Haswell Refresh Tempuratures

[Homeles]

Gimme a few days, gotta crawl through the last thousand or so threads in my history.

EDIT: Most of this thread.

http://forums.anandtech.com/showthread.php?t=2323720

Most of this thread.

http://forums.anandtech.com/showthread.php?t=2326941

There's not a single publicly-outed Intel engineer in either of those threads. Stop trolling and spreading false information.

Homeles, you really need to try to stop while you're ahead. You know our policies on thread crapping; trying to tell another member what to do isn't going to fly here.
-ViRGE

 

[Insomniator]

The idea that Intel planned all of this is completely ridiculous, and amusing.

Looking forward to DC :thumbsup:

 

[tential]

So we shaved a couple more watts off with DC.

How much more are we seeing with Broadwell? Anyone got any ideas/completely unfounded guesses they'd like to throw out?

 

[stuff_me_good]

Is the lower power consumption only due to lower temps, or actually some improvements on the core itself? :hmm:

 

[OCGuy]

Intel does these refreshes for their Motherboard and OEM system partners as well as themselves, so they can sell new product between ticks/tocks.

This is not the first time a multiplier bump has EOL'd a previous SKU, and won't be the last. I don't see any issue with it, and a TIM swap or stepping change is just a bonus.

 

[witeken]

So we shaved a couple more watts off with DC.

How much more are we seeing with Broadwell? Anyone got any ideas/completely unfounded guesses they'd like to throw out?

Broadwell-Y consumes 1.4x or 30% less power.

 

[BTRY B 529th FA BN]

They should of put better TIM on the K chips. That would mean Intel is listening to it's people who buy their products.. how literally ignorant

 

[Ajay]

From the same website where a 3770 performs like a 4790 in tests?

Lets try another review site before making any judgments.

Great point. Or we could wait for official reviews - gasp D:

We really are getting so desperate for news the past couple of years that any little snippet of information must be dissected and debated Ad Nauseam (myself included at times).

 

[24601]

I like how everyone comes out of the woodwork when I'm disallowed from pursuing the line of argument.

Cowards.

Look at the timestamp of the Mod edit.

Trolling is not allowed, and this also is a general insult.
Markfw900

 

[DominionSeraph]

I like how everyone comes out of the woodwork when I'm disallowed from pursuing the line of argument.

Cowards.

Look at the timestamp of the Mod edit.

 

[tential]

Broadwell-Y consumes 1.4x or 30% less power.

That's Broadwell-Y though (not your fault obviously intel tells us what they tell us. Becoming more interested in power consumption/heat as I leave my PC on all the time and the room slowly heats up. The slower/less heat there is to dissipate the better.

I'm excited for Broadwell on mobile (Wish it had been out this year in time for Fall), but really interested to see how intel will continue to lower power consumption, and what they'll do once they hit that "wall". When you've hit a performance wall and a power efficiency wall, where do you go next?

 

[Wall Street]

They should of put better TIM on the K chips. That would mean Intel is listening to it's people who buy their products.. how literally ignorant

Maybe you haven't heard, the refresh of the unlocked k-series is called "Devil's Canyon" and is rumored to be released on June 2. It will be interesting to see if that chip just matches the temperature improvement we see in this review, or if they went even further with enhancing the unlocked chips. The new CEO of Intel said that enthusiasts are a growth market that they are focused on, which is a huge plus.

 

[Idontcare]

http://www.tweaktown.com/articles/6...h-cpu-and-z97-performance-preview/index5.html

It looks like Haswell Refresh uses 15 watts less power and is 10 degrees C cooler than original Haswell. Let's see if Devil's Canyon is similar or if it is even better. This looks like the new TIM is paying off, let's hope all of the chips run this cool instead of the lottery effect.

Could be solely from improved thermal conductivity (be it improved TIM or be it from better control over the gap height distribution and mean) or it could really be from something as common as redoing a few masks as we saw occur with the 65nm Q6600 B3 stepping versus G0 stepping which yielded vastly improved thermals and power consumption.

Given that Haswell Refresh is guaranteed to have a reworked mask set, I'm betting we are going to see a power consumption reduction (which begets a temperature reduction provided the thermal conductivity remains unchanged at the worst) simply due to the layout engineers having had another year to optimize the routings and xtor blocks over that which the original Haswell layout team was given.

 

[OCGuy]

Could be solely from improved thermal conductivity (be it improved TIM or be it from better control over the gap height distribution and mean) or it could really be from something as common as redoing a few masks as we saw occur with the 65nm Q6600 B3 stepping versus G0 stepping which yielded vastly improved thermals and power consumption.

Given that Haswell Refresh is guaranteed to have a reworked mask set, I'm betting we are going to see a power consumption reduction (which begets a temperature reduction provided the thermal conductivity remains unchanged at the worst) simply due to the layout engineers having had another year to optimize the routings and xtor blocks over that which the original Haswell layout team was given.

Yea I don't see Intel bothering to replace the TIM on non-K parts. They already run well within spec and can't be pushed.

 

[witeken]

That's Broadwell-Y though (not your fault obviously intel tells us what they tell us. Becoming more interested in power consumption/heat as I leave my PC on all the time and the room slowly heats up. The slower/less heat there is to dissipate the better.

I'm excited for Broadwell on mobile (Wish it had been out this year in time for Fall), but really interested to see how intel will continue to lower power consumption, and what they'll do once they hit that "wall". When you've hit a performance wall and a power efficiency wall, where do you go next?

There are 2 options:

1. Make better transistors or other similar innovations.
2. Stop innovating like HDD companies have done; those would only shoot themselves in the foot if they make a 1PB HDD for 100 bucks.

 

[Dufus]

Is the lower power consumption only due to lower temps, or actually some improvements on the core itself? :hmm:

I think it's the other way around, lower power to keep temps under control due to higher power densities with die shrinks. Otherwise why not have a higher TDP and clocks that go with it while giving the customer the option to limit TDP if they want to conserve power.

 

[TuxDave]

I think it's the other way around, lower power to keep temps under control due to higher power densities with die shrinks. Otherwise why not have a higher TDP and clocks that go with it while giving the customer the option to limit TDP if they want to conserve power.

It sort of goes both ways. If you take a chip and compare water cooling vs air cooling, you'll find that, excluding the extra power for the water pump, the lower temperature on the CPU will result in lower power consumption. This is because leakage is affected by temperature. That's why if you look at the top 500 GREEN super computers, which measures performance per watt, besides seeing efficient chips and generally smaller server sizes, you're also starting to see exotic cooling solutions to help reduce power consumption (#1 uses oil immersion cooling)

And then yeah, faster clocks and all the usual stuff can increase temperature.

 

[ericloewe]

Any word on whether the refresh will lift the 8GB/DIMM RAM limitation?

 

[Magic Carpet]

As a 4570S/4770K owner. The figures are wrong, there is just no way 4790 consumes ~6w less at idle (that is only achievable with a different PSU/MB). Don't even wanna comment on the rest of the article. A joke. Need a more credible/professional source.

 

[ShintaiDK]

Any word on whether the refresh will lift the 8GB/DIMM RAM limitation?

Thats a DDR3 limitation rather than CPU. You need server memory to get higher densities or DDR4 for standard desktop.

 

[ericloewe]

Thats a DDR3 limitation rather than CPU. You need server memory to get higher densities or DDR4 for standard desktop.

Nope. 16GB unbuffered DIMMs are possible, the first ones are just appearing now. Intel processors older than the Avoton Atoms have some limitation (bug?) that keeps maximum DIMM size at 8GB.

 

[Magic Carpet]

Nope. 16GB unbuffered DIMMs are possible, the first ones are just appearing now. Intel processors older than the Avoton Atoms have some limitation (bug?) that keeps maximum DIMM size at 8GB.

There are even some AM3+ boards with 64GB memory support, but no modules to max them out. Crucial still doesn't list non-ecc 16GB memory sticks on their web site.

And with DDR4 due out later this year.... :hmm:

 

[cytg111]

Yes, shame on Intel for listening to the enthusiast community and giving them what they want. I too am depressed by this unfortunate development.

Mayhaps this chip wont throttle under avx loads? One can hope.

 

[mikeymikec]

I feel like I'm asking an idiotic question (presumably because I have missed a piece of information), but what unit of measurement are the temperatures in? I'm wondering what parallel universe I've just entered where an FX-8150 has the lowest temps.

I'm also having some trouble swallowing the power consumption readings. It suggests to me that an FX-8150 consumes a similar amount of power (load or idle) to an Intel i7-3770K. Whereas this page suggests that the FX-8150 consumes double the 3770K on load and about 20W more when idle:
http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/6

 

[ShintaiDK]

Nope. 16GB unbuffered DIMMs are possible, the first ones are just appearing now. Intel processors older than the Avoton Atoms have some limitation (bug?) that keeps maximum DIMM size at 8GB.

And who makes these modules? Its the same on Core series for that matter, 8GB per DIMM max for desktop.

And with DDR4 so close, a 16GB DIMM DDR3 product is just not gonna sell.