AMD - Piledriver chitchat

[CAT-THE-FIFTH]

I just realised pcgameshardware tests power consumption at the wall during gaming too:

http://www.pcgameshardware.de/screen...Power-pcgh.png

With an FX8350 and a Geforce Titan you are still consuming less than 300W at the wall,and around 90W more than say a Core i5 2500K,which is less than I expected.

[watercooled]

Broken link, they don't want direct linking to images it seems.

That also shows you don't need anything like a 1200W PSU to run something like Core i5 + 670!

[CAT-THE-FIFTH]

Broken link, they don't want direct linking to images it seems.

That also shows you don't need anything like a 1200W PSU to run something like Core i5 + 670!

Here it is:

http://www.pcgameshardware.de/Core-i...-Test-1077362/

Also its interesting that a Haswell 35W TDP Core i5 dual core is consuming more power than a 55W TDP Core i3 dual core.

Oh! Noes! It seems Intel is lying about TDP ratings - only AMD does that!!

[scaryjim]

CAT: look at Crysis performance and the i5 is 20% faster than the i3. That means that the GPU will be doing a lot more work rendering scenes, and hence probably using more power. It's perfectly possible that the CPU is actually drawing less power in the gaming test, even though the whole system is drawing more. Look at the other power usage scenarios and the i5 draws the same or slightly less power.

Besides, TDP doesn't have a linear relationship with power draw. It's an indication of the kind of heat generated by a chip that coolers need to deal with. A very efficient chip could actually draw the same power as a less efficient one but have a much lower TDP, because more of the drawn power is being converted into computational work and less is being dissipated as waste heat...

[CAT-THE-FIFTH]

CAT: look at Crysis performance and the i5 is 20% faster than the i3. That means that the GPU will be doing a lot more work rendering scenes, and hence probably using more power. It's perfectly possible that the CPU is actually drawing less power in the gaming test, even though the whole system is drawing more. Look at the other power usage scenarios and the i5 draws the same or slightly less power.

Besides, TDP doesn't have a linear relationship with power draw. It's an indication of the kind of heat generated by a chip that coolers need to deal with. A very efficient chip could actually draw the same power as a less efficient one but have a much lower TDP, because more of the drawn power is being converted into computational work and less is being dissipated as waste heat...

The thing is though people complain about AMD "lying" about TDP figures on other forums,when looking at power draw(when its not) but give Intel a pass when it happens in their case. They both different calculations for generating TDP,so this is why they are not directly comparable.

If you look at the X264 results,if anything the Core i5 4570T in the same class as the Core i3 3220 and G2120 for example. It technically is a 65W TDP chip using the metrics Intel uses for TDP definition.

Moreover,looking at the power consumption measurements,the actual added electricity costs seem to be more determined by usage pattern,so its why I really never understood why people think AMD rigs are very expensive to run.

[kalniel]

The thing is though people complain about AMD "lying" about TDP figures on other forums

Well I'd just avoid those forums then. There are plenty of decent forums out there who aren't so silly

[scaryjim]

True, true. It could still be be that Haswell dissipates less heat than Ivy Bridge for the same input load though. The only way to really tell how accurate a TDP is would be to measure the heat output through the IHS for different chips...

Sounds like a nice little project actually

[CAT-THE-FIFTH]

Well I'd just avoid those forums then. There are plenty of decent forums out there who aren't so silly

True.

True, true. It could still be be that Haswell dissipates less heat than Ivy Bridge for the same input load though. The only way to really tell how accurate a TDP is would be to measure the heat output through the IHS for different chips...

Sounds like a nice little project actually

That would be actually quite interesting.

[kalniel]

True, true. It could still be be that Haswell dissipates less heat than Ivy Bridge for the same input load though. The only way to really tell how accurate a TDP is would be to measure the heat output through the IHS for different chips...

Sounds like a nice little project actually

Or just heat output. You have pin-pad conduction to take account of as well.

[scaryjim]

Hmmm, tricky that - you'd need to some how discount the heat output of the other components on the motherboard, system RAM etc and try to find some way of isolating the area around the chip. I guess it'd be possible if you removed the socket and used fly leads to connect back to motherboard traces: you could seal the CPU and socket into a separate chamber that way....

[watercooled]

Actually electrical power draw and heat are essentially the same thing - a CPU drawing 65W of power will dissipate 65W of thermal energy, you can't destroy energy.

TDP and actual power draw are often different but for a different reason entirely; TDP is a sort of ballpark estimate given to a group of CPUs to give hardware manufacturers some general idea about how much heat they need to get rid of. However, individual CPUs in different scenarios can vary massively, for example sample variance, the type of code running, ambient temperature, etc. And TBH TDP is often way off anyway, for example I have a Celeron G540 with a 65W TDP which doesn't exceed 54W AC power even under OCCT PSU test. OTOH, a lot of CPUs are a lot closer to their TDP under normal loads and can easily exceed them under stress tests like this.

Edit: You can't just rip off a CPU socket and have it still work. Aside from the problems of connecting hundreds of leads back to the board, you'd ruin the finely tuned electrical characteristics of the PCB to the point it just wouldn't work anyway.

[Kumagoro]

I think they joking about ripping off the socket.

[watercooled]

I've also just realised 'finely tuned electrical characteristics' sounds like one of those 'sensitive electronics' phrases you're hear massively over-used.

But it's true here, impedance, capacitance, inductance, etc, all have to be taken into account when laying out high speed bus traces. Those serpentine PCB traces you see are an example of that, ensuring differential pairs are the correct/equal length to prevent things like signal reflections.

And then there's EMI to deal with, a lot of the traces are sandwiched between ground planes to shield them from the environment and each other.

[DanceswithUnix]

A CPU isn't like an engine, it doesn't have some of the input energy going to wheels and some heat. All the energy going in turns up as heat. If you use less gates or a better layout to do a given job, then you will generate less heat and therefore be more efficient at that job, but power in is power out.

Measuring power in just to the CPU isn't easy mind. With modern power management, you might be able to just ask the CPU if you know the right incantations

[CAT-THE-FIFTH]

The FX9590 is reviewed:

http://www.kitguru.net/components/cp...te-990fxa-ud5/

[HalloweenJack]

at the game settings they used you`ll be gpu limited , they need to lower them or drop the res to take the gpu out of the equation.

between a stock i7 SB and an i7 HB. (and this is a first) - both of those are 1/2 the price or less than this AMD `beast` ; if it was £300 then it would make sence - not £700.


317 watts system load - wonder if my gtx 480 uses that on its own lol