• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

The Intel Atom Thread

Page 186 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
http://forum.kodi.tv/showthread.php?tid=270298&pid=2444740#pid2444740

http://forum.kodi.tv/showthread.php?tid=270298&pid=2446410#pid2446410

https://www.youtube.com/watch?v=HjXbAPXf_Vo

Kodi users have confirmed Apollo Lake supports full fixed function HEVC Main10 hardware decoding with very low CPU usage, this is not hybrid/partial decoding like previous generation.

http://forum.kodi.tv/showthread.php?tid=231955&pid=2408646#pid2408646

Kodi developer posting what VAAPI reports on Linux.

"VAProfileHEVCMain10 : VAEntrypointVLD" VLD = Variable Length Decoding.
 
Nice find guys.


nvgpu said:
http://forum.kodi.tv/showthread.php?tid=270298&pid=2444740#pid2444740

http://forum.kodi.tv/showthread.php?tid=270298&pid=2446410#pid2446410

https://www.youtube.com/watch?v=HjXbAPXf_Vo

Kodi users have confirmed Apollo Lake supports full fixed function HEVC Main10 hardware decoding with very low CPU usage, this is not hybrid/partial decoding like previous generation.

http://forum.kodi.tv/showthread.php?tid=231955&pid=2408646#pid2408646

Kodi developer posting what VAAPI reports on Linux.

"VAProfileHEVCMain10 : VAEntrypointVLD" VLD = Variable Length Decoding.
Click to expand...

That will make Apollo Lake viable for low cost HTPCs, and hardware decoding does wonders in terms of power consumption (great for mobile!).


mikk said:
Cinebench R15

Celeron N3150 (Braswell): 35 ST/130 MT
Celeron J3455 (Apollo Lake): 50 ST +42,8%/181 MT +39,2%

https://www.heise.de/ct/artikel/Int...leron-schneller-und-mit-4K-60-Hz-3457002.html
Click to expand...

Respectable IPC gain here. Goldmont beating Puma+ performance per clock:

Celeron J3455 (Apollo Lake - 1.5-2.3 GHz): 50 ST / 181 MT
A8-6410 (Beema - 2.0-2.4 GHz): 49 ST / 156-166 MT*

*NotebookCheck results
 
Sweepr said:
Nice find guys.




That will make Apollo Lake viable for low cost HTPCs, and hardware decoding does wonders in terms of power consumption (great for mobile!).




Respectable IPC gain here. Goldmont beating Puma+ performance per clock:

Celeron J3455 (Apollo Lake - 1.5-2.3 GHz): 50 ST / 181 MT
A8-6410 (Beema - 2.0-2.4 GHz): 49 ST / 156-166 MT*

*NotebookCheck results
Click to expand...


Pretty much bang on what I estimated - slightly higher than Puma IPC

@ 10w though, it's not exactly earth shattering, nor particularly compelling. We've had 15w Puma based solutions for years now @ this performance level. 15 to 10w doesn't make a huge difference in total platform power for the sort of devices these would go in (i.e small laptops, and fanless motherboards) to get excited about.

Essentially, these chips have found themselves in an awkward position, with little mainstream appeal.
 
majord said:
Pretty much bang on what I estimated - slightly higher than Puma IPC

@ 10w though, it's not exactly earth shattering, nor particularly compelling. We've had 15w Puma based solutions for years now @ this performance level. 15 to 10w doesn't make a huge difference in total platform power for the sort of devices these would go in (i.e small laptops, and fanless motherboards) to get excited about.

Essentially, these chips have found themselves in an awkward position, with little mainstream appeal.
Click to expand...

Not to mention, even AMD's E-350 (Brazos), and AM1 (Kabini - is that with "Puma" cores, or was that before Puma?) didn't see huge successes. (Initially, Brazos did, because it served a fairly unique niche, and had "Good Enough" performance. 1080P video decoding back then, on a 15W power budget, was pretty amazing. The only alternative solution to doing that, was Atom with an NV ION/ION2 chipset.)

I'm still looking forward to a Brix unit, with quad-core Goldmont cores, and HDMI2.0 output, with HEVC Main10 decoding. Could still be a winner, for desktop / media-consumption tasks.

Has anyone conclusively compared the Goldmont cores? Are they finally equivalent performance to a similarly-clocked Core2? Minus the AES benchmarks, of course, because Core2 never had those features, and they tend to skew overall benchmarks.
 
Sweepr said:
Respectable IPC gain here. Goldmont beating Puma+ performance per clock:

Celeron J3455 (Apollo Lake - 1.5-2.3 GHz): 50 ST / 181 MT
A8-6410 (Beema - 2.0-2.4 GHz): 49 ST / 156-166 MT*

*NotebookCheck results
Click to expand...
Wait... isn't that Puma only?

BTW... seems that Carrizo-L is defeated too.

And is near Nehalem Core i3 levels on MT... however in ST the difference is still big.
 
majord said:
Pretty much bang on what I estimated - slightly higher than Puma IPC

@ 10w though, it's not exactly earth shattering, nor particularly compelling. We've had 15w Puma based solutions for years now @ this performance level. 15 to 10w doesn't make a huge difference in total platform power for the sort of devices these would go in (i.e small laptops, and fanless motherboards) to get excited about.

Essentially, these chips have found themselves in an awkward position, with little mainstream appeal.
Click to expand...

Puma/Jaguar were power hogs back in the real world.
 
majord said:
Pretty much bang on what I estimated - slightly higher than Puma IPC

@ 10w though, it's not exactly earth shattering, nor particularly compelling. We've had 15w Puma based solutions for years now @ this performance level. 15 to 10w doesn't make a huge difference in total platform power for the sort of devices these would go in (i.e small laptops, and fanless motherboards) to get excited about.

Essentially, these chips have found themselves in an awkward position, with little mainstream appeal.
Click to expand...

Apollo Lake for notebooks is only 6W TDP (4W SDP). That's for a quad-core with >Puma+ level of performance per clock, respectable 18 EUs Gen 9 iGPU and Main10 hardware decoding. If anything it's the best small core solution from Intel yet, and there's no direct competitor this time (Stoney Ridge is 1M/2C at 15W TDP).
 
First Passmark score for J3455:

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+J3455+@+1.50GHz

CPU marks: 1960 MT, 678 ST

That is lower than I expected considering Bay Trail J1900 scores 1883 MT and 533 ST

However, it should be noted the J3455 score is achieved by using a single 4GB DDR3 1333 SO-DIMM:

http://www.passmark.com/baselines/V9/display.php?id=70259937880

With dual channel, I would expect the score for J3455 with dual channel to be at least 200 CPU marks higher. This based on comparing passmark cores of J1900 with single channel and dual channel.

Still even with a passmark score of 2200 MT, the performance is disappointing IMO.
 
Last edited:
cbn said:
First Passmark score for J3455:

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+J3455+@+1.50GHz

CPU marks: 1960 MT, 678 ST

That is lower than I expected considering Bay Trail J1900 scores 1883 MT and 533 ST

However, it should be noted the J3455 score is achieved by using a single 4GB DDR3 1333 SO-DIMM:

http://www.passmark.com/baselines/V9/display.php?id=70259937880

With dual channel, I would expect the score for J3455 with dual channel to be at least 200 CPU marks higher. This based on comparing passmark cores of J1900 with single channel and dual channel.
Click to expand...
Was this J4205 score posted yet?

https://www.cpubenchmark.net/cpu.php?cpu=Intel+Pentium+J4205+@+1.50GHz&id=2877

Looks nice from a single thread perspective, but the multithreaded score is not all that great. I guess it simply has to lower its frequency a lot to keep within its TDP rating...
 
  • Like
Reactions: cbn
majord said:
Pretty much bang on what I estimated - slightly higher than Puma IPC

@ 10w though, it's not exactly earth shattering, nor particularly compelling. We've had 15w Puma based solutions for years now @ this performance level. 15 to 10w doesn't make a huge difference in total platform power for the sort of devices these would go in (i.e small laptops, and fanless motherboards) to get excited about.

Essentially, these chips have found themselves in an awkward position, with little mainstream appeal.
Click to expand...

The decode on the Apollo lake is much than the Puma core chips, but yeah from a CPU standpoint these Goldmont cores @ 10W don't provide a gain.

My suspicion is that Intel will need to boost TDP to 15W to bring out the true potential of Goldmont quad core.
 
Brunnis said:
I guess it simply has to lower its frequency a lot to keep within its TDP rating...
Click to expand...

Yeah, that is what I am thinking.

One reason is that the single thread score @ 901 is actually better than an Athlon 5350 @ 803......but the Athlon 5350 (with its higher 25W TDP) wins on MT 2566 vs. 2329.
 
Here's a review of the Asrock J4205-ITX: http://www.technikaffe.de/anleitung...test__apollo_lake_mit_hdmi_2.0_und_hevc_10bit

This review leaves some things to be desired. They only seem to test with Cinebench R15 and from that testing they conclude that the J4205 is 20% faster in CPU and 25% faster in GPU than the old ASRock N3710-ITX.

Also, they state that the board doesn't support more than 1600 MHz RAM, when it's explicitly stated in all Apollo Lake documentation and Asrocks own product page that 1866 MHz is the maximum supported speed. They mention that the board is finicky with RAM, though, so we'll see how mine likes Corsair's 1866 MHz CL10 RAM...

Interestingly, they claim (if I've understood correctly) that the CPU manages to run single threaded loads at 2.4-2.6 GHz and multithreaded loads at 2.4GHz. The passive cooling manages to keep the CPU at 70C during AIDA64's stability test.

Finally, media decoding seems to work great. They tested several clips (including HEVC 10-bit) and observed low CPU utilization and no issues.

They do say that running Linux on it shows "massive problems". Asrock states that the board is compatible with Ubuntu 16.10, but it's unclear if Technikaffe tested that.

I'd call this test slightly inconclusive. The lack of benchmarks makes it hard to draw any hard conclusions. I intend to remedy the benchmark situation when I get my board. I'll be going away on thursday for a short trip and will be back on Sunday. Hopefully I'll get the board before that so I can provide some results before I go.
 
Nice find Brunnis, looking forward to your results. 52 ST / 192 MT makes it tie with the fastest Carrizo-L part (A8-7410) in ST and beat it in MT. This is basically Core m3 MT performance, very capable for basic usage.
 
Brunnis said:
Here's a review of the Asrock J4205-ITX: http://www.technikaffe.de/anleitung...test__apollo_lake_mit_hdmi_2.0_und_hevc_10bit

This review leaves some things to be desired. They only seem to test with Cinebench R15 and from that testing they conclude that the J4205 is 20% faster in CPU and 25% faster in GPU than the old ASRock N3710-ITX.
Click to expand...


It's a bit more, 30% in MT compared to their J3710 test.

http://www.technikaffe.de/anleitung-372-asrock_j3710_itx_mit_braswell_refresh_4_kern_cpu_im_test
 
Some pre release publicity noted that Apollo Lake was designed for a reduced bill of materials. I am curious where prices settle for the ASRock motherboards, but something bothers me. ASRock motherboards are stating to drift into retail distribution (at inflated prices) but most major motherboard manufacturers have not yet announced their retail plans for Apollo Lake motherboards. Where are the retail motherboard product announcements, or are there basically none? 14nm is yielding OK or better so an availability issue is unlikely. Did ASRock get some sort of preferred status at release? Are manufacturers waiting for LPDDR4 module availability?
 
dealcorn said:
Some pre release publicity noted that Apollo Lake was designed for a reduced bill of materials. I am curious where prices settle for the ASRock motherboards, but something bothers me. ASRock motherboards are stating to drift into retail distribution (at inflated prices) but most major motherboard manufacturers have not yet announced their retail plans for Apollo Lake motherboards. Where are the retail motherboard product announcements, or are there basically none? 14nm is yielding OK or better so an availability issue is unlikely. Did ASRock get some sort of preferred status at release? Are manufacturers waiting for LPDDR4 module availability?
Click to expand...

LPDDR4 ain't going to be used in cheapo devices, that's premium memory right there. DDR3L will be used for cheap systems, especially DIY boards, IMO.
 
I now have all the parts for my J4205-ITX build, except the J4205-ITX itself. However, I just got a message that the supplier received the part and will send it today. That probably means it will arrive on Friday and I'll be out travelling from Friday morning to Sunday evenening. Typical. 😛 I'll see if I can get some results up on Sunday night, otherwise it'll be Monday night instead.
 
Sweepr said:
Apollo Lake for notebooks is only 6W TDP (4W SDP). That's for a quad-core with >Puma+ level of performance per clock, respectable 18 EUs Gen 9 iGPU and Main10 hardware decoding. If anything it's the best small core solution from Intel yet, and there's no direct competitor this time (Stoney Ridge is 1M/2C at 15W TDP).
Click to expand...

There is a prototype 6W stoney ridge, so that could be interesting.
 
VirtualLarry said:
Not to mention, even AMD's E-350 (Brazos), and AM1 (Kabini - is that with "Puma" cores, or was that before Puma?) didn't see huge successes. (Initially, Brazos did, because it served a fairly unique niche, and had "Good Enough" performance. 1080P video decoding back then, on a 15W power budget, was pretty amazing. The only alternative solution to doing that, was Atom with an NV ION/ION2 chipset.)

I'm still looking forward to a Brix unit, with quad-core Goldmont cores, and HDMI2.0 output, with HEVC Main10 decoding. Could still be a winner, for desktop / media-consumption tasks.

Has anyone conclusively compared the Goldmont cores? Are they finally equivalent performance to a similarly-clocked Core2? Minus the AES benchmarks, of course, because Core2 never had those features, and they tend to skew overall benchmarks.
Click to expand...

I'm also waiting for brix or zotac apollo lake, but it needs to be 4 core and passive.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top