HyperThreading => even hotter? 3.06GHz+HT => 96W

[DoctorBooze]

Just been re-reading this old article about CPU power consumption. It says the AMD chips produce more heat per MHz because of their higher IPC. This makes me wonder, if HyperThreading makes more efficient use of the P4 core, presumably that means it'll make them hotter too?

Interestingly in the table in that article about the P4, Intel didn't give a maximum thermal power, but just a little arithmetic suggests a 2.53GHz P4 would have a maximum thermal power of 80W - do the math from the max current, just the same as the AMD. Now, I haven't got Intel's figures for the P4 3.06GHz to hand, but I guess we all expected a typical thermal power of 72W or so, so can I suppose it might have a max thermal power of 96W? And if it could actually hit that in operation because of hyperthreading? We'll all really need huge new HSFs with 0.2C/W ratings. Ouch.

 

[overclocksomemore]

well then treat it like an athlon chip by extra cooling. Now you have yourself a Intel Athlon 4 souped up.. I heard from Digital Jesus that by enabling this Hyper thingy on the P4 gave some serious boost in 3dmark performance az well..

Use it well and may the force be with you.

 

[DoctorBooze]

Actually I don't have a P4 at all, my fastest PC is a T'bird 1333 - which is enough for now! I was just wondering what the techies thought, and if anyone who's got one can tell how much hotter they get.

 

[mechBgon]

The 2.8GHz has a theoretical maximum output of 86W, according to www.sandpile.org. Scaling that up to 3.06GHz gives about 94W. Because HT can only use as many transistors as the CPU actually has, I assume this would still hold as the max... HT is just going to make it easier for the CPU to approach maximum transistor usage, so in effect the CPU may cruise hotter (if all other factors were equal), but the max is probably not going to change with HT enabled.

BTW, reviews say Intel is supplying a beefier heatsink/fan unit with the retail-boxed 3.06GHz.

 

[aznskickass]

Originally posted by: mechBgon
The 2.8GHz has a theoretical maximum output of 86W, according to www.sandpile.org. Scaling that up to 3.06GHz gives about 94W. Because HT can only use as many transistors as the CPU actually has, I assume this would still hold as the max... HT is just going to make it easier for the CPU to approach maximum transistor usage, so in effect the CPU may cruise hotter (if all other factors were equal), but the max is probably not going to change with HT enabled.

BTW, reviews say Intel is supplying a beefier heatsink/fan unit with the retail-boxed 3.06GHz.

Please correct me if I'm wrong, but apparently the 'theoretical maximum thermal dissipation' level can never be reached with everyday software, mainly because the whole of the CPU (ALU/FPU etc) is never 100% in use at all times...

I'm sure HT keeps the CPU more busy in some cases, but definitely not 100%. Otherwise performance will go through the roof!

And btw, when I say 100% I don't mean 100% CPU load...

 

[mechBgon]

I think you're right. According to Wingnutz, who would know, their maxer-outer test program is beyond any real-world software's load level, hence Intel's 75%-of-max thermal-design target.

 

[RU482]

Processor Electrical Specifications

 

[zephyrprime]

I think the max power usage numbers woulnd't change with HT because the max numbers probably already include total processor resource usage. HT makes the average processor resource usage percentage higher but doesn't increase the maximum possible usage. At lesat not with the current P4. I guess they could add more processor execution units and such in future processors specifically to give HT more performance headroom even if these additional processor units were pretty useless for a single thread.

 

[DoctorBooze]

zeph - no they wouldn't do that; they'd improve single-thread performance before improving dual-thread performance. My question remains, given Intel's previous "typical" thermal power was way below any notional maximum thermal power, and HT would make more use of such execution units as there are, would HT push real thermal power past the previous "typical" thermal power level?

 

[rIpTOr]

Originally posted by: overclocksomemore
well then treat it like an athlon chip by extra cooling. Now you have yourself a Intel Athlon 4 souped up.. I heard from Digital Jesus that by enabling this Hyper thingy on the P4 gave some serious boost in 3dmark performance az well..

Use it well and may the force be with you.

HT only gave 20 more points in 3DMark2001 @ tomshardware.
I think its 100W when using HyperThreading according to tomshardware.

So far HyperThreading doesn't help performance at all for a gamer. We need multithreaded games out for it to soar in performance. Still some people are misunderstanding HyperThreading for Dual CPU's.

 

[mrman3k]

So I take it that the 3GHz P4HT is the hottest running processor now?

 

[mechBgon]

The 2.0GHz Williamette was right up there with it, at a thermal design power of 75W, but the 3.06GHz squeaks ahead with a TDP of 82W. (note that more heat output doesn't necessarily mean a higher core temperature reading, when there's simply more core)

What's really interesting is that a Xeon MP with 2Mb of L3 cache in addition to the 512Mb of L2 cache still has a TDP of only 57W at 2.0GHz. I wonder if the little chips around the edges are the L3 cache, and are excluded from the power of the CPU proper? Could explain it...

 

[Mingon]

The xeon will have a higher surface area, this will help in the cooling as per palimino>t'bred

 

[RazeOrc]

And if you'd looked around you'd see that they've redesigned the Intel Retail HSF for the 3+ GHz Procs, They're round not square, use a copper slug core with aluminum around it with bent fins, looks pretty nifty and it supposedly performs pretty well.

 

[DoctorBooze]

We don't have 3GHz P4's in the UK yet... or at least not many, I can't find one!

Whatever size the core is, the HSF has to shift an awful lot of heat. Still, glad to hear Intel have improved their stock HSF.

 

[Jhhnn]

Yeh, that new heatsink is the AVC sunflower, read a blurb a few months back about how their cooler was selected by Intel for future processors- Nice score for the AVC guys....