The Intel Atom Thread

[IntelUser2000]

Here are the two big issues I see with NUC and also laptops. They use a cheap platform but you have to replace all so the cost is not really far away from getting a DIY Core mITX setup. An mITX Pentium low power setup only needs a board replacement.

Also part of the throwaway culture. You'll likely end up with a weak link somewhere but you can't do anything about it while on an mITX board you get to choose.

By the way Jasperlake features 1.5MB L2 cache and 4MB L3 cache. So it's the first generation with a 3 level cache system.

 

[Roland00Address]

So is there any Intel Jasperlake reviews yet?

If I recall Intel was promising 6th Generation performance similar to Skylake. I wonder if we got an Intel Atom that is actually usable.

Not a review but there is two notebook check results with cinebench 11.5 scores and the results for the 6.0w tdp Intel n6000 quad core (goal is 4.8w SDP) is

1.36/1.46 single and 3.47/3.71 multi thread. For comparison the anandtech

0.31/1.24 z3740 T100
0.40/1.48 z3770 device not listened.

That is why I used a really old benchmark and not the more recent stuff for I was trying to compare old atom to new atom and that is 8.5 year old atom now.

 

[jpiniero]

Not a review but there is two notebook check results with cinebench 11.5 scores and the results for the 6.0w tdp Intel n6000 quad core (goal is 4.8w SDP) is

NBC does have two reviews of a laptop with the N6000.

Asus Vivobook 13 Slate tablet PC in review: OLED at its best

With low-end performance and a premium OLED and HDR display, the Asus Vivobook 13 Slate is a Surface contender on the one hand, but no competitor at all on the other. Jasper Lake focuses on content consumption, but performance is too weak for creators.

www.notebookcheck.net

Acer Swift 1 SF114-34 Review: Silent, long-lasting 14-inch laptop

We review the Acer Swift 1 SF114-34 with the Pentium Silver N6000, eMMC memory and a Full HD screen. How does the 14-inch notebook compare against the competition?

www.notebookcheck.net

 

[IntelUser2000]

If I recall Intel was promising 6th Generation performance similar to Skylake. I wonder if we got an Intel Atom that is actually usable.

That's for Gracemont, not Tremont. N6000 is Tremont. Gracemont is yet another 30% jump per clock from Tremont.

Bay Trail(Silvermont) to Cherry Trail(Airmont): Negligible
Cherry Trail to Apollo Lake(Goldmont): 30%
Apollo Lake to Gemini Lake(Goldmont Plus): 30%
Gemini Lake to Jasper Lake(Tremont): 30%
Tremont to Gracemont: 30%

The comparison ignores the clock speed gains. For comparison the N5000 gets 0.9 and 2.8 Cinebench R11.5 score.

Icelake 1065G7 gets 2.1 and 7.2 and it has Hyperthreading. 30% plus with Gracemont gets us to 1.9 and nipping at the heels of the chip that was considered top end in mobile back in 2019.

 

[DrMrLordX]

Gracemont is yet another 30% jump per clock from Tremont.

Gracemont hits much higher clocks than Tremont, at least as a part of Alder Lake. I wonder if the scaling is perfect for the last 900 MHz or so of Gracemont's clockspeed?

 

[Roland00Address]

That's for Gracemont, not Tremont. N6000 is Tremont. Gracemont is yet another 30% jump per clock from Tremont.

Bay Trail(Silvermont) to Cherry Trail(Airmont): Negligible
Cherry Trail to Apollo Lake(Goldmont): 30%
Apollo Lake to Gemini Lake(Goldmont Plus): 30%
Gemini Lake to Jasper Lake(Tremont): 30%
Tremont to Gracemont: 30%

The comparison ignores the clock speed gains. For comparison the N5000 gets 0.9 and 2.8 Cinebench R11.5 score.

Icelake 1065G7 gets 2.1 and 7.2 and it has Hyperthreading. 30% plus with Gracemont gets us to 1.9 and nipping at the heels of the chip that was considered top end in mobile back in 2019.

My mistake, thanks for the correction.

*sidenote* bad naming system where you have a two syllable word with a G, a three syllable word (with a plus), and a new two syllable word with G. All of them followed by Mont. (Gold vs Grace)

1.3*1.3*1.3*=2.47x multipler before clock speed, another 30% is 2.86% aka almost 3 times faster before clock speed 🤔

 

[dark zero]

Now makes me wonder how strong Alder Lake N is since is an efficient core uArch

 

[IntelUser2000]

Now makes me wonder how strong Alder Lake N is since is an efficient core uArch

It looks like they are putting it further up the stack, since it'll have up to 8 cores. That means the top chip isn't going to be at 10W either. I'm assuming it'll be raised to 25W.

 

[Exist50]

It looks like they are putting it further up the stack, since it'll have up to 8 cores. That means the top chip isn't going to be at 10W either. I'm assuming it'll be raised to 25W.

I doubt they go to 25W. Cheap laptops aren't going to want to support more than a 15W TDP, and the boost will probably be greatly limited compared to U/P. Hell, I bet a lot of their volume ends up being 4c configs.

 

[jpiniero]

I doubt they go to 25W. Cheap laptops aren't going to want to support more than a 15W TDP, and the boost will probably be greatly limited compared to U/P. Hell, I bet a lot of their volume ends up being 4c configs.

This is why I was kind of thinking that a 4 core die makes more sense. Maybe it will be 8 cores for embedded and the client parts would be cut down.

 

[Exist50]

This is why I was kind of thinking that a 4 core die makes more sense. Maybe it will be 8 cores for embedded and the client parts would be cut down.

Yeah, seems like a weird position. Cheapo Chromebooks and the like probably aren't going to benefit from 0+8 vs 1+4 or even 0+4. Would the embedded market, even?

 

[dark zero]

It looks like they are putting it further up the stack, since it'll have up to 8 cores. That means the top chip isn't going to be at 10W either. I'm assuming it'll be raised to 25W.

Maybe premium portables might go there?
Or... maybe it can be put only on Mini PC format, AIO and Mini ITX for some people who wants it.
But Octa Cores might be the way this time.

 

[NostaSeronx]

If it is based on int_rate...

Amber Lake-Y = 4c/8t 7W => Baseline
Alder Lake-N = 8c/8t 7W => ~1.8x performance at same power or ~1.4W at same performance, given TDP=CPU-task powerload only.

4c/8t + 24EUs => 8c/8t + 32EUs makes it a viable cost-orientated replacement of 14nm-Y series.

 

[IntelUser2000]

I doubt they go to 25W. Cheap laptops aren't going to want to support more than a 15W TDP, and the boost will probably be greatly limited compared to U/P. Hell, I bet a lot of their volume ends up being 4c configs.

I agree to an extent for laptops.

I was talking about taking the current 10W on desktops to 25W. For laptops I wouldn't be surprised at 10W but I doubt they'll go much above that. They have been raising TDPs ever so slightly. They went from 6W to 7W in the Tremont generation for few designs.

If it is based on int_rate...

Amber Lake-Y = 4c/8t 7W => Baseline
Alder Lake-N = 8c/8t 7W => ~1.8x performance at same power or ~1.4W at same performance, given TDP=CPU-task powerload only.

I don't know why you mention Amberlake. It wasn't that good of a product.

Tremont N6000 already outperforms it in multi core Cinebench by 25%+. Even in single thread it doesn't fall behind at all. The 4 core Gracemont successor should be close to beating Amberlake 10510Y by 60% in MT and 30%+ in ST.

 

[NostaSeronx]

I don't know why you mention Amberlake. It wasn't that good of a product.

Amber Lake-Y(4c/8t) is what Alder Lake-N is replacing.

Tremont N6000 already outperforms it in multi core Cinebench by 25%+. Even in single thread it doesn't fall behind at all. The 4 core Gracemont successor should be close to beating Amberlake 10510Y by 60% in MT and 30%+ in ST.

With Cinebench R20:

N6000 (20W/6W);
ST=278
MT=774

i7-10510y (24W/7W);
ST=328
MT=891

R23:
N6000 (20W/6W);
ST=759
MT=2036

i7-10510y (24W/7W);
ST=969
MT=2638

My understanding is that 8c/8t Gracemont is needed to beat 4c/8t Skylake. So, 4-core would probably just be marginal increase compared to N6000. While 8-core would give the same thread count and higher scores overall. 8c/8t should also replace Icelake-Y do to it doing horribly as well, specifically the bad configured variants.

128-bit 4c/8t Amber Lake-Y(Comet Lake-Y variants) -> 64-bit 8c/8t Alder Lake-N.
64-bit 4c/8t Ice Lake-Y(i5-1030G1 LowTDP variant of i5-1035G1) -> 64-bit 8c/8t Alder Lake-N.

 

[Exist50]

Amber Lake-Y(4c/8t) is what Alder Lake-N is replacing.
With Cinebench R20:

N6000 (20W/6W);
ST=278
MT=774

i7-10510y (24W/7W);
ST=328
MT=891

R23:
N6000 (20W/6W);
ST=759
MT=2036

i7-10510y (24W/7W);
ST=969
MT=2638

My understanding is that 8c/8t Gracemont is needed to beat 4c/8t Skylake. So, 4-core would probably just be marginal increase compared to N6000. While 8-core would give the same thread count and higher scores overall. 8c/8t should also replace Icelake-Y do to it doing horribly as well, specifically the bad configured variants.

128-bit 4c/8t Amber Lake-Y(Comet Lake-Y variants) -> 64-bit 8c/8t Alder Lake-N.
64-bit 4c/8t Ice Lake-Y(i5-1030G1 LowTDP variant of i5-1035G1) -> 64-bit 8c/8t Alder Lake-N.

N is the low cost platform. The Y series replacement is Alder Lake M.

 

[NostaSeronx]

N is the low cost platform. The Y series replacement is Alder Lake M.

While N is the low cost platform, the N-series is positioned to replace older Y-series with reduction of price.

Alder Lake M is the static ASP/higher performance replacement of prior Y-series.
Alder Lake N is the reduction ASP/same performance replacement of prior Y-series.

I can't find the sanctioned Intel slides of Jasper Lake vs 8th Gen Amber Lake, but there are somewhere, probably in Chinese.

8th Gen Y-series (2c/4t) => Jasper Lake Quad-core;

ASUSTeK COMPUTER INC. Vivobook Slate T3300KA_T3300KA vs Microsoft Corporation Surface Go 2 - Geekbench

10th Gen Y-series (4c/8t) => Tiger Lake-Y, but will most likely have a cost-reduced option with Alder Lake-N.

Since, Tiger Lake-Y is superseded by Alder Lake-M. The P-core addition maintains the $300-$400 price point, while the E-core only+64-bit DDR5+etc pushs the cost-reduction to sub-$250, etc.

 

[Brunnis]

Here are Geekbench 4 and 5 runs on Pentium N6005 (Shuttle DL20N) running a fresh Ubuntu 22.04 installation:

Shuttle Inc. DL20N - Geekbench

Benchmark results for a Shuttle Inc. DL20N with an Intel(R) Pentium(R) Silver N6005 @ 2.00GHz processor.

browser.geekbench.com

Shuttle Inc. DL20N - Geekbench

Benchmark results for a Shuttle Inc. DL20N with an Intel Pentium Silver N6005 processor.

browser.geekbench.com

These were run by my colleague that has access to this system. It's a pretty hefty upgrade compared to the J5005.

EDIT: The system is configured with 2x8GB DDR4-3200.

 

[LightningZ71]

Usable. That hits like an I5-3470 and slightly better than a ryzen 2700u. I get good use out of my 2500u, so, that looks decent.

 

[gdansk]

I think it is fine but damn is it late. If it was actually available in 2020 that would have been good. Instead only now am I starting to see Tremont designs available. And the rumored Alder Lake N would be much better for desktop use.

 

[jpiniero]

I think it is fine but damn is it late. If it was actually available in 2020 that would have been good. Instead only now am I starting to see Tremont designs available. And the rumored Alder Lake N would be much better for desktop use.

Launch was Jan 2021 and Intel clearly charges more so OEMs are more inclined to use the 14 nm ones. But it's well past two years since the launch of Gemini Lake Refresh and at some point Intel will EOL it. That might be why you are seeing more Tremont designs out there recently.

 

[Roland00Address]

Here are Geekbench 4 and 5 runs on Pentium N6005 (Shuttle DL20N) running a fresh Ubuntu 22.04 installation:

Shuttle Inc. DL20N - Geekbench

Benchmark results for a Shuttle Inc. DL20N with an Intel(R) Pentium(R) Silver N6005 @ 2.00GHz processor.

browser.geekbench.com

Shuttle Inc. DL20N - Geekbench

Benchmark results for a Shuttle Inc. DL20N with an Intel Pentium Silver N6005 processor.

browser.geekbench.com

These were run by my colleague that has access to this system. It's a pretty hefty upgrade compared to the J5005.

EDIT: The system is configured with 2x8GB DDR4-3200.

Amazing that is tied with ST in geekbench 5 with the A10 iPad I am using to type this forum post out, and almost twice as fast in multi

Shuttle Inc.
774 Single-Core Score
2415 Multi

7th Gen cheap iPad from 2019
747 Single-Core Score
1383 Multi

Of course the cheap iPad from 2021 is now a13 based and 1311 ST and 3077 Multi. Apple is using 2 Big Core and 4 Little Core on the a13, and 2+2 with the a10. And for almost $200 more you can get the ipad Mini with the a15.

What I am trying to signify is the worse of the worse cheap chips are now actually usable. Apple is still much faster for their low bin, and with even better power efficiency but Atom is useable at least. (Also remember Apple has better fabs with TSMC 7nm vs 10nm Intel but they are close enough where Intel is calling a newer version of 10nm Intel 7.). And the next gen Tremont is 30% faster.

 

[dark zero]

So... Alder Lake N would be the new bare minimun the industry needs?
I mean if Tremont already is usable... Alder Lake -N would be definately better.

 

[NostaSeronx]

081_ID1 = 3W to 15W cTDP, 35W Boost
041_ID2&3 = 3W to 6W cTDP, 25W Boost
021_ID4 = 3W to 6W cTDP, 25W Boost

 

[dark zero]

View attachment 62184

081_ID1 = 3W to 15W cTDP, 35W Boost
041_ID2&3 = 3W to 6W cTDP, 25W Boost
021_ID4 = 3W to 6W cTDP, 25W Boost

Mmm... and there are 4 of them.

Maybe the 1st one is the Octa Core going up to 3.5 Ghz single core and 3.2 Ghz all of them? because having 35W boost is higher than expecting. Is the Pentium Silver this time.

Also there are the 2 and 3 which might be Quad Cores but at different clock rate.
If that so, expecting 3.0 Ghz and 2.5 Ghz. Maybe they are the Celerons.

And the last one seems to be a Dual Core... but it makes no sense since the uArch states that the bare minimun is being a Quad Core.