way SMT. The binary floating-point unit incorporates “many microarchitectures, logic, circuit, latch and integration techniques to achieve [a] 6-cycle, 13-FO4 pipeline,” according to a company paper.[6] Unlike the servers from IBM's competitors, the POWER6 has hardware support for decimal arithmetic and will include the first decimal floating-point unit integrated in silicon. More than 50 new floating point instructions handle the decimal math and conversions between binary and decimal.[7] This is a feature being added to the processors powering IBM's System z.[8]


Fair enough.

What's the power envelope for these things?
December 4, 2007 12:06:38 AM

Just read over anandtech, in the review of the sparkle 8800Gt with passive cooling..

"you're not familiar with the QX9770, it's a quad-core 3.2GHz Penryn/Yorkfield based processor that won't be out until early 2008. It's got a TDP of 135W and manages to get quite warm"

Well, I guess that these TDP are normal, since IMHO135W is pretty close to 140w, given the memory controller is inside the CPU..
December 4, 2007 12:14:49 AM

Except that one's running at 2.6ghz and the other's at 3.2ghz. Go figure.
December 4, 2007 12:24:01 AM

jbj190 said:
Except that one's running at 2.6ghz and the other's at 3.2ghz. Go figure.


And more importantly, benchmarks.
December 4, 2007 12:47:24 AM

That I know, but I would have expected better for that one too..
December 4, 2007 12:51:09 AM

HAHAHA PWND! i knew something bad would happen to amd. god this is depressing. i'm very surprised there are still people wanting to see BTX cases in the game.
December 4, 2007 3:39:21 AM

Bummer. Let's hope this isn't true.

Then on the other hand, 2.8GHz Phenom has dropped off the roadmap, so maybe it is true.
December 4, 2007 4:32:14 AM

easyg said:
Bummer. Let's hope this isn't true.

Then on the other hand, 2.8GHz Phenom has dropped off the roadmap, so maybe it is true.


Perhaps AMD is labeling the 2.6GHz '9900' because that is as far as they think they can go? ;) 
December 4, 2007 4:49:04 AM

epsilon84 said:
Perhaps AMD is labeling the 2.6GHz '9900' because that is as far as they think they can go? ;) 


Untrue.

BM already stated that IBM is up to 5 GHz on their 65nm SOI process.

AMD will not be far behind.

I'm just going to put it in my attic during the winter. Could use another 1 kW heater.
December 4, 2007 5:24:16 AM

cnumartyr said:
Untrue.

BM already stated that IBM is up to 5 GHz on their 65nm SOI process.

AMD will not be far behind.

I'm just going to put it in my attic during the winter. Could use another 1 kW heater.


That's a bit optimistic if you ask me. Takes BM's comments with a world of salt. Blinded fanboy'ism can only get you so far. The PowerPC6 architecture is working fine for IBM. But don't expect it to give much help to AMD. They are in a tremendously bad situation atm. With their processors and IBM's heading in completely different directions. Just because IBM is having some success...doesn't mean it will filter down to AMD in their predicament. Even if they did ask IBM for help. This would not be a simple solution...It will also cost quite a bit of money which AMD isn't pulling in.

Personally...It will take some time for AMD to recover. Penryn isn't really that big of a problem. But if AMD doesn't find the lucky golden egg to solve their architecture issues by the time Nehalem comes around...They are going to be in big trouble. Who knows how big of a dent these Barcelona issues have placed on focused R&D research. It's most likely affected their roadmap and timeline for future releases. If they don't find a way out soon...This is going to be the worst possible situation for AMD. With their lack of R&D funding/staff/facilities...It might spell disaster if they don't get on the right track soon...Who's to blame?...Almost impossible to say. But, Hector Ruiz will definitely fall in at the top of the list.
December 4, 2007 5:34:15 AM

Quote:
Well, I guess that these TDP are normal, since IMHO135W is pretty close to 140w, given the memory controller is inside the CPU..

A lot closer than some think. AMD's chip is still being rated for max TDP, while Intel's is at average TDP.
As sonoran said, alot of the problems could be fixed by a really good redraw. Amd hasn't taken a serious look at component layout since t-bred B. Now they have to consider disimilar charges on the outside of the cores as well.
They probably could do with a little more r&d on thier process as well.
Why are they getting such bad thermals, when thier I d sat is so low?
Perhaps it's just time to up the number of pipeline stages.
December 4, 2007 7:09:07 AM

endyen said:
Quote:
Well, I guess that these TDP are normal, since IMHO135W is pretty close to 140w, given the memory controller is inside the CPU..

A lot closer than some think. AMD's chip is still being rated for max TDP, while Intel's is at average TDP.
As sonoran said, alot of the problems could be fixed by a really good redraw. Amd hasn't taken a serious look at component layout since t-bred B. Now they have to consider disimilar charges on the outside of the cores as well.
They probably could do with a little more r&d on thier process as well.
Why are they getting such bad thermals, when thier I d sat is so low?
Perhaps it's just time to up the number of pipeline stages.


Wrong. Intel has been measuring max TDP ever since Core 2 was released. In fact, Intel is generally quite conservative in their estimates, their '65W' CPUs are generally well under that, especially the lower end models.

December 4, 2007 7:25:39 AM

epsilon84 said:
Wrong. Intel has been measuring max TDP ever since Core 2 was released. In fact, Intel is generally quite conservative in their estimates, their '65W' CPUs are generally well under that, especially the lower end models.

Could you link it plz? Last time I had discussion about this, I proved Intel's TDP wasnt theoretical max, ever. Thats why AMD introduced ACP (average CPU power), to be comparable to Intel's TDP.
December 4, 2007 7:48:44 AM

Harrisson said:
Could you link it plz? Last time I had discussion about this, I proved Intel's TDP wasnt theoretical max, ever. Thats why AMD introduced ACP (average CPU power), to be comparable to Intel's TDP.


http://www.lostcircuits.com/cpu/intel_yorkfield/4.shtml


See how the '65W' rated E6700/E6750 are well below 65W? The X6800 is rated at 75W and pulls 69W. The QX6700 and QX6850 are both 130W TDP chips and pull 115 - 118W. All the Core2 based CPUs are pulling less power than the official TDP rating.

Contrast that with the P4 840EE pulling 147W despite having a 130W TDP. Clearly Intel has changed the TDP measurements since Core2, I don't know if it is documented or not (never looked) but it is clear that the rating is more a 'max' rating than an 'average' rating.
December 4, 2007 7:59:53 AM

Epsilon84, its just your assumption. Just because Intel left bigger headroom for C2D series, doesnt mean Intel's TDP is max, its just ACP analog. Notice BE2300 at 30W while its TDP 45W? C2D is about 15% below TDP in Prime, and BE is below 35%.
December 4, 2007 8:10:43 AM

For those who continues to cry "Intel's TDP is measured under "average" circumstance", here is the standard definition for Intel's TDP.

Quote:
Thermal Design Power: a power dissipation target based on worst-case applications.


Worst case applications, which means maximum TDP draw realistically possible.

ftp://download.intel.com/design/processor/designex/3178...

Note: This is different from AMD's definition of TDP, although I can't find the standard definition for AMD's TDP. It is, however, likely that AMD rated their TDP based on maximum power draw a processor can sustain, which I suspect they turn on every single transistor in the die.

December 4, 2007 8:16:23 AM

TDP (or Thermal Design power), as I understand it, isn't actually a measurement of CPU wattage, but rather a parameter to determine the level of cooling needed for a given family of CPUs. For example, '65W TDP' CPUs must have cooling solutions DESIGNED for operation up to 65W.

Both AMD and Intel currently have TDP groups that cover a wide range of processors, and are generally conservative in that part. For example, Intel rates the entire Core 2 Duo line at 65W, from the 1.6GHz E2140 to the 3.0GHz E6850. Clearly the E6850 would be a lot closer to 65W than an E2140 would.

December 4, 2007 8:19:46 AM

yomamafor1, dont twist words, unless its your prefered way of discussion, ofc ;)  Intel is measuring with some worst case apps (we dont know which, it may very be possible to push more W than that), but as I said, its not theoretical max as in AMD. Thats exactly why AMD introduced ACP, which is Intel's TDP analog.
December 4, 2007 8:24:36 AM

Thanks for the link yomamafor1.

Do you (or anyone else) have an official link to how AMD currently measures TDP?
a b à CPUs
December 4, 2007 8:52:15 AM

When Intel Released an oven they atleast had the name and power to back there product, AMD doesnt have this.

Iv seen Prescotts (CELERON D!) with the Heat Spreader Discolored from the Heat! - there a solid chip in that perspective.
December 4, 2007 8:56:06 AM

pat said:
Just read over anandtech, in the review of the sparkle 8800Gt with passive cooling..

"you're not familiar with the QX9770, it's a quad-core 3.2GHz Penryn/Yorkfield based processor that won't be out until early 2008. It's got a TDP of 135W and manages to get quite warm"

Well, I guess that these TDP are normal, since IMHO135W is pretty close to 140w, given the memory controller is inside the CPU..

OK, first of all... many wrong assumptions going on here. Intels and AMDs quoted TDPs are decidedly not comparable.

endyen said:
AMD's chip is still being rated for max TDP, while Intel's is at average TDP.

WRONG, exactly the opposite of that actually.

endyen said:
Perhaps it's just time to up the number of pipeline stages.

Thats what netburst was :non: 

epsilon84 said:
Wrong. Intel has been measuring max TDP ever since Core 2 was released. In fact, Intel is generally quite conservative in their estimates, their '65W' CPUs are generally well under that, especially the lower end models.

Indeed.

Harrisson said:
Could you link it plz? Last time I had discussion about this, I proved Intel's TDP wasnt theoretical max, ever. Thats why AMD introduced ACP (average CPU power), to be comparable to Intel's TDP.

To be fair, AMD introduced ACP because its maximum TDP is well above that of core2. AMD would like you to not hear about this though.
December 4, 2007 3:03:03 PM

piesquared said:
There does indeed seem to be something peculiar happening with the QX9770. This isn't the only site pondering why a mere 200mhz increase yields such a whopping increase in power consumption.

http://www.lostcircuits.com/cpu/intel_yorkfield9770/4.s...



I guess my suggestion that we won't see an actual retail 4GHz chip was right on the money. And just because this is a bash AMD thread, I guess I can say that Hi-K is Hi-Krap.
December 4, 2007 3:15:21 PM

BaronMatrix said:
I guess my suggestion that we won't see an actual retail 4GHz chip was right on the money. And just because this is a bash AMD thread, I guess I can say that Hi-K is Hi-Krap.


Well, let's not be too hasty. After all, this is Intel's first revision. Even though AMD doesn't seem to be allowed the same courtesy from most websites, forums, viral marketers (you know who you are... ;)  ), I imagine another revision should help supply a resonable 3.2 Ghz part. Hopefuly anyway, as they will soon need it.
December 4, 2007 3:56:17 PM

skittle said:

To be fair, AMD introduced ACP because its maximum TDP is well above that of core2. AMD would like you to not hear about this though.

AMD's TDP rating is higher than Intel's TDP and thats exactly what AMD was speaking for many years, where do you get "AMD would like you to not hear about this though"? ;)  AMD TDP > AMD ACP ~ Intel TDP, pretty much it. Still smiling over some ppl idea of Intel's TDP as max :hello: 
!