The Intel Atom Thread

[Brunnis]

I'm looking into getting the benchmarks I need for testing the J4205. What about version of x264 HD benchmark? Is it 5.0.1 that should be used? Also, what version of GFXBench is most interesting to get results from?

 

[Sweepr]

I'm looking into getting the benchmarks I need for testing the J4205. What about version of x264 HD benchmark? Is it 5.0.1 that should be used? Also, what version of GFXBench is most interesting to get results from?

For GFXBench try the latest version. Regarding X264, maybe an older one, so we can compare it with Braswell / Beema (5.0 / 5.0.1?).

 

[Brunnis]

For GFXBench try the latest version.

Okay, I'll get version 4 then.

Regarding X264, maybe an older one, so we can compare it with Braswell / Beema (5.0 / 5.0.1?).

5.0.1 is the most recent version I could find (and it's from 2012). I'll use this version unless I get any other recommendation.

 

[nvgpu]

http://bluesky23.yukishigure.com/en/DXVAChecker.html

Run the latest version of DXVA Checker and take a screenshot in PNG, would be good to see what it reports.

 

[nvgpu]

http://www.phoronix.com/scan.php?page=news_item&px=Intel-Geminilake-Enablement

 

[nvgpu]

https://downloadcenter.intel.com/download/26404

Intel 21.20.16.4542 driver for Apollo Lake.

 

[nvgpu]

https://twitter.com/TMFChipFool/status/796873235816136704

Intel's first 10-nanometer Atom is called MERCURY LAKE.

 

[Brunnis]

I actually got my J4205-ITX last night; one day earlier than expected. If everything had worked out, I would have posted a bunch of benchmarks yesterday. Unfortunately, things didn't work out. I spent a lot of time just getting the board to post. Turns out the only way to get it to post is with one DIMM in one specific slot. Both of the DIMMs work fine in that slot and both of them cause a failure to post in the other slot. Lowering the RAM speed has no effect. Right now, my hypothesis is that the slot is defective and that I need a new board.

I'm on a train to Stockholm now and won't be back until Sunday night, so not much more to report right now. I will rip out the 2x8GB Kingston DDR3L-1600 from my laptop when I get back and see if those work. If not, I'll be fairly certain the board is bust.

I did get Windows 10 up and running without issues, by the way (with the board in single channel mode and RAM at 1866 MHz). It was completely stable and I never saw the clock frequency fall below ~2.4 GHz, even during four core load. This was with the board just sitting on my desk. Need to do some more thorough and longer full load testing, though.

I made a run on Cinebench R15 and got 20.25 FPS, 147 MT and 50 ST. Seems the single channel memory hurts performance quite a bit in MT.

 

[hojnikb]

What were the maximum temps you saw ?

 

[DMSO]

Brunnis, maybe you need to enable the dual channel support in the BIOS ? (just a guess)

Regarding:
* Meanwhile I found time to bench my BayTrail DN2820FYK NUC (sysbench 0.4.12):

$ sysbench --num-threads=1 --test=cpu --cpu-max-prime=20000 --validate run
total time: 45.6506s
$ sysbench --num-threads=2 --test=cpu --cpu-max-prime=20000 --validate run
total time: 25.2435s

I know that these are rough CPU-only results, without using any of the CPU's features or such.
* But before I and other Linux users can replace -- in my case -- the DN2820FYK, it needs to be checked whether this bug is present: https://bugzilla.kernel.org/show_bug.cgi?id=109051.

 

[nvgpu]

https://software.intel.com/sites/de...4-ia-32-architectures-optimization-manual.pdf

14.1.1 Goldmont Microarchitecture
The Goldmont microarchitecture builds on the success of the Silvermont microarchitecture (see Section
14.1.2), and provides the following enhancements:
• An out-of-order execution engine with a 3-wide superscalar pipeline. Specifically:
— The decoder can decode 3 instructions per cycle.
— The microcode sequencer can send 3 uops per cycle for allocation into the reservation stations.
— Retirement supports a peak rate of 3 per cycle.
• Enhancement in branch prediction which de-couples the fetch pipeline from the instruction decoder.
• Larger out-of-order execution window and buffers that enable deeper out-of-order execution across
integer, FP/SIMD, and memory instruction types.
• Fully out-of-order memory execution and disambiguation. The Goldmont microarchitecture can
execute one load and one store per cycle (compared to one load or one store per cycle in the
Silvermont microarchitecture). The memory execution pipeline also includes a second level TLB
enhancement with 512 entries for 4KB pages.
• Integer execution cluster in the Goldmont microarchitecture provides three pipelines and can
execute up to three simple integer ALU operations per cycle.
• SIMD integer and floating-point instructions execute in a 128-bit wide engine. Throughput and
latency of many instructions have improved, including PSHUFB with 1-cycle throughput (versus 5
cycles for Silvermont microarchitecture) and many other SIMD instructions with doubled throughput;
see Table 14-14 for details.
• Throughput and latency of instructions for accelerating encryption/description (AES) and carry-less
multiplication (PCLMULQDQ) have been improved significantly in the Goldmont microarchitecture.
• The Goldmont microarchitecture provides new instructions with hardware accelerated secure
hashing algorithm, SHA1 and SHA256.
• The Goldmont microarchitecture also adds support for the RDSEED instruction for random number
generation meeting the NIST SP800-90C standard.
• PAUSE instruction latency is optimized to enable better power efficiency.

Goldmont microarchitecture finally documented by Intel.

 

[cbn]

I'm also waiting for brix or zotac apollo lake, but it needs to be 4 core and passive.

ASUS might also have something for you based on this fanless Vivo PC with N3700:

https://www.asus.com/Mini-PCs/VivoMini-UN45/specifications/

ASRock too:

http://www.asrock.com/microsite/beebox/

P.S. I found those PCs from the following fanless tech roundup---> http://www.fanlesstech.com/2016/10/fanless-nuc-roundup-2016.html

 

[hojnikb]

ASUS might also have something for you based on this fanless Vivo PC with N3700:

https://www.asus.com/Mini-PCs/VivoMini-UN45/specifications/

ASRock too:

http://www.asrock.com/microsite/beebox/

P.S. I found those PCs from the following fanless tech roundup---> http://www.fanlesstech.com/2016/10/fanless-nuc-roundup-2016.html

Vivo is passive cooled only with n3000 (dual core).

>>*Fanless design & 45W power adapter are only for Intel® Braswell Dual-Core Celeron N3000 Processor

 

[DMSO]

Wonder why/how the ASRock J4205-ITX doesn't support DDR4, since the Pentium J4205 clearly does. Maybe some other manufacturer will release an ITX board with this latest Pentium architecture CPU and DDR4 support.

 

[hojnikb]

Wonder why/how the ASRock J4205-ITX doesn't support DDR4, since the Pentium J4205 clearly does. Maybe some other manufacturer will release an ITX board with this latest Pentium architecture CPU and DDR4 support.

thats because j4205 only supports lpddr4, which last time i checked isnt available in dimm format.

 

[Brunnis]

Goldmont results inbound! I ripped the DIMMs out of my Dell laptop and the J4205-ITX booted right up in dual channel mode. The memory modules that work are Kingston KVR16LS11/8 and these are actually one of the few memory modules that Asrock has verified on this board. The specs of the modules are: DDR3L 1600 MHz, 8GB per module, CL11-11-11.

So, like Technikaffe.de said in their review, this board is probably quite picky when it comes to memory. I put the Corsair modules in the Dell laptop instead and they work fine there. I'm currently waiting for some Kingston 1866 MHz memory that I hope will work better in this board.

In the meantime, I've run some benchmarks. I will re-run the benchmarks once (if) I get the 1866 MHz memory working. Also, I don't have any graphics benchmarks yet, except for Cinebench. I got really low results in Fire Strike, but I also just realized that I don't have the very latest driver installed. I'll try to remedy that tonight. Now, the results so far:

Cinebench R15
OpenGL: 21.57 FPS
ST: 52
MT: 193

Dolphin Benchmark
Total: 1546 s

PassMark
CPU Mark: 2460
ST: 915
Memory: 1208
http://www.passmark.com/baselines/V9/display.php?id=71286660891

x264 Benchmark 5.0.1
Pass 1: 21.45
Pass 2: 5.39

Geekbench v3 (64-bit)
ST: 1715
MT: 5588
http://browser.primatelabs.com/geekbench3/8143500

Geekbench v4
ST: 1643
MT: 4679
https://browser.geekbench.com/v4/cpu/1038829

7-zip Benchmark (64-bit)
1968 / 7410 MIPS

I haven't tried to compare much to Bay Trail and Cherry Trail yet, but I think it looks like it's a decent step up performance-wise.

EDIT: More results, including graphics benchmarks:

3DMark Firestrike v1.1
Total: 603
Graphics: 648
Physics: 2793
http://www.3dmark.com/fs/10778133

GfxBench 4.0.13
Car Chase Offscreen: 14.8 FPS
1440p Manhattan 3.1.1 Offscreen: 13.7 FPS
1080p Manhattan 3.1 Offscreen: 23.7 FPS
1080p Manhattan Offscreen: 33.4 FPS
1080p T-Rex Offscreen: 67.5 FPS
1080p Texturing Offscreen: 5724 MTexels/s
1080p Driver Overhead 2 Offscreen: 35.9 FPS
1080p ALU 2 Offscreen: 61.5 FPS
Tessellation Offscreen: 51.8 FPS
Comparison to Pentium N3710

Fritz Chess Benchmark 12
Relative speed: 10.29
Kilo nodes/sec: 4937

POV-Ray 3.7
527 PPS
498 secs

 

[Nothingness]

z8750@2.56 GHz is 1137/3250
J2900@2.67GHz is 1258/3439, though most scores seem to be ~5% lower for ST.

Definitely a decent speed up.

 

[Sweepr]

Goldmont results inbound!

Thank you!

In the meantime, I've run some benchmarks. I will re-run the benchmarks once (if) I get the 1866 MHz memory working. Also, I don't have any graphics benchmarks yet, except for Cinebench. I got really low results in Fire Strike, but I also just realized that I don't have the very latest driver installed. I'll try to remedy that tonight.

Keep on your great work. GFXBench results tonight as well, please.

Is that Cinebench OpenGL score for CPU or iGPU?

If you don't mind, I will use your numbers for comparisons.

 

[Sweepr]

Goldmont vs Airmont vs Jaguar/Puma+ vs Core 2 Quad

Pentium J4205 - 4C/4T Goldmont @ 1.5GHz (base) / 2.6 GHz (turbo)
A8-7410 - 4C/4T Puma+ @ 2.2 GHz (base) / 2.5 GHz (turbo)*
Pentium N3710 - 4C/4C Airmont @ 1.6 GHz (base) / 2.56 GHz (turbo)*
A6-5200 - 4C/4T Jaguar @ 2.0 GHz (no turbo)
Athlon 5350 - 4C/4T Jaguar @ 2.05 GHz (no turbo)
Core 2 Quad Q6600 - 4C/4T 'Kentsfield' @ 2.4 GHz

*Picked the highest scores I could find for each @ NotebookCheck / Geekbench results are the highest Windows 64-bit scores

Cinebench R15
- ST:
Pentium J4205: 52
A8-7410: 52
Pentium N3710: 42

- MT:
Pentium J4205: 193
A8-7410: 174
Pentium N3710: 150
Core 2 Quad Q6600: 234

Dolphin Benchmark
Pentium J4205: 25,76 min

PassMark
Pentium J4205:
CPU Mark: 2460
ST: 915

A8-7410:
CPU Mark: 2620
ST: 918

Pentium N3710:
CPU Mark: 1839
ST: 552

http://www.passmark.com/baselines/V9/display.php?id=71286660891

x264 Benchmark 5.0.1
Pentium J4205:
Pass 1: 21.45
Pass 2: 5.39

Core 2 Quad Q6600:
Pass 1: 23.8
Pass 2: 5.7

http://www.legitreviews.com/upgrading-from-intel-core-2-quad-q6600-to-core-i7-4770k_2247/10

Geekbench v3

Pentium J4205:
ST: 1715
MT: 5588
http://browser.primatelabs.com/geekbench3/8143500

A8-7410 (highest score):
ST: 1379
MT: 4042

Pentium N3710 (highest score):
ST: 1077
MT: 3540

Core 2 Quad Q6600:
ST: 1306
MT: 4490

Geekbench v4

Pentium J4205:
ST: 1643
MT: 4679
https://browser.geekbench.com/v4/cpu/1038829

A8-7410 (highest score):
ST: 1360
MT: 3643

Pentium N3710 (highest score):
ST: 1207
MT: 3405

Core 2 Quad Q6600:
ST: 1543
MT: 4387

7-zip Benchmark

Pentium J4205: 1968 / 7410 MIPS

I haven't tried to compare much to Bay Trail and Cherry Trail yet, but I think it looks like it's a decent step up performance-wise.

If I'm not asking too much, run Cinebench 11.5 as well, much easier to compare to older CPUs.

 

[Brunnis]

Awesome job on the comparison, Sweepr!

Is that Cinebench OpenGL score for CPU or iGPU?

There's an Intel GPU driver installed and I just ran the test, without changing any settings, so it should be for the iGPU.

If I'm not asking too much, run Cinebench 11.5 as well, much easier to compare to older CPUs.

Sure, I already have it on the system. Just didn't have time to run it last night. The x264 benchmark took forever...

 

[Nothingness]

Geekbench v4

Pentium J4205:
ST: 1643
MT: 4679
https://browser.geekbench.com/v4/cpu/1038829

(...)

Pentium N3710 (highest score):
ST: 1214
MT: 3457

That's an Android score ;-) Though that doesn't make a lot of difference since the highest Windows 64-bit score is 1207/3405.

 

[majord]

Thanks (both) for your efforts , only requests I have would be, some classics, like Fitz chess, pov ray

 

[Brunnis]

More J4205 results! I installed the updated Intel driver and it resolved the 3DMark issue.

3DMark Firestrike v1.1
Total: 603
Graphics: 648
Physics: 2793
http://www.3dmark.com/fs/10778133

GfxBench 4.0.13
Car Chase Offscreen: 14.8 FPS
1440p Manhattan 3.1.1 Offscreen: 13.7 FPS
1080p Manhattan 3.1 Offscreen: 23.7 FPS
1080p Manhattan Offscreen: 33.4 FPS
1080p T-Rex Offscreen: 67.5 FPS
1080p Texturing Offscreen: 5724 MTexels/s
1080p Driver Overhead 2 Offscreen: 35.9 FPS
1080p ALU 2 Offscreen: 61.5 FPS
Tessellation Offscreen: 51.8 FPS
Comparison to Pentium N3710

Fritz Chess Benchmark 12
Relative speed: 10.29
Kilo nodes/sec: 4937

POV-Ray 3.7
527 PPS
498 secs

 

[Sweepr]

More J4205 results! I installed the updated Intel driver and it resolved the 3DMark issue.

Thanks again Brunnis. Cinebench 11.5 next!

Here's comes the long comparisons post. Will put all of them together in the next page.

3DMark Firestrike v1.1

Pentium J4205:
Total: 603
Graphics: 648

http://www.3dmark.com/fs/10778133

*Pentium N3700:
Total: 256
Graphics: 270

*A8-7410:
Total: 522
Graphics: 560

*Highest scores from NotebookCheck

GFXBench 4.0.13

Pentium J4205:
Car Chase Offscreen: 14.8 FPS
1440p Manhattan 3.1.1 Offscreen: 13.7 FPS
1080p Manhattan 3.1 Offscreen: 23.7 FPS
1080p Manhattan Offscreen: 33.4 FPS
1080p T-Rex Offscreen: 67.5 FPS
1080p Texturing Offscreen: 5724 MTexels/s
1080p Driver Overhead 2 Offscreen: 35.9 FPS
1080p ALU 2 Offscreen: 61.5 FPS
Tessellation Offscreen: 51.8 FPS

Pentium N3710:
Car Chase Offscreen: 9.9 FPS
1440p Manhattan 3.1.1 Offscreen: 8.7 FPS
1080p Manhattan 3.1 Offscreen: 15.3 FPS
1080p Manhattan Offscreen: 20.3 FPS
1080p T-Rex Offscreen: 41.3 FPS

Comparison to Pentium N3710

A4-7210:
Car Chase Offscreen: 11.1 FPS
1440p Manhattan 3.1.1 Offscreen: 6.8 FPS
1080p Manhattan 3.1 Offscreen: 10.5 FPS
1080p Manhattan Offscreen: 13.1 FPS
1080p T-Rex Offscreen: 46.9 FPS

*A9-9410:
Car Chase Offscreen: 15.6 FPS
1440p Manhattan 3.1.1 Offscreen: 12.1 FPS
1080p Manhattan 3.1 Offscreen: 18.4 FPS
1080p Manhattan Offscreen: 20.7 FPS
1080p T-Rex Offscreen: 70.6 FPS

*Top GFXBench scores for this part

Fritz Chess Benchmark 12

Pentium J4205:
Relative speed: 10.29
Kilo nodes/sec: 4937

Different version? That's the best I could find.

POV-Ray 3.7

Pentium J4205:
527 PPS / 498 secs

Very nice IPC gain relative to Pentium J2900!

Apollo Lake is basically 2.4x as fast as Braswell in Fire Strike Graphics, that's impressive. Relative to AMD it's faster than Carrizo-L and ahead of Stoney Ridge in most GFXBench subtests. It's actually close to HD Graphics 515 performance.

 

[DMSO]

Thanks Sweepr. I've just redone my Celeron N2820 NUC (7.5W TDP) in Geekbench v4 with the latest Arch updates and restart: https://browser.primatelabs.com/v4/cpu/1050479:
Geekbench v4:
Single-Core Score 1102
Multi-Core Score 1783
This time the "Multi-Core Score" scaled better than the last time.

I couldn't get the v3 to work because it said something like ~internet connection 500 upload error.

Sysbench 0.4.12:
$ sysbench --num-threads=1 --test=cpu --cpu-max-prime=20000 --validate run
total time: 45.6506s
$ sysbench --num-threads=2 --test=cpu --cpu-max-prime=20000 --validate run
total time: 25.2435s

For me it's 1102 ST vs 1715 ST that I care about more than MT when it comes to whether upgrading my slow NUC or not. 1715 is better but I wish it'd be something like >2000 ST. Maybe on an Arch system it would reach more than 1715 (for whatever reason). Of course the features of Apollo Lake should be considered too whether or not to upgrade my NUC. So, I'm not sure ATM.

edit:
And of course for Linux users this bug might be relevant: https://bugzilla.kernel.org/show_bug.cgi?id=109051