Trinity vs Ivy Bridge

[CarstenS]

Those definitions are essentially equivalent...

...ly worthless for comparison.

Which leads me to the conclusion that TDP is now only a measure for designing cooling aparati for typcial workloads. When demanding stuff is run, processors have circuitry to prevent them from "meltdown". Earlier, this was not the case and so TDP had to be defined with a safety margin to keep cooling sufficient under (hopefully) all circumstances.

 

[mczak]

You can pretty easily see why intel chips could exceed their TDP indeed (not sure exactly about AMD's Turbo with Trinity).
First, what the chip thinks it draws is an estimate, the datasheet explicitly mentions the actual value could be larger in case of a "power virus" and thermal monitoring must be enabled to prevent meltdown if this should happen (and your cooling can't handle more than TDP). (I don't know if this is still the case with Haswell as with integrated voltage regulation it theoretically could problably monitor actual power draw.)
Second, by default the chip may exceed its configured TDP (default value is just SKU TDP but this may be adjusted up or down) for 1 second (default value again but recommended for mobile 28 seconds) by a factor of 1.25 (again, default value).
I'd not be surprised if OEMs adjust some or all of these values. Haven't seen any reviews mentioning it though, it should be possible to read these values back from the cpu.

 

[Andrew Lauritzen]

This begets a need for dynamic vsync, I wonder what's the state of support or non support on AMD and Intel (and whether there's a measurable and useful power use difference if you run a game with vsync on, vsync off, dynamic vsync)

It's mostly (entirely?) a software feature, so it's just a question of API and driver support. CP overrides are always a non-ideal solution.

Which leads me to the conclusion that TDP is now only a measure for designing cooling aparati for typcial workloads.

Bingo. TDP is a good rough measure of how much power a processor draws, but it's for designing cooling, not for comparing architectural efficiency.

From a user point of view, what you care about is battery life (and heat/noise to some extent). The reality is that these days you can't really just look at a CPU SKU and determine what the performance is going to be... each individual OEM setup and cooling solution will affect everything from battery life to performance.

That of course makes it hard to do a head-to-head architectural efficiency comparison (unless you can convince an OEM to make exactly the same laptop with an AMD and an Intel part), which is why I told you guys exactly how to get power numbers from Intel platforms at least - presumably there's something similar for AMD. Again, I'll repeat that the information that you're looking for is readily available, so please go look at it instead of all these other flawed metrics.

(And yes, of course these numbers are only accurate to some error margin, but it's a much tighter and more useful margin than any system-level measurement for the purposes of measuring the CPU/GPU portion of power draw.)

 

[CarstenS]

You can pretty easily see why intel chips could exceed their TDP indeed (not sure exactly about AMD's Turbo with Trinity).

Wasn't this also touted as being a feature, utilizing the cooling solutions thermal inertia?

 

[mczak]

Wasn't this also touted as being a feature, utilizing the cooling solutions thermal inertia?

Yes but with the configurable TDP (and also because it is an estimate) you could exceed the SKU TDP limit for an unlimited amount of time in theory. Now of course if an OEM would configure the TDP differently you'd assume they know what they are doing (that is their cooling solution really is designed to handle a different value), and hopefully exceeding it due to being an estimate shouldn't really be much of a problem in practice.
Just saying that when you compare multiple notebooks with cpus seemingly having the same TDP, you could see power consumption discrepancies due to differently configured turbo stuff. But as said I don't know if that's actually the case, I haven't seen anyone verifying these values.

 

[DSC]

Intel 9.18.10.3165 drivers released for Ivy Bridge users.

32bit
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=22840&lang=eng

64bit
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=22842&lang=eng

 

[pMax]

Yes but with the configurable TDP (and also because it is an estimate) you could exceed the SKU TDP limit for an unlimited amount of time in theory.
...
Just saying that when you compare multiple notebooks with cpus seemingly having the same TDP, you could see power consumption discrepancies

...which leads to what I was meaning. You end up comparing a CPU that draws X under heavy workload with another that draws X+Y. And Y is not a minor fraction of X.

Which is very annoying in a notebook, as it kills your battery 2x faster than the other due to cheated TDP, enforce you to take a bigger power supply for your HTPC (-.-) and fakes comparison.

 

[Homeles]

Intel 9.18.10.3165 drivers released for Ivy Bridge users.

32bit
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=22840&lang=eng

64bit
https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=22842&lang=eng

This quick turnaround may be due to Haswell, but Intel's been stepping up their game. 2 month since the last release? Not bad. If only we had more performance improving drivers like the April 15 one...