Intel Says New 'Sierra Forest' Chip To More Than Double Power Efficiency

Intel says its upcoming "Sierra Forest" data center chip due out next year will have 240% better performance per watt than its current generation chip. It's the first time the company has disclosed such figures, notes Reuters. From the report: The company is for the first time splitting its data center chips into two categories: A "Granite Rapids" chip that will focus on performance but consume more power, and the more efficient "Sierra Forest" chip. Ronak Singhal, a senior fellow at Intel, said the company's customers can consolidate older software onto a smaller number of computers inside a data center.

"I may have things that are four or five, six years old. I can get power savings by moving something that's currently on five, 10 or 15 different servers into a single" new chip, Singhal said. "That density drives their total cost of ownership. The higher the density, the fewer systems they need."

intel says a lot of things

[smash]

10nm shipping in volume in 2016, etc wake me up when it ships and actually posts performance numbers.

Re:

[willy_me]

I'm left wondering why they haven't been better at performance/W before now. If they've suddenly found 240% savings, what the hell have they been doing wrong all these years?

Those power efficiency gains are easy to achieve if you a) use a more advanced processing node, b) optimize the core design for efficiency, and c) reduce the clock frequencies. Now the question is, will "Sierra Forest" be as fast as the previous generation?

Looks like "Granite Rapids" will be the performance chip and "Sierra Forest" will be for efficiency. It is great to have this as an option but I wouldn't expect "Sierra Forest" to break any benchmarks -- at least those benchmarks that ignore efficien

With a nod to Spiderman

[echo123]

Singhal said. "That density drives their total cost of ownership. The higher the density, the fewer systems they need."

With less power, comes less reason to become chill.

Re:

[thegarbz]

You have a reading disability. None of what you said in any way invalidates the claim that this is the first time Intel has marketed a future product in relation to performance per watt over a current product.

uAmWayDumberThanUs

Re:

[iAmWaySmarterThanYou]

Try again. It's right there in the summary as a direct quote.

You're the dumbest person on this site.

Re:

[thegarbz]

Try again. It's right there in the summary as a direct quote.

You're the dumbest person on this site.

Whatever you say Dunning Kruger. (And since I am talking to you Dunning Kruger is an insult given to people who are so confidentially wrong that they are an embarrassment to the human race. I normally don't point this out to people since generally people know what the Dunning Kruger effect is, but ... this is you we're talking about).

Here's a good site for you to help yourself: https://learnenglish.britishco... [britishcouncil.org]

Re:

[iAmWaySmarterThanYou]

Yawn. Did you babble something again? It;s your MO. So anyway, as usual you stepped up, were wrong then went way off topic to pure ad hominem attack based on nothing and even pulled the DK card. You're such a loser. Big L.

TLDR: you're wrong again, like always, cried like the bitch you are, and went home to celebrate your loss. Again.

I should just write a script to reply to you when I'm offline. It's always the same with you. Maybe I'll just write a generic reply and copy paste it. You really are th

less

[Ol Olsoc]

Less targets for the bad guys

why not?

[v1]

I don't see why you wouldn't make this a major part of your marketing? So much of the consumer market is now in laptops, where battery life is a huge talking point and more efficient processors make them lighter, thinner, longer lasting, and better performers.

Re:why not?

[alvinrod]

Sierra Forest isn't a consumer chip, it's for the server market. Power efficiency is an even more important part of the product there because over the lifespan of the server it's going to be operating and drawing power to some degree almost continuously and the power costs start to become a larger part of the overall cost for a datacenter. That power also generates waste heat that needs to be removed, so cooling needs to be considered as well.

Re:

[Junta]

This is about the datacenter market, and, specifically, bringing over a concept already in the desktops.

Effectively, you can either get a single thread behemoth with AVX-512 and all the trimmings (basically the "high performance" chips is made of what might be considered equivalent to a "P core" in Intel's hybrid architecture). You will have a lot of cores, but the cores will be fairly beefy.

In the 'efficient' offering, it's basically "Ooops, all E cores". Lower clock and less function, but more cores (ho

Re:

[guruevi]

But as we've seen with ARM in servers, many workloads is exactly this, a few microseconds on the chip per click/thread/page hit doesn't need a ton of power, you just need to process/store/retrieve a small amount of data but across many thousands of clients per second.

The actual 'scientific' workloads where current Intel chips shine above all others is a very small part of the modern server market. Sure, back in the day when your LAMP stack was on a single server, this was relevant, now with microservices ev

Re:

[ChunderDownunder]

These chips use E-cores which has been sold with various brandings in low end 'Atom' consumer products for the past decade and a half.

The only announcement here is they've chained 144 of them together.

New Chips, New Security Flaws

[NoWayNoShapeNoForm]

The Intel Way

At what cost?

[WaffleMonster]

What they don't say is they achieve this by producing a CPU with only E cores each having about a third of the single thread performance of a normal P core only Xeon processor. No AMX or AVX512. They are simply lowering cost per watt by dramatically reducing performance per core. It comes down to your applications whether or not this kind of tradeoff is worthwhile. At some point you will probably run into a wall where applications in this category are increasingly better served running on GPUs with far better performance per watt than anything intel CPUs can deliver.

Laptops

[backslashdot]

When is this tech going into laptops? AKA where it is needed.

Re:

[CAIMLAS]

They're taking this tech from the mobile chips - it's just a nerf'ing of core and instruction capabilities. There's nothing special about it.

If you want power/efficiency while retaining performance, you'll have to get off the archaic Intel architectures and move to ARM.

Re:

[tlhIngan]

They're taking this tech from the mobile chips - it's just a nerf'ing of core and instruction capabilities. There's nothing special about it.

If you want power/efficiency while retaining performance, you'll have to get off the archaic Intel architectures and move to ARM.

A lot of workloads are such that ultimate performance isn't needed - after all, if you're just tossing up files that doesn't need much performance since each core can handle many users at once.

Or, you can virtualize a lot of servers at once w

How they'll do it

[CAIMLAS]

So can I assume that, based on the performance of the last half dozen chipsets, this will mean a significant decrease in performance to achieve that threshold?

Double! power efficiency

[groobly]

Previous efficiency was 70%. New efficiency is 140%. It produces energy!

AVX512 Cake is a lie!

[cb88]

You can get double power efficiency and density scaling on Intel next YEAR!

Or you can just buy AMD and get it right now (and have your AVX512 cake too).

Apple Silicon

[gnasher719]

I remember vaguely that their efficiency cores have about three as much power per watt as the performance cores. I wonder how they would compare. And ARM server chips have quite high performance per watt as well.