The Intel Atom Thread

[coercitiv]

I think Intel would be the first to spill the beans on how good their 10SF yields are.

 

[moinmoin]

I think Intel would be the first to spill the beans on how good their 10SF yields are.

Aren't they SuperFine?

 

[Thala]

ARK page doesn't say SuperFin.

Also 70% over Apollo Lake in ST and 50% in MT is not a lot. That suggests similar performance to Gemini Lake in ST and MT.

And what's up with the separate PCH? So Jasper Lake will be a different config? They needed the additional features is that it?

Should still be faster than Gemini Lake - slightly higher IPC plus higher Turbo. What is more disturbing is that the 4 cores are at least rated 9W - thats quite a lot.

 

[LightningZ71]

I don't know about those "stellar" yields. That atom die is tiny from what I saw in another post. Tiny dies are very friendly for poorly yielding foundry nodes...

 

[Exist50]

https://twitter.com/x/status/1309093615684276225

Yeah Ian said that Intel provided an updated press release after he followed up with them. It appears engineering - marketing got wires mixed and I guess that’s why Ark says what it does

Still wrong. Marketing still has their wires crossed, apparently.

 

[mikk]

https://twitter.com/x/status/1309260853678682112

As expected Elkhart Lake is using the Icelake 10nm process. Looking good for Gracemont. If they can boost up to 3.0 Ghz on this subpar 10nm and 10W design a 4.0+ Ghz boost for Gracemont should be doable.

 

[jpiniero]

I don't know about those "stellar" yields. That atom die is tiny from what I saw in another post. Tiny dies are very friendly for poorly yielding foundry nodes...

And yet they felt the need to have SKUs cutting down to 2 cores and 16 EUs... and there's one model with no IGP at all. The driver said that it's possible to go down to 8 EUs, but it doesn't look like Intel went there.

 

[LightningZ71]

Very friendly = able to produce at least something that can generate revenue. It doesn't mean that it's going to set the world on fire.

 

[IntelUser2000]

As expected Elkhart Lake is using the Icelake 10nm process. Looking good for Gracemont. If they can boost up to 3.0 Ghz on this subpar 10nm and 10W design a 4.0+ Ghz boost for Gracemont should be doable.

Technical sources like ARK, are always more correct over the rest. No mention of SF on Intel press was another reason.

Never take things at a face value.

 

[borandi]

Technical sources like ARK, are always more correct over the rest. No mention of SF on Intel press was another reason.

Never take things at a face value.

Incorrect. The number of times I'm emailing corrections to the Ark team. They simply put the details in as given to them by the respective business units.If the BUs make an error, it goes through into Ark.

 

[IntelUser2000]

Results for Elkhart and Jasper Lake show the ~30% per clock gain over Goldmont Plus.

Pentium Silver N6000(6W, laptop)
Cinebench R20 1T: 279
MT: 774

40-50% over the N5040 result and 50-60% over the best N5000 result.

 

[jpiniero]

Results for Elkhart and Jasper Lake show the ~30% per clock gain over Goldmont Plus.

Pentium Silver N6000(6W, laptop)
Cinebench R20 1T: 279
MT: 774

40-50% over the N5040 result and 50-60% over the best N5000 result.

What's the actual model number? Intel hasn't announced any Pentium Silvers yet.

 

[IntelUser2000]

That IS the actual model name. I didn't make it up.

 

[jpiniero]

That IS the actual model name. I didn't make it up.

The MT score is also a lot higher than what NBC got with Lakefield, and that's with the Sunny Cove core. It even beat the Lakefield result in single core.

 

[IntelUser2000]

The MT score is also a lot higher than what NBC got with Lakefield, and that's with the Sunny Cove core. It even beat the Lakefield result in single core.

Lakefield scores crap for some reason.

By your logic Lakefield should be at top, but its not. Lakefield's R15 MT scores the same as Goldmont Plus-based N5000.

 

[LightningZ71]

That Mt score is in between the Athlon 200g and 220g. That’s... not bad.

 

[Bouowmx]

The 279/774 Cinebench R20 score got entered into CPU-Monkey database, which also has the Pentium Silver N6005 (10 W) scoring 294/925: CPU-Monkey
On the AnandTech Bench, a Core i3-4130T (2.9 GHz Haswell) scores 277 cb single-thread.

 

[LightningZ71]

Not bad company for the 6005. The i7-3770 scores just a few more points in single thread, and the Ryzen 2700u scores just ten more points in MT CR20. This is a solid little chip for those few watts.

(EDIT: The numbers in the CPU-Monkey database are incorrect for the 2700u and have it listed at around 50% of its real performance)

 

[IntelUser2000]

The Ryzen 7 2700U gets nearly double the N6000 score. It's a wider uarch plus it supports SMT after all.

Just in Geekbench Integer(4 and 5) Goldmont Plus outperforms Nehalem by 5% or so. In Cinebench its on the level of Core 2.

Translating that to Tremont, in Geekbench it'll be on par with Broadwell. Only another 10% is needed to put it on par with Skylake. It's more realistic to expect Ivy Bridge levels based on Cinebench IMO.

 

[mikk]

If the source is CPU-Monkey, it's a bad source. They usually use estimated scores, it's not a real score. It happened several times in the past that their scores were higher than the actual product launched later. Their identical iGPU - FP32 scores looks nonsense to me and calling it Tigerlake G4 is wrong as well.

 

[LightningZ71]

Yes, I'd like to retract my earlier comparison with the 2700u. CPU-Monkey has the wrong score in for the 2700u multi. Still, not a bad showing for a 10w processor. What I found surprising is the inclusion of the G1 spec Xe iGPU at 48EU. That's quite an upgrade over previous generations.

 

[Roland00Address]

Results for Elkhart and Jasper Lake show the ~30% per clock gain over Goldmont Plus.

Pentium Silver N6000(6W, laptop)
Cinebench R20 1T: 279
MT: 774

40-50% over the N5040 result and 50-60% over the best N5000 result.

So if I understand correctly (no expert on modern cinebench) this is roughly comparable performance as a dual core i5* with hyper threading between Haswell (4th Gen) and Kabylake (7th Gen) in the U series Ultrabook form factor. Thus cheap stuff will now have cpu performance as medium range ultra books of roughly 4 years ago. What was $700 then will now cost $300 maybe $400?

Is this everyone else’s sense?

(To achieve these numbers with 4 core / 4 threads comparable to 2 cores / 4 threads I sense better turbo and other energy optimization.)

 

[IntelUser2000]

@Roland00Address The 6th and 7th gen Core chips are still faster in single thread. Skylake is 30% faster per clock.

It has a multi-threading win only because it has 4 cores which is better than Hyperthreading, rather than clocks.

But in a general sense you are right.

Also Jasperlake has Gen 11 based graphics, not Xe. The only Xe graphics chip is still Tigerlake.

 

[LightningZ71]

Tha does make more sense.

 

[mikk]

Celeron N5100 https://browser.geekbench.com/v5/cpu/4277542
Celeron N4500 https://browser.geekbench.com/v5/cpu/4240214

This is Jasper Lake I believe.

N5000 https://browser.geekbench.com/v5/cpu/3183632

Google Android 9 N5000 for comparison.