Saturday, April 15, 2006

Individual Processor TDP / TcaseMax

Starting with it’s Revision E cores, AMD has begun programming individual chip’s TDP rating into their processors. In the past AMD used a conservative maximum TDP rating that was often the highest theoretical TDP across several models and all production batches. TDP was often used to compare how power hungry certain brands/lines of processors were against others, but now with this chip specific TDP, there is a way to quantify the sample-to-sample production variance and determine how a particular chip will perform in terms of power consumption. This becomes much more interesting to overclocking and silent PC enthusisasts. A low TDP chip will consume less power leading to less heat that needs to be dissipated (from both the CPU as well as the PSU) which means it can be cooled more easily or quietly. Conversely a higher TDP or a chip's ability to consume more power, is also speculated to indicate a better overclocking chip.

The majority of the information is compiled from AMD Opteron Processor Power and Thermal Data Sheet (pdf) and AMD Athlon 64 Processor Power and Thermal Data Sheet (pdf). Any clarifications, corrections, and new information is welcome.

TDP stands for Thermal Design Power, or also called Thermal Dissipation Power.

In AMD support forums we find this definition:

This is the maximum theoretical amount of power (in Watts) that a processor may consume and therefore dissipate as heat. TDP-values are crucial when it comes to designing cooling solutions as they specify the maximum amount of power (=heat)
that a cooling solution must be able to dissipate.Note that
AMD's TDP-values are absolute maximum, i.e. 'worst case' ratings.During normal operation, an AMD Processor will typically not reach its specified TDP. Other manufacturers may have a different definition for their TDP-values, e.g. they might give typical instead of absolute maximum ratings, i.e. they might specify the amount of power dissipated during normal operation (='typical' conditions).

In AMD’s Processor Power and Thermal Data Sheets (APPTDS) we find:

Thermal Design Power (TDP) is measured under the conditions of Tcase Max, IDD Max, and VDD=VID_VDD, and include all power dissipated on-die from VDD, VDDIO, VLDT, VTT, and VDDA.

Unfortunately the individual chip’s TDP is not indicated on the exterior or packaging of the CPU. It is however programmed into the chip as a TcaseMax value. From APPTDS

Tcase max is the maximum case temperature specification which is a physical value in degrees Celsius. This value is programmed into Rev D and later processors..
Tcase max is programmed during device manufacturing with part-specific values for Rev E and later processors with 'Variable'
indicated by the Case Temperature OPN character, and can be any valid Tcase max value in the range specified for the corresponding OPN

How is TcaseMax related to TDP? In APPTDS we find a series of thermal profile tables that translates Tcase Max to TDP based on the specific thermal profile of the processor.

The thermal profile is used to define the relationship between Tcase max and devicespecific Thermal Design Power for Rev. E and later processors with “Variable” indicated by the Case Temperature OPN character.

From the above definitions, it seems that during the manufacturing process, AMD will load each chip at the rated voltage/speed and measure the maximum case temperature and/or power consumption to obtain TcaseMax/TDP.

So how to read the TcaseMax from a Revision E or later CPU? A nice little program called AMD64 TcaseMax will read this value off your chip and automatically translate the value based on APPTDS tables. This handy utility can be downloaded here.

Will the TDP rating be affected by the actual clock or voltage the CPU is being run at? No, TDP is measured at the rated voltage, max P-state (or rated/stock clock speed), and assume under maximum load so regardless of your clock speed or voltage, the TcaseMax utility should give the same TDP. The actual power consumption of the chip however will vary with changes in voltage and clock.

So does this mean that a dual-core Opteron 180 with a chip rating of 1.35v, 49C TcaseMax, and 35.0W TDP consumes only 35.0W when running at stock speed of 2.4Ghz under CPU load? Less than half the power of a slower dual-core Athlon X2 3800+ with a smaller L2 cache rated at 1.35v, 71C TcaseMax, and 89.0W TDP running at 2.0Ghz?! Even nearly half the power of a single-core Athlon 3500+ with a TcaseMax of 65C, 67.0W TDP running at 2.2Ghz? Well if our interpretation of TDP is correct, then yes that is exactly what it means.

Borrowing numbers from SPCR’s Desktop CPU Power Survey we see various A64 CPU’s with their TDP read by TcaseMax and their actual CPU power consumption. Although the measured CPU power draw does not exactly match up to the rated TDP, it is within -2.3 to 13.2W. The relative ranking in terms of measured power consumption matches the TDP rating except for the 3500+ Venice which has a slightly higher TDP rating by 1.4W than the X2 3800+ but measured in at 4.2W less. Considering that the motherboard used may not have been giving the exact rated voltage, the temperature/cooling differences of the CPU, as well as power measurement error, overall it seems that rated TDP does give a very good idea how the chip will perform in terms of power consumption.

In addition, even with all CPUs idling at the same clock and voltage (1Ghz 1.1v), we see the ranking of measured power consumption also matching that of the TDP. Again the only exception being the 3500+ Venice, which this time is drawing slightly more power than the higher TDP X2 4800+, but again this could easily be due to the factors mentioned above.

How does TDP affect the ability of a particular chip to be overclocked or undervolted? Well that is a question that unfortunately APPTDS does not seem to address. Any input that can help answer this question is welcome.

Studying the APPTDS we find several figures that seem to determine the processor’s TDP. Each model will have a Thermal Profile that gives a specific Thermal Resistance (case to ambient) in C/W as well as a Tcase Max range. In addition there is a Local Ambient Temperature or Tambient. Plotting the various the Processor Thermal Profile tables, it seems that the relationship between Tcase Max and TDP is linear and can be roughly calculated using the following formula:

TDP = ( TcaseMax – Tambient ) / Thermal Resistance

Each series in the graph represents a different specific Thermal Resistance (C/W). We can see AMD slightly rounded down the highest TDP for some profiles. Most processor profiles are based on a Tambient of 42C, except the FX series which has a Tambient of 40C.

As we can see that different models have different Thermal Profiles and in general lower Thermal Resistance will give a higher TDP for the equivalent Tcase Max (and vice-versa). Will this give us any clues which models will have lower or higher TDP? Let’s look at a summary of APPTDS Thermal/Power Specifications tables. I’ve only included the Rev E chips, and this is per the last update of November 2005 for the Opterons, and March 2006 for all other models. In addition to giving the Thermal profile/Thermal resistance, the voltage range (VID_VDD) and TcaseMax range are also given for each CPU model/revision.

** Data points corrected as these seemed to be typos on the APPTDS.

Before we start salivating at the 20.6 TDP (for silent PC enthusisasts) or the 110.0 TDP (for overclockers), please keep in mind this is just a general range given in AMD’s specs and does not mean that there are even any chips that fall on these extremes. However we do see that in general the dual-core chips have a lower Thermal Resistance of 0.20, notably the higher speed X2s and Opterons. This would suggest a higher TDP, but from user’s posting on the TcaseMax utility forum and on SPCR, it seems most dual-core Opterons have an exceptionally low 35.0W TDPs although one Opteron 165 also had a rating that went up to 105.0W. It seems that it would be too soon to draw any generalizations with the limited data on actual chip TDPs.

However it is interesting to note we do find an Opteron Thermal Profile with a very high Thermal Resistance of 0.95C/W and an extremely low 7.4W-30.0W TDP range, but alas none of the listed models have this profile. The lowest TDP range can be found for the s940 dual-core 260/860 series OSAxxxFAA6CB/CC with 0.51C/W Thermal Resistance, 1.15/1.20v rating, and 13.7-55.0W TDP range; as well as the s940 single-core 240/840 series OSAxxxFAA5BL/BM with 0.53C/W Thermal Resistance, 1.35/1.40v rating, and 13.2-54.7W TDP range.

None of the s754 processors seem to have the individual chip TDP ratings despite having also moved to Rev E. However we can look at their maximum TDP rating just for comparison purposes.

The Rev E Athlon 64 3000 (ADA3000AIK4BX) as well as the Sempron 64s both have a rating of 1.4v, 42C Tambient, and 0.45C/W Thermal Resistance. The TcaseMax is 65C giving a 51W TDP for the Athlon, and TcaseMax of 69-70C giving a 59/62W TDP for the Semprons. Again this is the overall maximum TDP and not the TDP of the individual chip which will likely to be lower. The high Thermal Resistance, however does suggest that these chips in general should have lower TDPs.

The Turion (MT only, no ML) are also included in the APPTDS. The Thermal Resistance of these chips are amazingly high at 2.00-2.08 C/W. However instead of TcaseMax we find a very high TdieMax of 95C, and also the Tambinet is a lower 36C. Possibly this is because Turions come without the heat spreader found on other AMD64 processors. The maximum TDP is listed at 24-25W. Using the formula from above we get slightly higher figures of (28.4-29.5W).


Post a Comment

<< Home