Intel Unveils All New Tremont Low Power CPU Architecture

MojoKid writes: Intel took the wraps off a new CPU architecture today, one that is employed in its next-gen Lakefield series CPUs with Foveros chip tech that will power some next-gen Microsoft Surface devices, among others in 2020. Codenamed Tremont, it's a low-power, 10nm x86 microarchitecture that is the successor to Goldmont Plus, which was is found in current-gen Atom, Pentium Silver, and some Celeron series processors. Tremont targets compact, low-power packages and incorporates a number of updates to the instruction set, enhanced security features, and more advanced power management. Tremont delivers significant IPC (Instructions per Cycle) improvements as well, gen-over-gen versus Intel's current low power x86 architectures, to the tune of about 30%. Intel beleives Tremont-based processors will enable a new wave of innovative form factors for client devices, more efficient data center products, and new-found applications for the internet of things (IoT).

Intel Atom

[sexconker]

Intel Atom. It won't suck this time, we promise!*

*Not an actual promise.

See also Intel's iGPUs.

Re:

[Rockoon]

Imagine an Atom running with all the performance-sucking software-based OS hacks needed to mitigate the security nightmare of Intels flawed designs.

Re:

[tepples]

Atom originated as an inorder processor without all this dangerous speculation. The intent was to use thread-level parallelism (Hyper-Threading Technology) to compensate for dropping the transistors needed for reordering and speculation. At the time, it worked; an Atom N450 was about as fast as a similarly clocked (1.6 GHz) Pentium 4 processor at a much lower TDP (single digit watts).

Re:

[drinkypoo]

Unfortunately, what didn't work was the absolute garbage GPU they used with those processors. Even full screen video which is within the realm of the processor's capabilities is abusive to that junk.

Re:

[UnknownSoldier]

Haha! Would be sad if it was so funny!

What IS the market for these chips?

Re:

[lactose99]

Atom doesn't suck, especially since Silvermont. Its great for a number of uses, particularly low-power ones. Beats the pants of AMD's APU in my benchmarks.

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[bluefoxlucid]

Waiting for AMD to release an FPGA.

Re:

[Tough Love]

10 nanometer. It won't suck This time! We! Promise!! **

** Also not an actual promise

I have trouble with all the code names...

[AbRASiON]

Is this going to mean a new Intel Nuc (low power) model? There's about 2 now I think, very low power very good price which are nearly excellent htpc devices. They just need HDMI 2.1.

Mind you since I started reading this stuff, even if Intel address 4k HDMI 2.1 HDR etc, the next big thing will be an av1 decoder in hardware and I can't see these chips doing that.

Re:

[sexconker]

NUCs run laptop Core CPUs, not Atom CPUs. Atom CPUs are in things like NAS boxes and other crap that runs like ass.

Re:

[AbRASiON]

You're right, I'm thinking of the Celeron and Pentium NUCs (prob still 'Core' cores on it)

https://ark.intel.com/content/... [intel.com]

https://ark.intel.com/content/... [intel.com]

One or two of those are particularly well priced and decent for a HTPC - but unfortunately, lacking a couple of key features to buy long really long term viability for it.
I bet they finally update Denverton with something decent, shocked how slow my NAS is - I thought Denverton was going to be great from the articles, but man no, Avoton must have just b

Re:

[Tough Love]

Is this going to mean a new Intel Nuc (low power) model?

I don't really care, to tell the truth. I got an Atom NUC last time round, that's the last one. Runs too hot and slow. I am comfortable waiting for the AMD NUC now :)

Re:

[AbRASiON]

Low power and performance isn't their strong suit.

The current 8 series bean canyon product is very fast, reliable and feature packed.

I'd very happy if AMD release a competing product and would gladly consider it, but it does need to match up.

Especially in video out capability, input output options etc

Re:

[Tough Love]

Oh right, video out... you mean, like displayport hooked up to NAVI?

All new?

[AmiMoJo]

It doesn't seem all new. Same x86 architecture from the 70s, and once again it's an iteration on the previous generation and some borrowed bits from other CPUs in their line-up.

Re:

[Anonymous Coward]

It's not the x86 architecture from the 1970s. The only thing a modern Intel chip has in common with the 8086/8 is the ISA, but the ISA is implemented in an entirely different way.

Re:All new?

[RightSaidFred99]

Yeah, like those stupid cars we still drive to this day. Almost all of them have 4 wheels, a steering wheel, the radio in the middle of the front console, etc... Same old early 20th century design!

Re:

[AmiMoJo]

I didn't say it was bad... I mean, it is, Atom chips have never been good.

Re:

[AmiMoJo]

Hmm, Intel shilling now? That's a new one for Slashdot.

Re:

[ArchieBunker]

You should read up on the AS/400 architecture. It's so different than anything else. https://www.kernel.org/doc/ols... [kernel.org]

Re:

[drinkypoo]

You should read up on the AS/400 architecture. It's so different than anything else. https://www.kernel.org/doc/ols [kernel.org]...

Except it isn't any more. It's a decoder on the front of a PowerPC.

Re:

[Rockoon]

Except it isn't any more. It's a decoder on the front of a PowerPC.

..and isnt one of the big complaints against x86 that it is also a big decoder in front of risc core?

The fact of the matter is that RISC cant deliver performance once you go superscaler. Those "complex" instructions arent doing unnecessary operations.. on the contrary they are delivering necessary operations to the pipeline more efficiently (performance) than "reduced" instructions. They tried to widen instruction bandwidth on the Alpha, and found out that its not good enough.

Re:

[drinkypoo]

..and isnt one of the big complaints against x86 that it is also a big decoder in front of risc core?

It is, but it's nonsense. The x86 decoder is a minuscule portion of a modern CPU. I don't know how big the decoder is on AS/400, and I don't really care actually; It's not a complaint, simply an observation.

The fact of the matter is that RISC cant deliver performance once you go superscaler.

All x86 CPUs have been internally RISCy since the K5, so RISC has already taken over the part where it's valuable; and amd64 has solved the big problems with x86 (with more registers which are actually general purpose, not like in x86 where operations have to source from and deliver the result to specifi

Re:

[Rockoon]

wooosh

Only advertising.

[Gravis Zero]

There are only basic specs and propaganda numbers. There is no definitive release date much less a real performance comparison.

This is just an advertisement because Intel is flailing as they are losing traction in the market.

Re:

[gaelfx]

Not to mention the lack of any info about actual power draw on the processors. Is it low compared to Intel's other processors? Probably. Compared to ARM based processors? Probably not even close.

Vulnerable to MELTDOWN and SPECTRE?

[drinkypoo]

No mention is made in the article whether this architecture was designed to do access checks before accesses, so that it won't be vulnerable to MELTDOWN attacks, nor is any mention of SPECTRE made. However, the marketing image does say "Intel(R) Core(TM) class branch prediction", which implies that it will be vulnerable — since Core's branch prediction is vulnerable.

Re:

[drinkypoo]

Guess what? it's almost certainly going to be vulnerable [arstechnica.com]:

Unlike earlier Intel mobile CPUs, Tremont offers what Intel refers to as "Core-class branch prediction"â"meaning it's as good at speculative execution as the mainstream Core class of processors. I pressed Intel Senior Principal Engineer Stephen Robinson pretty hard on the security aspects of this pointâ"specifically, does Core-class branch prediction mean Core-class speculative execution vulnerabilities to exploits such as Spectre and Meltdo