Phenom, debuting at 2.4GHz

[EternityZX9]

The funny thing is that INQ is often right.

Surely, you must be joking.

Everyone is likely to hit a bullseye if you throw 50 darts at the board.

[BrowncoatGR]

http://www.amd.com/us-en/assets/cont...A_ACP_WPv7.pdf

i've seen this pdf and while it goes into more detail it still doesnt mention the equation red posted. Or maybe im missing it. But i reread it and i didnt see it.

[informal]

i've seen this pdf and while it goes into more detail it still doesnt mention the equation red posted. Or maybe im missing it. But i reread it and i didnt see it.

Here you go :

A suite of industry accepted server workloads was
selected as a basis for determining ACP. These include
fl oating point, integer, java, Web, memory bandwidth,
and transactional workloads. Using a suite of server
workloads that represents the breadth of typical server
applications allows for a better representation of
enterprise server class workloads used by end
customers. These workloads were Transaction
Processing Performance Council (TPC-C), SPECcpu2006,
SPECjbb2005, and STREAM. The geometric mean of
measurements, taken during these workloads, is the ACP.

The results across the suite of workloads are used
to derive the ACP number. The ACP value for each
processor power band is representative of the
geometric mean for the entire suite of benchmark
applications plus a margin based on AMD historical
manufacturing experience.

[BrowncoatGR]

Here you go :

well the doctor told me i need to buy glasses

[red]

Excuse my confusion with geometric mean Yep, results multiplied, then n rooted, n is how many results, right?

http://www.math.utoronto.ca/mathnet/...r/geomean.html
I'm not getting geometric mean, never mind why AMD would use it for ACP. The only reason I can see AMD using it is to obfuscate matters.

Shintai
http://download.intel.com/design/Xeo...s/30235501.pdf
68/96 there is a max power specified as well as a TDP.

From the same PDF, dispelling belief that TDP=average power consumed...

It is the highest expected sustainable power while running known power-intensive real
applications.

And for Shintai again

TDP is not the maximum power that the processor/chipset can dissipate.

[oldblue]

Excuse my confusion with geometric mean Yep, results multiplied, then n rooted, n is how many results, right?

Right.

http://www.math.utoronto.ca/mathnet/...r/geomean.html
I'm not getting geometric mean, never mind why AMD would use it for ACP. The only reason I can see AMD using it is to obfuscate matters.

The geometric mean is probably the best way to average a bunch of benchmarks. For example, let's say you were evaluating a processor with two benchmarks that are considered equally important. The time taken is:

Benchmark A: 10 s
Benchmark B: 1000 s

Now imagine you're comparing it to another processor, and the scores for the second processor are:

Benchmark A: 20 s
Benchmark B: 500 s

If you used the arithmetic mean (average), it would look like processor #2 was better. But if you replaced "Benchmark A" with "Loop benchmark A 100 times", it would look like processor #1 was better.

But this is what the benchmarks really tell us:

Processor #2 is 1/2 as fast on Benchmark A and 2x as fast on Benchmark B.
Processor #1 is 1/2 as fast on Benchmark B and 2x as fast on Benchmark A.

So there is no way to say one is better than the other on average, and their scores should be the same. And sure enough, the geometric mean for each processor is 100s. It doesn't matter if you replace "Benchmark A" with "Loop benchmark A 100 times", the geometric mean for processor #1 would still be the same as the geometric mean for processor #2.

This is why the geometric mean is used for industry-standard benchmarks. For example, the SPEC score is really the geometric mean of many different benchmarks.

I don't know why AMD would use this for ACP. On one hand, you could make an argument similar to benchmarks: if a processor #1 uses 1/2 has much power for task A and 2x as much on task B, it should be equivalent to processor #2. But this doesn't make a lot of sense to me, because benchmarks are of arbitrary length, whereas power consumption is not. AMD is saying that if a processor consumes 1W on task A and 100 W on task B, it's equivalent to a processor that consumes 2W on task A and 50 W on task B. Hmm....

The cynic in me suggests that maybe they use the geometric mean because it's always less than or equal to the arithmetic mean (for positive numbers).

[Shintai]

From a sense that it's an engineering target for both companies, that's right. But the targets are set in different ways. There's no standard for how to define TDP for your processors, and each company uses its own method.



That makes no sense. If they sold Yorkfield with a lower TDP, it would be more valuable in the eyes of consumers (and system builders) and they could charge more for it. Competition or not, Intel is not going to leave money on the table.

Oh im quite sure there is a standard on how to define TDP. Thats the max possible thermal heat from a CPU to be removed by a cooling and system solution. In other words, its max power consumption in ANY worst case.
ACP however is a whole other ballgame. Quite the marketing gimmick when you cant follow the competitor.

Why would they make more money on saying, label the Yorkfield EE at 95W TDP? The people paying 999$+ for a desktop CPU with unlocked multiplier usually dont care much about that. For any other CPU almost...sure. The Yorkfield EE just got a standard 130W TDP. Nomatter how little it use.

[red]

I looked up AMD's definition of TDP...
http://www.amd.com/us-en/assets/cont...docs/33954.pdf

TDP. Thermal Design Power. The thermal design power is the maximum power a processor can draw for a thermally significant period while running commercially useful software.

What I infer from this is that AMD's TDP is NOT the max that the chip can possibly draw.

http://www.amd.com/us-en/assets/cont...ESENTATION.pdf
Why here though they ramble about DE-RATING?

Sounds practically identical to Intel's definition of

It is the highest expected sustainable power while running known power-intensive real
applications.

Yeah, and some time ago, I saw AMD boohooing about Intel TDP, but rather than focusing on the methodology, they focused on the added wattage of the northbridge...

[oldblue]

Oh im quite sure there is a standard on how to define TDP. Thats the max possible thermal heat from a CPU to be removed by a cooling and system solution. In other words, its max power consumption in ANY worst case.

From a cooling perspective, that's correct. But for a given processor, how do you set the TDP? That is the question.

Intel may set the TDP below what the processor actually consumes. Hence, the processor may not be able to run at full speed for some applications without overheating. (Intel processors will slow down to avoid this.)

AMD uses a different way of setting the target TDP, which ensures that their processors should never have to slow down to avoid overheating.

Why would they make more money on saying, label the Yorkfield EE at 95W TDP? The people paying 999$+ for a desktop CPU with unlocked multiplier usually dont care much about that. For any other CPU almost...sure. The Yorkfield EE just got a standard 130W TDP. Nomatter how little it use.

If a 95W TDP makes the CPU more appealing to even one user, why not set it? I repeat: Intel will not leave money on the table. If they could have done something to make these processors more appealing to the market, at no cost to themselves, they would have done it in a heartbeat. And yes, from a marketing perspective, 95 W TDP is more appealing than 130 W.

[informal]

I looked up AMD's definition of TDP...
http://www.amd.com/us-en/assets/cont...docs/33954.pdf


What I infer from this is that AMD's TDP is NOT the max that the chip can possibly draw.

http://www.amd.com/us-en/assets/cont...ESENTATION.pdf
Why here though they ramble about DE-RATING?

Sounds practically identical to Intel's definition of

Yeah, and some time ago, I saw AMD boohooing about Intel TDP, but rather than focusing on the methodology, they focused on the added wattage of the northbridge...

Well at the end,all that matters is system level power consumption .If AMD is better (due to intel forcing FB-dimms),all this rambling about ACP/TDP means jack ***.

[Dimitriman]

if its cheap (sub 250$) I'll buy one.

[Shintai]

From a cooling perspective, that's correct. But for a given processor, how do you set the TDP? That is the question.

Intel may set the TDP below what the processor actually consumes. Hence, the processor may not be able to run at full speed for some applications without overheating. (Intel processors will slow down to avoid this.)

AMD uses a different way of setting the target TDP, which ensures that their processors should never have to slow down to avoid overheating.



If a 95W TDP makes the CPU more appealing to even one user, why not set it? I repeat: Intel will not leave money on the table. If they could have done something to make these processors more appealing to the market, at no cost to themselves, they would have done it in a heartbeat. And yes, from a marketing perspective, 95 W TDP is more appealing than 130 W.

Do you seriously even believe that yourself? The only way a CPU will throttle is with improper cooling solution or faulty sooling solution on stock speeds.

The TDP rating is a standard set that ALL CPUs can be under. Some can be alot lower, some just a small amount under. But all CPUs are under.

Again...Intel is not leavign ANY money on the table for not labeling it 95W. Wakeup and start realize who buys these CPUs. Also they keep family TDP classes. Do you think they also leave money on the table when they mark singlecore and lower end Core 2 Duo as 65W, yet they could get a 35 and 45W TDP? Thats just how it works!

[oldblue]

Do you seriously even believe that yourself? The only way a CPU will throttle is with improper cooling solution or faulty sooling solution on stock speeds.

That's not what Intel says: "A thermal solution designed to meet the thermal profile specifications should rarely experience activation of the TCC as indicated by the PROCHOT# signal going active."

Rarely is not the same as never.

Again...Intel is not leavign ANY money on the table for not labeling it 95W. Wakeup and start realize who buys these CPUs. Also they keep family TDP classes. Do you think they also leave money on the table when they mark singlecore and lower end Core 2 Duo as 65W, yet they could get a 35 and 45W TDP? Thats just how it works!

I understand that they keep family TDP classes. The reason these CPUs are not in the 95 W family is because there are some (not necessarily all) that don't qualify for the 95 W family. If they could call these 95 W CPUs, they would be more appealing (even to overclockers) and would make more money. It would also be a marketing coup.

Does Intel have a desktop TDP family lower than 65 W? They label CPU's in the 35 W TDP class as mobile CPUs. They also charge more for these, as would anybody. (Compare an E6600 to a T7700)

[Hornet331]

overclockers care jack about tdp, they run it out of specs anyway, so tdp are meaningless for them.

only desktop cpu from intel i know thats ts below 65W is the singel core celeron (35W with conroe core).

[Shintai]

That's not what Intel says: "A thermal solution designed to meet the thermal profile specifications should rarely experience activation of the TCC as indicated by the PROCHOT# signal going active."

Rarely is not the same as never.



I understand that they keep family TDP classes. The reason these CPUs are not in the 95 W family is because there are some (not necessarily all) that don't qualify for the 95 W family. If they could call these 95 W CPUs, they would be more appealing (even to overclockers) and would make more money. It would also be a marketing coup.

Does Intel have a desktop TDP family lower than 65 W? They label CPU's in the 35 W TDP class as mobile CPUs. They also charge more for these, as would anybody. (Compare an E6600 to a T7700)

Desktop and Mobile also got one other huge difference. Their packaging and housing. Unlike a desktop CPU, a mobile CPU can stand 100C. You simply cant compare directly.

I´m quite sure you could fit alot of 1.8Ghz Core 2 Duos under the 35W TDP.
Just like alot of AMD CPUs could be lower aswell. But thats the purpose of TDP families. Else the OEM would get 117 specs for CPUs instead of less than 10.

I know what you mean with rarely vs never. But you are on a dead end. Think ambient temperature...specially since we talk about TCC. It simply dont have anything to do by its TDP itself.

[Richie P]

http://my.ocworkbench.com/bbs/showth...697#post422697

vCore at 0.976v for 2.3GHz?? Sounds good...

No-one notice the

(c) 2006

On that IHS? Maybe just re-using old IHS's? Slightly dubious, no?

[oldblue]

Desktop and Mobile also got one other huge difference. Their packaging and housing. Unlike a desktop CPU, a mobile CPU can stand 100C. You simply cant compare directly.

I´m quite sure you could fit alot of 1.8Ghz Core 2 Duos under the 35W TDP.
Just like alot of AMD CPUs could be lower aswell.

AMD does have lower desktop TDP families. And they charge more money for those processors than the ones in the 65 W family. Why wouldn't they?

I know what you mean with rarely vs never. Think ambient temperature...specially since we talk about TCC.

The rarely vs. never was in response to your comment that "The only way a CPU will throttle is with improper cooling solution or faulty sooling solution on stock speeds." Intel's own cooling specifications show that this is wrong. And it's not because of "ambient temperature":

"The thermal control circuit is intended to protect against short term thermal excursions that exceed the capability of a well designed processor thermal solution."

These "short term thermal excursions" occur when a processor exceeds its TDP due to an unusually power-hungry workload, which a "well designed processor thermal solution" can not necessarily handle, causing the CPU to overheat and throttle.

[Nedjo]

No-one notice the

(c) 2006

On that IHS? Maybe just re-using old IHS's? Slightly dubious, no?

It's allways like that. Check out A64 IHS's...

[Shintai]

AMD does have lower desktop TDP families. And they charge more money for those processors than the ones in the 65 W family. Why wouldn't they?



The rarely vs. never was in response to your comment that "The only way a CPU will throttle is with improper cooling solution or faulty sooling solution on stock speeds." Intel's own cooling specifications show that this is wrong. And it's not because of "ambient temperature":

"The thermal control circuit is intended to protect against short term thermal excursions that exceed the capability of a well designed processor thermal solution."

These "short term thermal excursions" occur when a processor exceeds its TDP due to an unusually power-hungry workload, which a "well designed processor thermal solution" can not necessarily handle, causing the CPU to overheat and throttle.

Those are special family CPUs that also got radically other specs than the normal line. Just like LV Xeons and mobiles. Try and compare apples to apples.
Just because CPUs can be in a lower TDP, then it doesnt mean they have to to avoid losing money. So slamming a 95W TDP rating on a Yorkfield EE got 0 impact on sales.

If there was money to be made on lower change TDP values, without being special purpose CPUS. Then you would see CPUs like Core 2 Duo with 65W TDP from 1.8Ghz to 2.67Ghz being divided aggresively from 40W to 65W.

Also from your own document:

Thermal Monitor should not be relied upon to compensate for a thermal solution that does not meet the thermal profile up to the thermal design power (TDP).

So TCCs functionality is also to help under fanspeed changes. But never to compensate for power consumption beyond the TDP.

[oldblue]

So slamming a 95W TDP rating on a Yorkfield EE got 0 impact on sales.

I think at this point it's pretty clear that you have your opinion and I have mine. I think an Extreme processor with a 95 W TDP has more market appeal than an identical one with a 130 W TDP, and you don't. I don't think we're going to convince each other.

Thermal Monitor should not be relied upon to compensate for a thermal solution that does not meet the thermal profile up to the thermal design power (TDP).

So TCCs functionality is also to help under fanspeed changes. But never to compensate for power consumption beyond the TDP.

You misread that. It says "Thermal Monitor should not be relied upon to compensate for a thermal solution that does not meet the thermal profile up to the thermal design power (TDP)."

So for an example:

If the TDP is 95 W, and your thermal solution can only handle up to 90 W, don't count on the thermal monitor to save you when the processor hits 91 W, because the chip could stay there for a while.

But if the TDP is 95 W, and your thermal solution can handle up to 95 W, then the thermal monitor is designed to help you when the chip temporarily exceeds 95 W. This should happen rarely.

[User9498]

TDP is meant as an average rule-of-the-thumb or suggestion for OEM builders, and not law. It's mostly derived by marketers and doesn't approach real-world usage.

http://www.silentpcreview.com/article169-page3.html

I think AMD would be smart in having their marketing department create a thermal/power rating system for not just the cpu, but their individual platforms. Might as well tout the current advantages they have over Intel, while they still exist....

[KTE]

I would never rely on a marketing companies end word, making their product look better at any non-monetary cost is their religion - that goes for both AMD and Intel. Most products under an SKU of a certain TDP are those which just about meet that particular thermal envelope at "maximum tested loads" and a tiny bit more oc or ov will make it break the barrier, sometimes quite extravagantly (C2Q after 3GHz). However, "some" products turn out to be manufacturing bliss that manage to stay at very low TDP without much extra engineering tweaks - we can see this with many C2D CPUs. I don't think they binned especially for many of the lower TDP models, but the design was so good that it ended up with naturally lower TDP with lower speeds. Tell me someone who tests real TDP for all released CPUs though, with a consistent methodology?

It's the same as PSU ratings and HDD ratings for me until properly tested by a neutral third party. Look at the Seagate Barracuda 7200.9 500GB for example; MFG only lists a maximum draw of 13W during Read/Write operations but at spinup it was tested to draw 31W and idles at 9W. Not very straight-up are they but they work on the principle "if we didn't mention it, you can't say we lied".

Server market differs in that they test in-house for maximum power consumption with their specific loads on a given CPU when bulk buying.

[red]

I just thought of something on why AMD might be using geometric mean. One would expect that power consumption stays similar for CPU intensive loads. Then I thought, hey, if it's not threaded, only some cores use power.. Like the single threaded SPEC CPU tests.. tsk tsk.