The Intel Atom Thread

[Jan Olšan]

Oho, that's going to be something then

 

[imported_bman]

Absolutely. Gemini Lake features Gen 9 graphics / Gen 10 display.

Forgot about eMMC 5.1 support too.

What is Gen 10 display? Is it just HDMI 2.0 or does it include Freesync/VRR/DP 1.4?

 

[nvgpu]

http://images.anandtech.com/doci/10635/NGAtom.png

Gen10 Display Engine from Cannonlake 10nm was backported to 14nm++ so native HDMI 2.0 support can be a feature for Gemini Lake while still using the Gen9 graphics architecture from Skylake.

 

[jpiniero]

http://images.anandtech.com/doci/10635/NGAtom.png

Gen10 Display Engine from Cannonlake 10nm was backported to 14nm++ so native HDMI 2.0 support can be a feature for Gemini Lake while still using the Gen9 graphics architecture from Skylake.

Interesting.. Coffee Lake does not have this apparently.

 

[dark zero]

Interesting.. Coffee Lake does not have this apparently.

Ouch... Another gen to skip it.

 

[Sweepr]

Intel Compute Card specs leaked (Apollo Lake and Kaby Lake models)

The Kaby Lake models will also likely be more expensive. The last I’d heard, Apollo Lake Compute Cards would probably cost around $150, while top-of-the-line models would be closer to $500. But the cards aren’t really aimed at consumers at this point. They’d most likely be sold to device makers and developers. Keep in min, they don’t have any standard USB or HDMI ports and instead use a special connector for USB and display connectivity.

Anyway, here’s a run-down of what to expect from the first four models:

• CD1C64GK – Celeron N3450 quad-core CPU/4GB RAM/64GB eMMC storage
• CD1P64GK – Pentium N4200 quad-core CPU/4GB RAM/64GB eMMC storage
• CD1M3128MK – Core M3-7Y30 dual-core CPU/4GB RAM/128GB PCIe SSD storage
• CD1IV128MK – Core i5-7Y57 dual-core CPU/4GB/128GB PCIe SSD storagequote]

https://liliputing.com/2017/05/intel-compute-card-specs-leaked-apollo-lake-kaby-lake-models.html

 

[nvgpu]

http://pc.watch.impress.co.jp/docs/news/event/1063203.html

Gemini Lake products under development according to roadmaps from Asrock.

 

[Arachnotronic]

http://pc.watch.impress.co.jp/docs/news/event/1063203.html

Gemini Lake products under development according to roadmaps from Asrock.

It's also worth noting that the Z270 boards are shown as SKL + KBL, but Coffee Lake is a platform of its own. I think this suggests lack of CFL compatibility with Z270. Bummer.

 

[nvgpu]

https://downloadcenter.intel.com/download/26836/Graphics-Intel-Graphics-Driver-for-Windows-15-45-

https://communities.intel.com/thread/115054

Intel 21.20.16.4678 driver for Apollo Lake.

 

[nvgpu]

https://downloadcenter.intel.com/download/26941/Graphics-Intel-Graphics-Driver-for-Windows-15-46-

https://communities.intel.com/thread/116390

Intel 22.20.16.4729 driver for Apollo Lake. New driver branch with OpenGL 4.5 support & WDDM 2.2 support on Creators Update 1703.

 

[Sweepr]

Intel's little cores are growing up; Goldmont Plus (core name #GeminiLake) appear to feature a 4-way decode - 2x width of Silvermont

For reference, Goldmont (core name Apollo Lake) has a 3-way decode. This uarch is undergoing radical changes. Within same power envelope.

https://twitter.com/WikiChip

Thanks to nvgpu for the heads up. The 31% CPU performance uplift compared to Apollo Lake makes more sense now.

 

[Nothingness]

Previous leaks said Goldmont Plus was the same as Goldmont with L2 cache size increased from 2MB to 4MB.

The 31% performance increase seems to be against Braswell not Apollo Lake. And do you really think such a different uarch would be named Goldmont Plus? I bet that tweet is wrong.

EDIT: I might be the one wrong here!

https://patchwork.kernel.org/patch/9836747/

 - Goldmont Plus has 4-wide pipeline for Topdown

Unexpected to say the least!

 

[Arachnotronic]

Previous leaks said Goldmont Plus was the same as Goldmont with L2 cache size increased from 2MB to 4MB.

The 31% performance increase seems to be against Braswell not Apollo Lake. And do you really think such a different uarch would be named Goldmont Plus? I bet that tweet is wrong.

EDIT: I might be the one wrong here!

https://patchwork.kernel.org/patch/9836747/

 - Goldmont Plus has 4-wide pipeline for Topdown

Unexpected to say the least!

More innovation from Intel's Atom team than from its Core teams these days, lol

 

[mikk]

The 31% performance increase seems to be against Braswell not Apollo Lake.

It is compared to Apollo Lake.

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/the-intel-atom-thread.2341969/page-192#post-38911168

 

[IntelUser2000]

More innovation from Intel's Atom team than from its Core teams these days, lol

That coincides with Charlie's claim(at S|A) that Intel's going to converge some of their cores. Also its much easier to improve low power/low end performance than absolute high end where thermals and clock is against a red brick wall.

I'd like to see a true big core, because Core seems like a medium core nowadays. Ever since they were able to stick the Core chips into a 4.5W power envelope, it stopped being a big core.

On a side note Intel needs all three decode pipelines(since its based on Goldmont) capable of executing AVX-512 instructions for Knights Hill to be anywhere near competitive with Nvidia by that time. Yes, 3 FMA capable AVX-512 units per Knights Hill core.

 

[Roland00Address]

I'd like to see a true big core, because Core seems like a medium core nowadays. Ever since they were able to stick the Core chips into a 4.5W power envelope, it stopped being a big core.

So I think I understand what you meant by this comment IntelUser2000, I think I understand your intent.

But let me have fun anyway and ask a question that is not your intent via redirecting the words you used in the quote I just requoted

Currently we have the small Intel Atom Core

And we have the medium size core i series kabylake which in its lowest tdp version we have a dual core plus graphics in a 4.5w tdp (and I would argue the tdp is even lower since with c TDP down these processors can have a 3.5w tdp core m3 7y30, i5 7y54, i5 7y55, i7 7y75)

So my question is, while it is not an Intel design core and done by a different team, what would AMD Ryzen / Zen chips be?

Would that be a medium core, or a high end core which focuses on total performance instead of trying to balance performance per watt / performance per tdp?

 

[dark zero]

Wondering how strong will be Gemini Lake against the ARM A75

 

[Sweepr]

More bits about Gemini Lake, thanks FanlessTech (and @nvgpu for the heads up):

"Add support for Gemini Lake (aka. GLK)
- Decoding: H.264/MPEG-2/VC-1/JPEG/VP8/HEVC/HEVC 10-bit/VP9/VP9 10-bit"

https://01.org/linuxgraphics/downloads/2017q2-intel-graphics-stack-recipe

"25 x 24 mm package size and 9.5 x 9.9 die size for Gemini Lake. It's coming."

https://twitter.com/FanlessTech/status/890887475186610177

 

[VirtualLarry]

Maybe my next HTPC? I do so love my Brix J1900 unit though.

 

[IntelUser2000]

More bits about Gemini Lake, thanks FanlessTech (and @nvgpu for the heads up):

https://01.org/linuxgraphics/downloads/2017q2-intel-graphics-stack-recipe

https://twitter.com/FanlessTech/status/890887475186610177

9.5mm x 9.9mm die size for Geminilake(Goldmont Plus). That's nice. They must have lived up to their promises according to here: https://seekingalpha.com/article/1882851-intels-broxton-die-size-revealed

Bay Trail: 102mm2
Cherry Trail(25% smaller): 76.5mm2
Apollo Lake(Goldmont): ??
Gemini Lake(Goldmont Plus): 94.05mm2

So they increased die size by 23% on the same 14nm process while going from 2-wide issue to a 4-wide issue CPU core, improve performance by 60-70%(1.3x + 1.3x), and the graphics would go from Gen 7 16EU to Gen 9 18EU. That's really nice.

 

[FanlessTech]

Thanks guys! Yes, most exciting release since the out-of-order Bay Trail 4 years ago. The "Atom" is doing better than ever.

Intel is also giving tips to its partners regarding passive cooling, which is music to my ears

 

[IntelUser2000]

Somewhat off topic:

Intel uses Atom cores for Xeon Phi.

The current Knights Landing implementation uses Silvermont-based cores. The next one called Knights Hill are supposed to use Goldmont cores. Though that was before the 14/10nm problem was known. Could it be possible they'll use Goldmont Plus on Knights Hill?

 

[Arachnotronic]

Somewhat off topic:

Intel uses Atom cores for Xeon Phi.

The current Knights Landing implementation uses Silvermont-based cores. The next one called Knights Hill are supposed to use Goldmont cores. Though that was before the 14/10nm problem was known. Could it be possible they'll use Goldmont Plus on Knights Hill?

I think Knights Hill has been cancelled.

 

[CHADBOGA]

I think Knights Hill has been cancelled.

So what does this mean for future products of Xeon Phi?

Has Intel decided there isn't a market for Xeon Phi?

 

[majord]

Interesting! Would this make Goldmont a finalist for the list of shortest lived uArch in history award?

Very strange it was launched less than a year ago, has barely been put into anything, and is being replaced by a heavily revised core. Must have been a huge overlap in design phase