Intel: Chips To Become Slower But More Energy Efficient

[MarcWolfe]

8 hours ago, Albatross said:

I don't like the idea of my products getting slower/less performance for the sake of energy proficiency but whatever I guess.

Watch this video from Intel, and don't confuse the word power with performance. Skip to 4:40.   https://youtu.be/cMWGeJyrc9w

[vanished]

4 hours ago, MarcWolfe said:

Watch this video from Intel, and don't confuse the word power with performance. Skip to 4:40.   https://youtu.be/cMWGeJyrc9w

6:24 what was that?  "look at our research chip with 48 cores"?

[patrickjp93]

1 hour ago, Ryan_Vickers said:

6:24 what was that?  "look at our research chip with 48 cores"?

Yup, that was an early version of Big Avoton, or a sort of middle ground between the Avoton (Atom-based server chips with 8-16 cores) and their Xeon Phi designs (with extra vector units to be used in compute, going up to 61 cores in the previous generation and 72 in Knight's Landing). There's a 32-core Avoton chip coming based on Airmont which will be going head to head with Qualcomm's and Cavium's ARM solutions as an even lower power alternative to Xeon D.

[vanished]

3 minutes ago, patrickjp93 said:

Yup, that was an early version of Big Avoton, or a sort of middle ground between the Avoton (Atom-based server chips with 8-16 cores) and their Xeon Phi designs (with extra vector units to be used in compute, going up to 61 cores in the previous generation and 72 in Knight's Landing). There's a 32-core Avoton chip coming based on Airmont which will be going head to head with Qualcomm's and Cavium's ARM solutions as an even lower power alternative to Xeon D.

I guess I knew it must be something like that but I just couldn't avoid thinking about the 0.1% possibility of some new X*9 CPU to replace the 5960X

[patrickjp93]

Just now, Ryan_Vickers said:

I guess I knew it must be something like that but I just couldn't avoid thinking about the 0.1% possibility of some new X*9 CPU to replace the 5960X

Well, there is the 6950X coming with 10 cores, but yeah, it's going to be a long time before we've got the software that will be built to take advantage of all of this, much less do it optimally. Sure, some professional software does it already, but consumerville doesn't have the cash to make it worthwhile for the R&D (even though OpenMP makes it so damn easy...)

[MarcWolfe]

On 2/24/2016 at 1:14 AM, Ryan_Vickers said:

I guess I knew it must be something like that but I just couldn't avoid thinking about the 0.1% possibility of some new X*9 CPU to replace the 5960X

Its their research chip. It probably is an x86 over powered behemoth i7/Xeon. It probably has some 300 watt tdp which is why they don't make a consumer version. Theh very well could make a 60 core 4 Ghz i7/Xeon, but just won't put out such a power hungry CPUs.

Edited February 28, 2016 by MarcWolfe

[patrickjp93]

12 minutes ago, MarcWolfe said:

Its their research chip. It probably is an x86 over powered behemoth i7/Xeon. It probably has some 300 watt tdp which is why they don't make a consumer version. Theh very well could make a 60 core 4 Ghz i7/Xeon, but just won't put out such a power hungry CPUs.

IBM can only get 12 cores to go at 4.3-4.5GHz, and that's a Power 8 RISC chip on 22nm FDSOI, sacrificing power efficiency for clocks. They're 250W chips. I doubt Intel can get near this level of nuts performance.

[MarcWolfe]

Just now, patrickjp93 said:

IBM can only get 12 cores to go at 4.3-4.5GHz, and that's a Power 8 RISC chip on 22nm FDSOI, sacrificing power efficiency for clocks. They're 250W chips. I doubt Intel can get near this level of nuts performance.

I'm sure they can. Imagine all the R+D money they must have. They only have to do it once to have a research CPU. May of been some trial and error.

[patrickjp93]

Just now, MarcWolfe said:

I'm sure they can. Imagine all the R+D money they must have. They only have to do it once to have a research CPU. May of been some trial and error.

I think you vastly overestimate Intel's abilities to defy the laws of physics. It would be nightmarishly hot and require chilled liquid cooling.

[MarcWolfe]

2 hours ago, patrickjp93 said:

I think you vastly overestimate Intel's abilities to defy the laws of physics. It would be nightmarishly hot and require chilled liquid cooling.

You say it would defy the laws of physics, then you say what it would need. Make up your mind. Its a research CPU meaning it might not follow most of imtels standards. Its just for research, they jave the money and technology, so why not?

[patrickjp93]

8 hours ago, MarcWolfe said:

You say it would defy the laws of physics, then you say what it would need. Make up your mind. Its a research CPU meaning it might not follow most of imtels standards. Its just for research, they jave the money and technology, so why not?

Because if there's a promise of no ROI at all, their shareholders will start revolting and firing executives. I'm sorry but Intel has nothing like this anywhere in its labs.

[vanished]

16 hours ago, patrickjp93 said:

IBM can only get 12 cores to go at 4.3-4.5GHz, and that's a Power 8 RISC chip on 22nm FDSOI, sacrificing power efficiency for clocks. They're 250W chips. I doubt Intel can get near this level of nuts performance.

I wonder actually... I've thought about this before: what if actually creating a reasonable consumer product was not necessary; what if cost, heat output, and size were all totally unimportant and they were allowed to just create whatever they could.  Could they make a chip that's like 4" x 4" so you've got plenty of room for tons of cores and enough space to spread out the probably 1000+ W TDP with liquid nitrogen or something like that?

[patrickjp93]

1 hour ago, Ryan_Vickers said:

I wonder actually... I've thought about this before: what if actually creating a reasonable consumer product was not necessary; what if cost, heat output, and size were all totally unimportant and they were allowed to just create whatever they could.  Could they make a chip that's like 4" x 4" so you've got plenty of room for tons of cores and enough space to spread out the probably 1000+ W TDP with liquid nitrogen or something like that?

If fab costs were much lower and the yield on such a chip was in the 99% range, sure.

[MarcWolfe]

7 hours ago, Ryan_Vickers said:

I wonder actually... I've thought about this before: what if actually creating a reasonable consumer product was not necessary; what if cost, heat output, and size were all totally unimportant and they were allowed to just create whatever they could.  Could they make a chip that's like 4" x 4" so you've got plenty of room for tons of cores and enough space to spread out the probably 1000+ W TDP with liquid nitrogen or something like that?

Well this is supposed to come out soon. They already make 60 core models. They could make them faster, but they focus on less heat and less power consumption.

[MarcWolfe]

13 hours ago, patrickjp93 said:

Because if there's a promise of no ROI at all, their shareholders will start revolting and firing executives. I'm sorry but Intel has nothing like this anywhere in its labs.

I assume you mean, return on investment. How much you think they would have to imvest for ONE research CPU? Sure as hell not millions or some shit. They get info/data from research that helps devolop future products.... thats the return.... development.

[patrickjp93]

54 minutes ago, MarcWolfe said:

I assume you mean, return on investment. How much you think they would have to imvest for ONE research CPU? Sure as hell not millions or some shit. They get info/data from research that helps devolop future products.... thats the return.... development.

Yes millions. Just to tool one fabrication line with all the proper masks will cost north of 10 million dollars, and then to retool it for something else costs more money. You literally have no idea how expensive this business is.