Intel's 14nm chips.

[patrickjp93]

The heat I was talking about was Broadwell-K. Relax. You didn't need to void your warranty and risk damaging a chip with a solder.

Anyway I'm hoping for an even bigger drop in TDP for the Broadwell-U class for ultrabooks, completely silent laptops utilising dual core +HT and awesome iGPU performance would be great to see.

The newest Chrome Books are already fanless. Also, the lowest SKU has 24 iGPU cores with the improved FPU in the CPU cores as well (multiply/divide instructions used to take 5 cycles, now they take 3).

The top Broadwell SKU is supposed to have 96 GPU cores, or 2.4x the core count of Iris 5200 at the same clock speed. You're going to be highly satisfied. Intel managed to cut the heat output in half from Haswell.

[patrickjp93]

So when intel used soldered IHS in x99 chips they were wrong???

Yup. Naked mount or liquid pro/ultra under the good or go home.

[shiftberry]

i gotta admit that Intel is progressing very well in term of graphics chips! 

 

They perform very well on optimized games like Codemasters'!

[patrickjp93]

i gotta admit that Intel is progressing very well in term of graphics chips!

They perform very well on optimized games like Codemasters'!

They have a deficit on geometry budget they need to fix on the gaming end, as well as provide some easier/more thorough API, and make better drivers, but Intel is due to start off on the right foot with Skylake given the amount of graphics code they added to Linux for using the heterogeneous Skylake iGPU.

[patrickjp93]

Edit: double post error.

[LukaP]

Yup. Naked mount or liquid pro/ultra under the good or go home.

or you just use solder, which is basically 96%metal, so highly heat conductive, and call it a day, maybe gaining 1C from using the CLP

[shiftberry]

They have a deficit on geometry budget they need to fix on the gaming end, as well as provide some easier/more thorough API, and make better drivers, but Intel is due to start off on the right foot with Skylake given the amount of graphics code they added to Linux for using the heterogeneous Skylake iGPU.

I hope so and Skylake would feature more iGPU performances than the previous chips! 

[Builder]

If Broadwell is going to continue the low-power trend of Haswell, then they got quite a competitor in that chip.

Not with smartphones. Intel chips are still coming down from a much higher TDP.

 

ARM chip wattages and TDP are both coming up while Intel's have been starkly falling since Sandy Bridge.

They may be going up, (source?) but Intel has a much longer way to fall to get down to their level in terms of performance/watt.

 

You're a big CISC believer though, aren't you?

[juretrn]

Not with smartphones. Intel chips are still coming down from a much higher TDP.

 

They may be going up, (source?) but Intel has a much longer way to fall to get down to their level in terms of performance/watt.

 

You're a big CISC believer though, aren't you?

1. Maybe... but they have the new Atoms for that anyway.

 

2. According to some tests, RISC and CISC hardly matters anymore when it comes to perf/W unless we're talking some super low-power chips.

http://www.extremetech.com/extreme/188396-the-final-isa-showdown-is-arm-x86-or-mips-intrinsically-more-power-efficient

[LukaP]

1. Maybe... but they have the new Atoms for that anyway.

 

 

Not with smartphones. Intel chips are still coming down from a much higher TDP.

well a 4.5W tdp is very close to some arm designs already... (k1 springs to mind) so i wouldnt be surprised if by skylake they manage to get it into that 1-2 watt range

 

 

edit: snapdragon 800 was apparently a 2-4W part...

[patrickjp93]

or you just use solder, which is basically 96%metal, so highly heat conductive, and call it a day, maybe gaining 1C from using the CLP

I gained 3*C over the solder in my 4960x under full load. And the metal in solder isn't great at conducting as metals go. If it contained some copper or gold we'd be speaking of a very different story.

[LukaP]

I gained 3*C over the solder in my 4960x under full load. And the metal in solder isn't great at conducting as metals go. If it contained some copper or gold we'd be speaking of a very different story.

wow. 3 fucking degrees... a different world huh?

[patrickjp93]

Not with smartphones. Intel chips are still coming down from a much higher TDP.

 

They may be going up, (source?) but Intel has a much longer way to fall to get down to their level in terms of performance/watt.

 

You're a big CISC believer though, aren't you?

Snapdragon is a 2-4W TDP, Broadwell-Y is a 4.8W TDP. Braswell is supposedly a 2W part. ARM is going to feel the pinch of not having good branch prediction come March. ARM used to be <1W processors across the board except for on servers. They've been ticking up as companies realized they needed Out of Order Processing, branch prediction, and some of the amenities which make Intel's designs higher wattage in the first place, and they just implement them to a lesser extent to avoid gaining all the heat. Compare the current snapdragon to the ARM 6 designs. 100% TDP difference (1-2W vs. 2-4W).

 

And ARM will continues to tick up TDP as their logic is forced to rise to higher complexity to fight against x86 which can now perform in their TDP range.

 

As per RISC vs. CISC, since ARM is moving steadily toward becoming CISC, yes. Starting from the ground up on a new ISA to see what we could get when we scrapped legacy instructions for intelligent new designs has been great for everyone to see. Frankly I think it's time get got rid of dedicated 8-bit and 16-bit registers and just allowed people to access all 4 or 8 8-bit sections of a 32/64-bit register. That would be much more efficient and flexible design.

[patrickjp93]

wow. 3 fucking degrees... a different world huh?

Under full load. That gained me 200MHz in my overclock.

[Builder]

1. Maybe... but they have the new Atoms for that anyway.

 

2. According to some tests, RISC and CISC hardly matters anymore when it comes to perf/W unless we're talking some super low-power chips.

I wasn't trying to suggest that RISC vs CISC does matter, however I believe there are other elements of RISC design that make it superior to CISC outside of power consumption, but I was pretty sure he was a CISC guy.

 

 well a 4.5W tdp is very close to some arm designs already... (k1 springs to mind) so i wouldnt be surprised if by skylake they manage to get it into that 1-2 watt range

 

 

edit: snapdragon 800 was apparently a 2-4W part...

Getting down to a 1-2w TDP is much harder than getting to a 4.5w TDP. Also TDP is an inaccurate measure of power consumption because it refers mainly to heat and only serves to demonstrate the minimum power consumption of a given part. 

[LukaP]

Under full load. That gained me 200MHz in my overclock.

and that is rare. also example here (math alert)

 

3GHz part. overclock to 4.5 GHz. that is a 50% increase. delid, CLU/P, oc to 4.7. that is another 4%. really worth it? for those 4 percent?

 

Getting down to a 1-2w TDP is much harder than getting to a 4.5w TDP. Also TDP is an inaccurate measure of power consumption because it refers mainly to heat and only serves to demonstrate the minimum power consumption of a given part. 

so waht if it is, when there are no HP parts in that TDP? when they are all close to 4W.

 

and TDP is pretty much a very accurate way to compare power consumption, on similar processes

[Builder]

Snapdragon is a 2-4W TDP, Broadwell-Y is a 4.8W TDP. Braswell is supposedly a 2W part. ARM is going to feel the pinch of not having good branch prediction come March. ARM used to be <1W processors across the board except for on servers. They've been ticking up as companies realized they needed Out of Order Processing, branch prediction, and some of the amenities which make Intel's designs higher wattage in the first place, and they just implement them to a lesser extent to avoid gaining all the heat. Compare the current snapdragon to the ARM 6 designs. 100% TDP difference (1-2W vs. 2-4W).

 

And ARM will continues to tick up TDP as their logic is forced to rise to higher complexity to fight against x86 which can now perform in their TDP range.

But "come March" is largely the problem here. It's taking longer and longer for Intel to spit out chips. If they're having trouble with 14nm, just imagine what the push to Cannonlake will be like. Their entire roadmap is being pushed back by Broadwell. Skylake was supposed to be a 2015 release, then Cannonlake in 2016.

 

Intel is having a lot of trouble reducing the power and heat on their latest chips, and it's not entirely clear to me that they will be able to come down to ARM's performance/watt in time. 

 

For companies that are vertically integrated like Apple Broadwell is an enormous problem. Broadwell's delay is the reason why Apple is vertically integrated. They don't want to be dependent on other companies for components of their devices. They might switch to ARM just so they can consistently make their own stuff, they did a BIL (Rosetta) with the x86 switch so I wouldn't be surprised to see them do it again.

[LukaP]

But "come March" is largely the problem here. It's taking longer and longer for Intel to spit out chips. If they're having trouble with 14nm, just imagine what the push to Cannonlake will be like. Their entire roadmap is being pushed back by Broadwell. Skylake was supposed to be a 2015 release, then Cannonlake in 2016.

it will be hard, but they managed alright till now

 

Intel is having a lot of trouble reducing the power and heat on their latest chips, and it's not entirely clear to me that they will be able to come down to ARM's performance/watt in time.

they already are. their chips are faster from arm designs, at slightly higher TDPs. they just need to improve the arch now to reduce power consumption a bit more. skylake wil do that

For companies that are vertically integrated like Apple Broadwell is an enormous problem. Broadwell's delay is the reason why Apple is vertically integrated. They don't want to be dependent on other companies for components of their devices. They might switch to ARM just so they can consistently make their own stuff, they did a BIL (Rosetta) with the x86 switch so I wouldn't be surprised to see them do it again.

what does this have to do with intel catching up to arm again?

[patrickjp93]

and that is rare. also example here (math alert)

 

3GHz part. overclock to 4.5 GHz. that is a 50% increase. delid, CLU/P, oc to 4.7. that is another 4%. really worth it? for those 4 percent?

 

so waht if it is, when there are no HP parts in that TDP? when they are all close to 4W.

 

and TDP is pretty much a very accurate way to compare power consumption, on similar processes

When extending the lifetime of the chip, yes, especially as the new instructions that came on Sandybridge enter wide usage in consumer software (mainly AVX stuff).

[LukaP]

When extending the lifetime of the chip, yes, especially as the new instructions that came on Sandybridge enter wide usage in consumer software (mainly AVX stuff).

its still only 4 percent. even when the new instructions start getting used, and you get some boost, the comparison between 4.5 and 4.7 will be the same. 4 percent. and to me, that is not worth delliding

[patrickjp93]

But "come March" is largely the problem here. It's taking longer and longer for Intel to spit out chips. If they're having trouble with 14nm, just imagine what the push to Cannonlake will be like. Their entire roadmap is being pushed back by Broadwell. Skylake was supposed to be a 2015 release, then Cannonlake in 2016.

 

Intel is having a lot of trouble reducing the power and heat on their latest chips, and it's not entirely clear to me that they will be able to come down to ARM's performance/watt in time. 

 

For companies that are vertically integrated like Apple Broadwell is an enormous problem. Broadwell's delay is the reason why Apple is vertically integrated. They don't want to be dependent on other companies for components of their devices. They might switch to ARM just so they can consistently make their own stuff, they did a BIL (Rosetta) with the x86 switch so I wouldn't be surprised to see them do it again.

They're not having problems with 14nm. It's the highest yield node yet. They had a small bump in the road at the first tape out, whereas TSMC had 3 failed tape outs at 20nm but supposedly got 16nm working in good yields on the first try. Also, Skylake still releases Q2 2015. Intel has confirmed the Broadwell delays have not affected Skylake whatsoever, as recently as July.

[patrickjp93]

its still only 4 percent. even when the new instructions start getting used, and you get some boost, the comparison between 4.5 and 4.7 will be the same. 4 percent. and to me, that is not worth delliding

Casual >.>

[LukaP]

They're not having problems with 14nm. It's the highest yield node yet. They had a small bump in the road at the first tape out, whereas TSMC had 3 failed tape outs at 20nm but supposedly got 16nm working in good yields on the first try. Also, Skylake still releases Q2 2015. Intel has confirmed the Broadwell delays have not affected Skylake whatsoever, as recently as July.

thank you for some sense

[LukaP]

Casual >.>

naah, just dont like delidding if not necessary and if the chip clocks well already, im not even considering it. and if i consider it, ill never reuse the IHS. baremountmasterrace

[Builder]

They're not having problems with 14nm. It's the highest yield node yet. They had a small bump in the road at the first tape out, whereas TSMC had 3 failed tape outs at 20nm but supposedly got 16nm working in good yields on the first try. Also, Skylake still releases Q2 2015. Intel has confirmed the Broadwell delays have not affected Skylake whatsoever, as recently as July.

So we're going to have Broadwell and then a new socket within six months of each other? That's embarrassing.

 

it will be hard, but they managed alright till now

 

they already are. their chips are faster from arm designs, at slightly higher TDPs. they just need to improve the arch now to reduce power consumption a bit more. skylake wil do that

what does this have to do with intel catching up to arm again?

They managed alright until now.

 

They're chips aren't that much faster than ARM's designs. http://www.anandtech.com/show/6536/arm-vs-x86-the-real-showdown/14 They're all extremely competitive with each other.