The ODROID H-series is back

[Raybuntu]

The ODROID H-series is back
It is more powerful, offers higher performance and comes in two brand new models.

A PDF format file is also provided to make reading this introductory document more comfortable.
https://dn.odroid.com/ODROID-H2/H3_pics ... uction.pdf


Introducing the ODROID-H3 and ODROID-H3+
Hardkernel is introducing the ODROID-H3 and H3+, which both have the same form factor and similar power efficiency as their predecessor, the ODROID-H2+.








The major characteristics of the ODROID-H3 and H3+ compared to the ODROID-H2+ are:

1. Intel 10nm Jasper Lake vs. Intel 14nm Gemini Lake processors.
2. Maximum memory (DDR4 2933MT/s) is 64GB vs. 32 GB (DDR4 2400MT/s).
3. Higher base and boost CPU frequencies and more powerful iGPU.
4. PCIe Gen 3 x4 NVMe vs. PCIe Gen 2.
5. An Unlimited Performance mode allowing the CPU to run in sustained Turbo Boost mode.

We also implemented little details following the ODROID-H2+ feedback we received from all of our users, this means you. Examples:

1. The ODROID-H3 and H3+ use a standard PC 12V PWM fan. Yes, you can use whatever 3rd party fan you prefer and plug it into the board with zero hassles.
2. The BIOS fTPM is enabled by default: Windows 11 compatibility out of the box.

Without further ado, let’s look at the detailed table shown below.



Notable Facts

Performance
While the CPU base frequency increases by 16% (H3) and 32% (H3+) compared to the H2+, the memory bandwidth increases by 22%, and the disk I/O increases by 97% compared to the H2+, one should not forget that the Jasper Lake generation also brings node intrinsic optimizations compared to the Gemini Lake generation. In addition, both the H3 and H3+ can be set to run in turbo boost mode with no time limit, a mode we call "Unlimited Performance" and describe further.
As the saying goes, the whole is greater than the sum of its parts: While running 300+ benchmarks, a great number of them being non synthetic, we witnessed a performance increase ranging in average from 45% to 73%, with maximum increases being more than twice faster.

Compatibility
The ODROID-H3 and H3+ boards use the same physical format as the H2+ board. As a consequence the H2+ blue acrylic cases and all the 3rd party cases, e.g. users 3D printed cases, are compatible with the ODROID-H3 and H3+. This also means that the ODROID H-series Net card works on the H2+, H3 and H3+ out of the box. Finally 3rd party PCIe cards (Network cards, RAID cards, Graphics cards) that one used and uses with the H2+ can also be used with the H3 and H3+.

Versatility
We have seen and still see an incredible broad range of ODROID-H2+ usage from users like you. We expect to witness the same range with ODROID-H3 and H3+. But we also expect to see users add more use cases thanks to the higher CPU performance as well as graphics and PCIe increased performance which definitely bring new opportunities. The doubling of the maximum memory from 32GB to 64GB offers new possibilities too. We believe the success of the ODROID H-series is in part due to its original DIY design goal with a board that does not restrict you to one kind of application, e.g. TV box. Let us recall the common features through all the ODROID H-series models:

Design
An SBC design that makes sense: All the connectivity is on the rear side, simplifying case design and reducing footprint on a desk.

H-series Net Card
Using the NVMe port, provides 4 additional 2.5 GbE ports, thus tripling the number of 2.5 GbE ports to 6 ports.

Do It Yourself
The ODROID H-series offers you a lot of freedom. You are free to chose:

1. The amount and brand of memory. No soldered memory.
2. The size of the eMMC (including not using one). No soldered eMMC.
3. The size of the NVMe PCIe Gen 3 x4 SSD (including not using one)(*).
4. To transform the NVMe slot into a PCIe Gen 3 x4 slot for using PCIe cards via optional adapter cable(*).
5. The size of the 1 or 2 SATA III disks or SSDs (including not using them).
6. A case among 7 types of Hardkernel cases, or use a custom one you or another user designed (**).
7. Hardkernel cases allow the usage of an optional silent 92mm fan for optimal thermal performance.
8. Any x86-64 flavor of Windows, Linux or BSD operating systems, etc. Plus Android.
9. To upgrade the hardware later with more memory, more NVMe or SSD or hard disk space.

(*) PCIe Gen 2 on the H2/H2+.
(**) The acrylic blue cases can be used with the H3/H3+ and conversely the new PCB cases can be used with the H2/H2+.


Comparing the H3 and H3+ to the H2+
In order to evaluate the performance of the H3 and H3+ and compare them to their predecessor, the H2+, we proceeded with real application benchmarks rather than synthetic ones. To do so we ran a battery of Phoronix testing suites. These testing suites are listed below:

• Compilation
• Compression
• Java
• Python
• Imaging
• Audio Encoding
• Databases
• Unigine GPU
• Cryptography
• Video Encoding

The Phoronix Testing Suite is available here: https://www.phoronix-test-suite.com/. As stated on its web site, we quote: “The Phoronix Test Suite [is an OSS project that] makes the process of carrying out automated tests incredibly simple. The Phoronix Test Suite will take care of the entire test process from dependency management to test download/installation, execution, and result aggregation.”

Let us examine the results we obtained with tables and charts showing the H2+ as base 100.
Example: Using the line Timed Eigen Compilation in the Compilation Benchmark table shown below, the H3 is 63% faster, the H3+ 83%, the H3’up 67% and the H3+’up 85%.

What are the H3’up and H3+’up?
The suffix ‘up is a shortcut notation to indicate that the CPU is running in “Unlimited Performance Mode”. This is a mode where the CPU can run in Turbo Boost mode with no time limit, hence the name. The Unlimited Performance Mode is described in the next section.

Compilation Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/compilation


Compression Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/compression


Java Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/java


Python Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/python


Imaging Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/imaging


Audio Encoding Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/audio-encoding


Databases Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/database (note: we only ran a subset)


Unigine GPU Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/unigine (This benchmark was run in 1280x720 resolution)


Cryptography Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/cryptography


Video Encoding Benchmark


Benchmark description: https://openbenchmarking.org/suite/pts/video-encoding


Summary Benchmark
To summarize, we completed 313 benchmarks on the H2+ and H3 and 259 on the H3+, H3’up and H3+’up. The Average line in the table shown below is the flat average on all the tests.
For each platform H3/H3+ system, we counted the number of tests where the optimization was more than 25% up to more than 150%.




Benchmarks Configuration
The same WD BLACK 500GB SN750 NVMe was used on all systems.
ODROID-H2+
Ubuntu 22.04, 32GB of DDR4 2400MT/s memory.
ODROID-H3, H3+, H3’up, H3+’up
Ubuntu 22.04, 64GB of DDR4 2933MT/s memory.



PCIe Gen 3 vs. PCIe Gen 2
To illustrate the quasi double performance of the PCIe bus on the ODROID-H3 and H3+ compared to the ODROID-H2+, we ran iozone3 tests with relatively big data blocks:
iozone -e -I -a -s 100M -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2

The two tables shown below list the relative I/O acceleration on the H3 and H3+.
One can see that the bigger the data block the closer we get to twice the speed, within margin of error, plus a few % points because the CPUs also run faster.



The 3 tables shown below list the benchmark transfer speed values (in bytes):





Introducing the Unlimited Performance Mode

Starting with the Core 10th generation INTEL introduced Power Limit 4 (PL4) and made it user configurable via the BIOS. What is it? PL4 is the SoC's maximum power limit at the package level. No matter what the CPU is actually doing, it will not pass this limit. The interesting side of the story is that as a user you can set it to 0, which means no limit.

The ODROID-H3 and H3+ BIOS allows you to set this limit to 0. This is what we call Unlimited Performance mode. The default value is 30,000 corresponding to the Balanced mode, meaning around a SoC's maximum power limit of 10W.

Using the Unlimited Performance mode (annotated 'up) with the H3 and H3+ enables the CPU to turbo boost indefinitely: 2.6 GHz all cores and 2.9 GHz one core for the H3 , 2.8 GHz all cores and 3.3 GHz one core for the H3. This results in a significant increase in performance, especially for the H3+. You can see the effect in the benchmarks we included above. In a few cases the H3’up can reach and pass double performance for the same task compared to the H2+.

As you may expect the CPU will get hot quickly (in a matter of minutes) and get close to its T Junction (Tj) temperature which will trigger the emergency shutdown as thermal protection.
But the CPU will not reach Tj because it will automatically throttle down when it is about 5 degrees Celsius away from Tj (we tested this multiple times).
As soon as the CPU thermally throttles down you start losing the increased performance you were aiming at while still consuming more power compared to the Balanced mode. Not ideal.

In order to prevent thermal throttling when using the Unlimited Performance mode, the solution is simple: Active cooling with a fan.

We designed the H3/H3+ heat sink to make it very efficient: (a) You do not need a fan in Balanced mode and (b) it has a high rate of thermal exchange when coupled with a fan.

Using a fan will decrease the maximal CPU temperature by about 25 to 30 degrees Celsius depending on factors such as the ambient temperature. It is difficult for us to publish precise temperature values because what one witnesses depends on many factors: As already mentioned the ambient temperature, the CPU BGA soldering thickness error, the heat sink assembly tolerance, the type of thermal paste and quantity applied, and the cooling fan speed RPM error margin (which can be as high as 5 to 10%). All of these factors can result in a 10+ degrees Celsius difference between one setting and another.

The important point is that with active cooling you get the increased performance you aim at while the CPU stays just comfortably warm while turbo boosting indefinitely, way below temperatures close to Tj. In other words the fan active cooling brings you the best of both worlds. This is what we witnessed and validated while performing many tests in different locations.

Last point: In Unlimited Performance mode, the CPU (and the fan) use more power than they do in Balanced mode, easily reaching 20+ Watts. However this happens only when the CPU is indeed turbo boosting. When idle, the system will use the same power as in Balanced mode. If your goal is to minimize energy consumption, use Balanced mode. If your goal is to maximize performance, use Unlimited Performance mode and again use active cooling with a fan to avoid the CPU to constantly throttle down.

For learning how to change PL4 in the BIOS, as well as change the fan settings, please refer to the related Wiki page.

The official 92x92x25mm 12V PWM cooling fan or a similar 3rd party cooling fan should be mounted on the official case's venting hole to avoid thermal throttling of the CPU in Unlimited Performance mode. We have tested the following 3rd party cooling fan samples.

• Noctua NF-A9 PWM
• Noctua NF-A19x4 PWM
• Noctua NF-B9 REDUX PWM
• Thermalright TL-9015W

Here is the ODROID Stock Fan and a 3rd Party Fan (a Noctua):







Demo video

To test the CPU computing power and GPU rendering speed of the ODROID-H3/H3+, we tested running Wii U and PS2 games emulation.
This emulation is still very difficult to run on a low power SBC. The games were extremely slow and very far from a playable level on the ODROID-H2+.
On the other hand, during our testing, we enjoyed playing PS2 games on the ODROID-H3+ running in Unlimited Performance mode
https://www.youtube.com/watch?v=JSbaRCdkgeQ





Dual Head 4K Monitor Demo

We can connect two 4K/60Hz monitors to the H3 for both fun and productive work. Thanks to hardware virtualization, Linux and Windows can be operated at the same time. Note also that with a maximum memory of 64GB running guest OSes is no problem.


Picture 1 : Two different 4K YouTube videos play back flawlessly and simultaneously with Chrome browser on Ubuntu desktop.


Picture 2 : The monitor on the left shows Ubuntu Desktop host OS and the one on the right shows Windows 10 running as a guest OS, using the hardware virtualization VT-x technology.


Picture 3 : The monitor on the left shows the PCB designing KiCAD application running in Ubuntu while the monitor on the right shows the Edge browser in Windows running as a guest OS.



New Cases
The blue acrylic cases we have made so far had the advantage of being translucid, enabling you to see the inside of the system. However we witnessed too many occurrences of broken or cracked panels due to external impact during delivery.
Accordingly, we developed new cases using a relatively thin and rigid PCB made of glass fiber epoxy to prevent these issues. We have seven types of new cases as listed in the table shown below:




Case Type 1
No Net card and space for 2 x 3.5” SATA drives.




Case Type 2
No Net card, no space for SATA storage.




Case Type 3
No Net card and space for 2 x 2.5” SATA drives (max. 15mm thick).




Case Type 4
No Net card and space for 1 x 3.5” SATA drive or 2 x 2.5” SATA drives.




Case Type 5
Space for Net card and 2 x 3.5” SATA drives.




Case Type 6
Space for Net card, no space for SATA storage.




Case Type 7
Space for Net card and 2 x 2.5” SATA drives (max. 15mm thick).





Hardware Details

Board Description

1. CPU (Intel Celeron N5105(H3+: N6005) )
2. 2 x DDR4 SO-DIMM slots (Dual channel memory support)
3. 1 x M.2 PCI Express Module Socket (NGFF-2280)
4. 1 x eMMC (Embedded Multimedia-Card) Socket
5. 2 x SATA Power Connectors (2.5mm pitch, JST-XH compatible connector)
6. 2 x SATA3 6.0 Gb/s Data Connectors
7. 1 x DC Power Jack
8. 2 x USB 3.0
9. 2 x USB 2.0
10. 1 x HDMI 2.0
11. 1 x DisplayPort 1.2
12. 2 x RJ45 Ethernet Ports (10/100/1000/2500)
13. 5 x System LED Indicators
14. 1 x Peripheral Expansion Header (24-pin)
15. 1 x Power Switch
16. 1 x Reset Switch
17. 1 x Backup Battery Connector (2-pin)
18. 1 x Active Cooling Fan Connector (4-pin)
19. 1 x Audio out, 1 x Audio in, 1 x SPDIF out

On the H3+ board, a yellow round sticker is attached on the product serial label, and on the H3 board, a green round sticker is attached.


Block diagram




Specifications




Fan Connector
The ODROID-H3 and H3+ use a PC standard 12V PWM 4-pin connector instead of the proprietary 5V mini connector used on the H2/H2+.
Therefore, anyone can install a third-party cooling fan that can be easily purchased in the market. A 92mm size of cooling fan is recommended for the H3 cases compatibility.




Power consumption
We used our SmartPower3 (see https://www.hardkernel.com/shop/smartpower-iii/) to test and measure the ODROID-H3 power consumption while performing specific activities. We used an M.2 NVMe storage device, 4K HDMI monitor, Ethernet cable and USB combo keyboard + mouse while measuring the power consumption. The table shown below and its corresponding chart detail the power consumption we witnessed:





Notes

• In headless mode, the idle state power consumption should be lower than 2 Watt.
• If one runs the H3+ in Unlimited Performance mode and stresses the CPU to its maximum, the system peak power consumption can be near 22 Watt.

More detail technical information are available on the WiKi pages:
https://wiki.odroid.com/odroid-h3/start



The Making of the ODROID-H3 and ODROID-H3+

Until the introduction of the ODROID-H2, we were well-known for our ARM-based SBC such as the XU4, C-series, N-series and more recently the M1.

However the x86 platform brings many advantages:

• Most GPU and VPU hardware acceleration drivers are working perfectly fine on the latest Linux Kernel releases and GNU software out of the box.
• Large scale memory expandability, up to 64GB of RAM.
• We can have stable and more hardware connectivity: two 2.5GbE ports, two SATA III ports, four lanes of PCIe Gen 3, etc.
• Hardware virtualization VT-x powered virtual machines allow a user to run different operating systems, test applications, and experiment with specific features without worrying about system crashes related to specific hardware differences.
• On the ARM platform, it takes non-trivial human and time resources to port the mainline kernel to a level where the GPU and VPU hardware acceleration functions work properly. In contrast, on the x86 platform, you can usually run the latest operating system with full hardware acceleration.
• As a journalist wrote about the ODROID-H2+, we quote: “x86 tends to do very well with legacy support. We cannot predict the future, but there is a good chance that 5-7 years from now, the hardware we have on the ODROID-H2+ will still be well supported and OSes will be installed out-of-the-box. That is not necessarily the same for the ARM maker board ecosystem to date.”(*)

(*) Quoted from https://www.servethehome.com/odroid-h2- ... -6x-2-5gbe

Obviously, the ARM platform has numerous benefits of its own and a large number of applications in many markets: usually low cost, very low power allowing usage in millions of devices running on battery (e.g. smartphones, remote controls, sensors, tiny robots, etc) as well as 24x7 appliances where energy consumption must be as low as possible (e.g. TV boxes, home NAS, gateways, home routers, etc). Finally ARM-based SBCs enable DIY users like you or industrial users to design custom boxes, servers, instrumentation devices with sensors, etc. This is why Hardkernel makes a lot of ARM-based SBC.

Our first endeavor into the x86 world with the ODROID-H2 and H2+ was both an engineering and commercial success. This was reflected in multiple articles at https://www.cnx-software.com/ as well as at STH (see link cited above).
Due to the continuous shortage of semiconductors during the COVID-19 pandemic, we had no choice but to discontinue the popular ODROID-H2+ at the end of last year.
However, there has been continued customer demand for the x86 platform. So, earlier this year, we started developing new ODROID-H series models which leverage 11th-gen Intel processors made with an advanced 10nm (Intel 7) semiconductor fab and other key components that we can purchase much more reliably.

Many of our B2B customers have requests for long-term stable supply.
The previous generation processor J4115 package size was 25x24mm with 1090 BGA pins while the new N5150 and N6005 package size is 35mmx24mm with 1338 pins. Due to the big difference of the SoC packages, we had a very hard time keeping the same PCB form factor and connectors' positions while we designed the hardware. We have achieved backward compatibility through this difficult design process and efforts.
Thanks to this mechanical and electrical compatibility, H2 series cases and the ODROID H-series Net card can be used on the H3/H3+ as well. Since Intel doesn't guarantee the period of availability, we can't say about the longevity of the H3 series. Perhaps we can only supply the H3 series for 2~3 years in the worst case.
However, we will always do our best to maintain mechanical and electrical compatibility when Intel's next-generation low-power processors are released. We believe that this method enables long-term supply in other ways.

So here we are today with the ODROID-H3 and ODROID-H3+!



Availability

Orders for the H3 and H3+ boards with wide range peripheral devices are available now with shipping from early next week.

ODROID-H3 : https://www.hardkernel.com/shop/odroid-h3/
ODROID-H3+ : https://www.hardkernel.com/shop/odroid-h3-plus/

60W PSU : https://www.hardkernel.com/shop/15v-4a- ... y-us-plug/
133W PSU: https://www.hardkernel.com/shop/19v-7a-power-supply/
DDR4 SO-DIMM 4GB: https://www.hardkernel.com/shop/samsung ... 0-so-dimm/
DDR4 SO-DIMM 8GB: https://www.hardkernel.com/shop/samsung ... 0-so-dimm/
DDR4 SO-DIMM 16GB: https://www.hardkernel.com/shop/samsung ... 0-so-dimm/
DDR4 SO-DIMM 32GB: https://www.hardkernel.com/shop/samsung ... 0-so-dimm/
92mm 12V cooling fan : https://www.hardkernel.com/shop/92x92x2 ... r-tacho-2/
2.5GbE 4-ports NetCard : https://www.hardkernel.com/shop/h2-net-card/
WiFi + Bluetooth module :https://www.hardkernel.com/shop/wifi-module-5bk/
SATA + Power cables : https://www.hardkernel.com/shop/sata-da ... wer-cable/
BIOS/RTC backup battery : https://www.hardkernel.com/shop/rtc-backup-battery/ (One battery is included in the package as a free bundle. You may want to buy one as a spare.)
Case Type 1 : https://www.hardkernel.com/shop/odroid-h3-case-type-1/
Case Type 2 : https://www.hardkernel.com/shop/odroid-h3-case-type-2/
Case Type 3 : https://www.hardkernel.com/shop/odroid-h3-case-type-3/
Case Type 4 : https://www.hardkernel.com/shop/odroid-h3-case-type-4/
Case Type 5 : https://www.hardkernel.com/shop/odroid-h3-case-type-5/
Case Type 6 : https://www.hardkernel.com/shop/odroid-h3-case-type-6/
Case Type 7 : https://www.hardkernel.com/shop/odroid-h3-case-type-7/
LED Power button : https://www.hardkernel.com/shop/led-power-button/
VESA Mount Kit : https://www.hardkernel.com/shop/vesa-mount-kit/
20x4 Character LCD module : https://www.hardkernel.com/shop/i2c-lcd-module/
SmartPower3 : https://www.hardkernel.com/shop/smartpower-iii/
8GB eMMC : https://www.hardkernel.com/shop/8gb-emmc-module-h2/
16GB eMMC : https://www.hardkernel.com/shop/16gb-emmc-module-h2/
32GB eMMC : https://www.hardkernel.com/shop/32gb-emmc-module-h2/
64GB eMMC : https://www.hardkernel.com/shop/64gb-emmc-module-h2/
128GB eMMC : https://www.hardkernel.com/shop/128gb-emmc-module-h2-2/

[odroid]

The H3 and H3+ are very similar to the H2 and H2+.
There are many compatible items, so we will continue to use the existing H2 sub-forum.

I've just added "H3/H3+" to the sub-forum title.
viewforum.php?f=167

[odroid]

Reserved #2

Update:
We've tested the following discrete video cards and they worked fine.
We used a M.2 NVMe Converter To PCIE x16 graphics card adapter with an extra PC PSU for the test. Although there are only four PCIe 3.0 lanes actually connected.

AMD RX 6400 : XFX Speedster SWFT105 RDNA2 4GB GDDR6
AMD RX 6500 XT : GIGABYTE RADEON RDNA2 4GB GDDR6
AMD RX 590 : SAPPHIRE NITRO+ Radeon RX 590 8GB GDDR5
NVIDIA GT 710 : Asus GT710-4H-SL-2GD5 2GB GDDR5 (4 ports HDMI)

[odroid]

Reserved #3

[Raybuntu]

Awesome news. I have ordered the H3+ as a new buildserver for CoreELEC builds.

Have been waiting for this!
Thanks

[rooted]

Very nice update. I know a lot of people will certainly be happy to see this.

I'm surprised at the + performance, it's roughly double in most places versus the H2+.

[domih]

[mad_ady]

Congratulations on the new product! The best thing about it is compatibility with previous cards/cases!

[mctom]

Hey, that's faster than my desktop!

Congratulations to everyone in HK, everything about this product must have been an enormous challenge - not only a backward-compatible HW design, but component procurement, certification, testing...
I'm sure this will pay off faster than you think!

[joerg]

Also from my side, congratulations for great product!

[tmihai20]

Wow, I did not expect that and the prices are very competitive. I did not expect a successor the H2+ with an Intel chipset.

[hominoid]

Nice approach to extending the H-Series, I like the backward compatibility. This should serve the user base very well and cover a lot of uses. Excellent and thorough product write up too.

[fvolk]

Does the H3 series have the 5V 2.5" disk power problem like the H2 - or not? viewtopic.php?p=268006#p268006

Did you try a complex/longer load on the H3 without a fan? The H2 is/was thermally limited without a fan: viewtopic.php?p=248543#p248543

Can we have a "cat /proc/cpuinfo" output?

Edit: What is the power use with: headless idle (so only power and network attached) and 2x 2.5" disks?

(My H2 as NAS increasingly showed its limits, I've planned to upgrade to something stronger, now I'm really torn to stick to my plan or be again one of the first customers of a H3, just like I got a H2 from the first batch...)

[PigLover]

Absolutely fabulous news! I love my H2+. An absolute workhorse in my home lab setup, my go-to box for stable operation. I was very sad when Intel's allocation system cut you off - hopefully you'll get treated better this time around.

[tkaiser]

I'm surprised at the + performance, it's roughly double in most places versus the H2+.

And I'm surprised that people believe this. Look at the comparison chart. A N5105 has a 16% higher CPU clock and 22% higher clocked memory. The N6005 is at a 32% higher CPU clock and again 22% higher clocked memory. How should this result in 'roughly double' the CPU performance? That's simply impossible if those Jasper Lake CPU cores aren't completely different beasts than Gemini Lake Refresh (which they aren't).

https://www.cpubenchmark.net/compare/45 ... eron-J4115

[tkaiser]

Can we have a "cat /proc/cpuinfo" output?

N6005: http://ix.io/4aUc
N5105: http://ix.io/4aSR

[odroid]

Does the H3 series have the 5V 2.5" disk power problem like the H2 - or not? viewtopic.php?p=268006#p268006

We've tested only 12V 3.5" HDDs. A few different high-power required 7200rpm 3.5" disks worked fine when we used the 19V/7A PSU.
We've not test any 2.5" HDDs.

Did you try a complex/longer load on the H3 without a fan? The H2 is/was thermally limited without a fan: viewtopic.php?p=248543#p248543

If we ran stress-ng --cpu 4 --cpu-method matrixprod command for 10 minutes on the H3 without cooling fan, we could see 2.4~2.8Ghz swing instead of sustained 2.9Ghz turbo-boost at 26°C ambient temperature.
The throttled frequency level is ~20% higher than the H2 since the new H3 CPU is made with 10nm fab instead of 14nm.
But, the frequency throttling does still exist if you keep running a full-core heavy load over several minutes without a cooling fan.

Can we have a "cat /proc/cpuinfo" output?

tkaier has posted the outputs even I don't know where he could get it though. Probably, from his sbc-bench log server?

Edit: What is the power use with: headless idle (so only power and network attached) and 2x 2.5" disks?

Likely 2.0 ~ 2.5Watt if your HDDs can be in spin-down mode.
Actually measured values in various states were available in "Power Consumption" section on the OP.

(My H2 as NAS increasingly showed its limits, I've planned to upgrade to something stronger, now I'm really torn to stick to my plan or be again one of the first customers of a H3, just like I got a H2 from the first batch...)

If you still prefer using two small 2.5" HDDs to build a extremely low idle power consuming NAS, I think a low-power ARM platform should be considered.

[odroid]

And I'm surprised that people believe this. Look at the comparison chart. A N5105 has a 16% higher CPU clock and 22% higher clocked memory. The N6005 is at a 32% higher CPU clock and again 22% higher clocked memory. How should this result in 'roughly double' the CPU performance? That's simply impossible if those Jasper Lake CPU cores aren't completely different beasts than Gemini Lake Refresh (which they aren't).

Welcome back. Long time no see.

I think your calculations and assumptions are reasonable since the micro-architecture differences between Goldmont and Tremont are near zero.
But many of Multi-threaded benchmark scores are about 1.4~1.8 times actually because the H2/H2+ had no feature to unlock the Power Limit 4 (we call it "Unlimited Performance mode").
Thanks to the major difference, the sustained max clock frequency of the H2+ was around 2.3Ghz when ran a multi-thread load while the H3+ could run at sustained 3.3Ghz turbo boost frequency if the Unlimited Performance mode was enabled and a cooling fan equipped.
Therefore, the H3+ could get 40~45% more computing power than the H2+ and additionally 22% faster DRAM might increase the benchmark scores here and there obviously.
So I think some people can say it "roughly double".

Because of our non-synthetic benchmark tests, the comparison graphs should be fair enough.
We ran the same Ubuntu OS image, the same kernel version, the same governor, the same ASPM settings and the same gcc/gblic version on H2+, H3 and H3+.

BTW, we couldn't implement the Unlimited Performance mode on the H2 series due to a hardware limitation in the core PMIC circuitry.

[tkaiser]

Therefore, the H3+ could get 40~45% more computing power than the H2+ and additionally 22% faster DRAM might increase the benchmark scores here and there obviously.
So I think some people can say it "roughly double".

I guess it's more about the relationship between benchmarks and real-world use cases. Unless you want to use H3/H3+ as number cruncher running 24/7 at full load the performance difference between Gemini Lake and Jasper Lake will not be that high. Performance is higher of course, it might be even noticeable (at least between H2/H2+ and H3+ in up mode -- every performance difference below 20% is hard to 'feel') but honestly H2/H2+ were already fast enough for what they were used and for the average user the performance per watt being now even higher might be more important?

The benchmark comparison charts represent a 'use case' that's probably not existing: running full load over hours since only then H2/H2+ get throttled down so the relative performance increase could be drawn in such a dramatic way in those charts.

As for /proc/cpuinfo: that's the same with every member of a specific Intel family anyway since those CPUs come all from the same wafers and only binning at the factory creates the different SKUs. Those with more working CPU cores and Execution Units clocking higher at lower supply voltages become the more expensive variants.

[istanbulls]

I guess it's more about the relationship between benchmarks and real-world use cases. Unless you want to use H3/H3+ as number cruncher running 24/7 at full load the performance difference between Gemini Lake and Jasper Lake will not be that high. Performance is higher of course, it might be even noticeable (at least between H2/H2+ and H3+ in up mode -- every performance difference below 20% is hard to 'feel') but honestly H2/H2+ were already fast enough for what they were used and for the average user the performance per watt being now even higher might be more important?
The benchmark comparison charts represent a 'use case' that's probably not existing: running full load over hours since only then H2/H2+ get throttled down so the relative performance increase could be drawn in such a dramatic way in those charts.

I guess your real-world use cases seem to be different from mine.
I can assume your generic use cases might be NAS, firewall, intelligent switch/router, web servers and so on.
Probably, those headless applications might not make me feel a noticeable difference even there is triple times faster computing power inside.

For my real-world use cases

<1> When I built an Android 11 BSP-SDK for M1 on my H2+, it took slightly over 8 hours.
On my H3+ in 'up mode, I can build it in 5 hours now.
I noticeably felt that difference of 3 hours was quite big for me.
I also saved a lot of electricity energy obviously thanks to the much faster compiling and linking.
Note that, the idle mode power consumption are not quite different between Performance Mode and Balanced Mode as you may know well.
At the very least, if the build time of small Arduino project files is reduced by 30-40%, you can feel that the development environment becomes much more comfortable.

<2> When I tried to play some Wii U and PS2 retro games on my H2+, the games were extremely slow and very far from a playable level.
But I can play those games on my H3+ in 'up mode at quite playable framerate now.
However, I had to spend some electricity while playing games for a few hours.
Anyway, I don't think the H2+ was fast enough for those games.

<3> When I designed a relatively complex 4-layers PCBs on my H2+, it took nearly 15 seconds to run an advanced DRC test add-on program, but now it's done in less than 8 seconds.
I definitely feel that it helps a lot in improving my productivity.

Of course, our company wants to keep high productivity by designing a 10-layers PCB and building Android on a 1000-watt Threadripper 64-core powered workstation system at work.
But personally, for my home hobby, I need to find a sweet point between power consumption and productivity.
Also, I think that running a 300 watt gaming PC to play retro games is obviously a waste of energy.

This isn't a defense/attack against your opinion, but rather an attempt to acknowledge that there are a lot more different use cases in the real-world than you & we guess.

Please open a new topic/thread on the H3 sub-forum if you need more discussion.
viewforum.php?f=168

[mctom]

I was just about to say that 100% CPU use for extended periods is not that unusual. Some terribly coded web pages even cause that.

There is however one very important benchmark missing, and that is of course a framerate of SuperTuxKart

[fvolk]

We've not test any 2.5" HDDs.

That is... unfortunate.
I run a H2 as NAS with 2.5" disks in case 3: viewtopic.php?p=336777#p336777
Note that I have removed the side panels for better cooling (my wish for a next-gen vertical design was not granted...)
It's been running for >20000h now - I don't dare to shut it down and power disconnect it when I'm away, because then every power reconnect jolts the disks?
So this is not solved for H3...

But, the frequency throttling does still exist if you keep running a full-core heavy load over several minutes without a cooling fan.

I see.
My H2 NAS runs 24/7 (because see above), I limited all cores to 2GHz to keep a safety margin to enforced thermal throttling.
I run overnight some builds/CI/webscraping etc.
Unfortunately, these take longer and longer, so the H2 is no longer sufficient.
H3 will be about the same, only a bit faster.

Edit: What is the power use with: headless idle (so only power and network attached) and 2x 2.5" disks?

Likely 2.0 ~ 2.5Watt if your HDDs can be in spin-down mode.
Actually measured values in various states were available in "Power Consumption" section on the OP.

With the H2 I have ~1W suspend to ram; ~2.5W H2 on and idle but no disks; ~3.1W H2 with 2x 2.5" disks that are spun down; ~3.8W disks running but heads parked; ~5.3W disks heads active on platter
It appears the H3 can go a bit lower?
So for a pure NAS 24/7 the H3 is still a good choice.
This is an argument for an easy H2->H3 swap... Hmmm.
(Energy is expected to be ~0.50€/kWh for quite some time :-/... )

If you still prefer using two small 2.5" HDDs to build a extremely low idle power consuming NAS, I think a low-power ARM platform should be considered.

For a pure 2x 2.5" NAS on ARM, the Odroid HC4 is too clunky IMHO.


The internet also wonders if the H3 can use M2->SATA expander cards:
https://old.reddit.com/r/selfhosted/com ... r/is2a6ow/
https://old.reddit.com/r/selfhosted/com ... r/is0jtmp/
Would one then be able to power multiple 2.5" drives from the new H3 stronger power supply with some Y-cables?

[tkaiser]

The internet also wonders if the H3 can use M2->SATA expander cards:

The Internet is asking about PCIe attached SATA host controllers. Both ASM1166 (6 SATA ports) and JMB585 (5 SATA ports) are Gen3 x2 and will of course work.

There are also SATA port multipliers like JMB575 available that turn one SATA port into many ofc sharing both latency and bandwidth of the single SATA port they're attached too. Both are entirely different things even if they could both be called 'M2->SATA expander cards'

In ODROID-M1 forum an unfortunate user already showed up having slapped a M.2 SATA port multiplier into the M1's M.2 slot which of course resulted in zero functionality.

Whether SATA port multipliers also work with Jasper Lake is something Odroid could tell us (connected to a SATA port then ofc). At least they tested this with two JMB575 port multipliers on H2 back then. And should have these PM still lying around somewhere in a drawer.

[powerful owl]

Since Intel doesn't guarantee the period of availability, we can't say about the longevity of the H3 series. Perhaps we can only supply the H3 series for 2~3 years in the worst case.

Congratulations, it looks like a real nice update. What will be the longevity of BIOS updates for this version?

[odroid]

Whether SATA port multipliers also work with Jasper Lake is something Odroid could tell us (connected to a SATA port then ofc). At least they tested this with two JMB575 port multipliers on H2 back then. And should have these PM still lying around somewhere in a drawer.

We briefly tested a JMB575 based SATA port multiplier board. It worked fine with H2 as well as H3. But we didn't take pictures of the H3 & PM and couldn't add any picture to this WiKi page due to our laziness.
https://wiki.odroid.com/odroid-h3/appli ... ata_drives

We will test a couple of different PCIe-to-SATA bridge boards in M.2 form factor in this month.

[odroid]

Since Intel doesn't guarantee the period of availability, we can't say about the longevity of the H3 series. Perhaps we can only supply the H3 series for 2~3 years in the worst case.

Congratulations, it looks like a real nice update. What will be the longevity of BIOS updates for this version?

We've released a minor BIOS update yesterday. https://wiki.odroid.com/odroid-h3/hardw ... os_release
Anyway, I think we can keep updating the BIOS two more years based on two years of contract with the BIOS vendor.

[odroid]

That is... unfortunate.
I run a H2 as NAS with 2.5" disks in case 3: viewtopic.php?p=336777#p336777
Note that I have removed the side panels for better cooling (my wish for a next-gen vertical design was not granted...)
It's been running for >20000h now - I don't dare to shut it down and power disconnect it when I'm away, because then every power reconnect jolts the disks?
So this is not solved for H3...

Probably right if you don't use 3.5" HDDs because 2.5" HDD power input shares with the always-on 5V rails.

I see.
My H2 NAS runs 24/7 (because see above), I limited all cores to 2GHz to keep a safety margin to enforced thermal throttling.
I run overnight some builds/CI/webscraping etc.
Unfortunately, these take longer and longer, so the H2 is no longer sufficient.
H3 will be about the same, only a bit faster.

If you don't use a cooling fan with "Unlimited Performance Mode", your estimation might be correct.
You might need to wait a few more years until a very decent 3nm or 5nm fabricated new Intel SoC appears which can run at steady 3.3Ghz without any active cooler.

With the H2 I have ~1W suspend to ram; ~2.5W H2 on and idle but no disks; ~3.1W H2 with 2x 2.5" disks that are spun down; ~3.8W disks running but heads parked; ~5.3W disks heads active on platter
It appears the H3 can go a bit lower?
So for a pure NAS 24/7 the H3 is still a good choice.
This is an argument for an easy H2->H3 swap... Hmmm.
(Energy is expected to be ~0.50€/kWh for quite some time :-/... )

We measured the values from a DC power supply via SmartPower3 as we informed.
If you measured the values from AC power sources, consider the efficiency of your AC-DC converters in power bricks.
You might need to add 0.5 to 1.5 Watt to our values for fair comparison.
BTW, is it time to switch to SSDs more aggressively instead of using HDDs to save electricity energy?

[fvolk]

The internet also wonders if the H3 can use M2->SATA expander cards:

[...]
Both are entirely different things even if they could both be called 'M2->SATA expander cards'

Ah, of course, I remember again!
I propose to document these two different types in the Wiki, and also collect a list of "known-good" cards.
...this will help the newcomers to the H3 universe and save on support time...

[Atlantis95]

Good day,

I have 2 questions:

1. Is it correct that that H3+ is able to run 64GB (2 x 32GB) or is this not correct? Following the Odroid-H3+ product page the support for memory is max 64GB (I read :"Two SO-DIMM slots, up to 32GB per slot"), but following the cpu n6005-processor]product page of Intel the n6005 cpu the support for memory is max 16GB (I read: "Max Memory Size (dependent on memory type) 16 GB").
2. Are the new cases to be expected for the H2. I like the case of the M1. I personally find the H2 cases a little bit dodgy. Can we expect a case for the H3 as there is a case for the M1??

Kind regards, Atlantis

[odroid]

@Atlantis95,

1. Intel documentation have had some wrong information from time to time.
The actual maximum memory address space is 32GB on the Gemini Lake platform (J4105/J4115 on H2/H2+) while their document says it is 8GB.
The actual maximum memory address space is 64GB on the Jasper Lake platform (N5105/N6005 on H3/H3+) while their document says it is 16GB.
My personal opinion is that Intel has a (bad?) habit of deliberately understating the specs to prevent their low-power processors from impacting the mainstream processors market.

2. We have no plan to make/develop other new cases for the H2 and H3 series at this moment.

[venkatbo]

Congrats HK team, you've done it again
If possible, in the future, pl explore using the Intel i225 NICs. thanks.

[tkaiser]

Great to hear SATA port multipliers are still supported with Jasper Lake Intel SoCs!

We will test a couple of different PCIe-to-SATA bridge boards in M.2 form factor in this month.

Though no idea why you would do this (H3/H3+ are neither ARM nor RISC-V, it's PCIe, the reasonable options are all Gen3 x2, all of them will work flawlessly with all major OS) but if you do this and document the outcome please refer to those cards as SATA host controllers or host bus adapters. A bridge is usually something else (e.g. translating between one storage protocol like USB mass storage or UAS and a different storage protocol like SATA or NVMe).

Congrats HK team, you've done it again
If possible, in the future, pl explore using the Intel i225 NICs. thanks.

Those buggy NICs that had severe problems implementing this https://de.wikipedia.org/wiki/Interpacket_Gap correctly?

You are aware that most of this RealTek bashing is just repeating the same old BS over and over again? See for example https://www.cnx-software.com/2022/10/12 ... ent-599057

[odroid]

We will test a couple of different PCIe-to-SATA bridge boards in M.2 form factor in this month.

Though no idea why you would do this (H3/H3+ are neither ARM nor RISC-V, it's PCIe, the reasonable options are all Gen3 x2, all of them will work flawlessly with all major OS) but if you do this and document the outcome please refer to those cards as SATA host controllers or host bus adapters. A bridge is usually something else (e.g. translating between one storage protocol like USB mass storage or UAS and a different storage protocol like SATA or NVMe).

Since you can't just connect a SATA device directly to a PCIe bus, so the word bridge isn't exactly a misnomer.
But I know well that bridge might cause some confusions and misunderstaings. Perhaps a more accurate expression would be PCIe to SATA host controller.
However, Asmedia and Jmicron, which are the most market dominant controller vendors, have been using the word bridge in their datasheet and marketing materials for over 10 years.
They probably might copy the inaccurate word each other.

Anyway, it seem to be a too much off-topic stuff.
I will create a new topic thread if you need further discussion.

[back2future]

...
If you measured the values from AC power sources, consider the efficiency of your AC-DC converters in power bricks.
You might need to add 0.5 to 1.5 Watt to our values for fair comparison.
BTW, is it time to switch to SSDs more aggressively instead of using HDDs to save electricity energy?

[ energy consumption comparing HDD and SSD (eMMC) is dependent on storage capacity for each device (6.5W/10TB_HDD and 6W/5TB_SSD, investment/service/maintenance cost aside), data transfer bandwidth GB/s (dependent on data type, data processing, file system, &c.) and storage support system (e.g. SoC/xPUs, mainboard, network, hardware accelerators, power supplies, etc.)

some recommendation seen for most efficient data storage was 'cloud storage' (about 8kWh/(TB*y)¹ for US_~2020 8TWh/1000MioTB, predictive to ~'1/10'_203x, no closer definitions for data availability/safety/redundancy) compared to guesses for individual local systems on some 10-35kWh/(TB*y)

page 13, figure 7 gives an idea for software related opportunities (SaaS, compared to IaaS storage clouds) about difference between local common server structures and optimized cloud data centers for the US, ~2013.

1) for US data center optimized situation (2020 forecast numbers from y2016 report): HDDs (~650MioTB) ~7.2kWh/TBy, SSDs (~310MioTB) ~11.6kWh/TBy, mainly because of storage capacity differences per each device for HDD and SSD, only storage(~12%), not included infrastructure(~54%)&network(~2.5%)&servers(~32%)
2) WD black 7200rpm 10TB ~4-7.4kWh/TBy ]

[Rafaello7]

How is it possible that RAM capacity is 64GB, since processors specification says that RAM capacity (of both N5105 and N6005) is 16GB?

[tkaiser]

How is it possible that RAM capacity is 64GB, since processors specification says that RAM capacity (of both N5105 and N6005) is 16GB?

Explained just few posts above: viewtopic.php?p=357066#p357066

Adding to that: ODROID-H2 users experienced problems with certain DIMM modules of larger size with low CAS latency settings so Intel might play here 'better safe than sorry'. This even resulted in a list of compatible and incompatible DIMM modules which I always used as a general source of information for all Gemini Lake (refresh) thingies out there even though different UEFI versions/fixes might make an important difference: https://wiki.odroid.com/odroid-h2/hardw ... am_modules

Today only a 'confirmed to work' list is available for Jasper Lake: https://wiki.odroid.com/odroid-h3/hardw ... am_modules

[mctom]

My personal opinion is that Intel has a (bad?) habit of deliberately understating the specs to prevent their low-power processors from impacting the mainstream processors market.

those CPUs come all from the same wafers and only binning at the factory creates the different SKUs. Those with more working CPU cores and Execution Units clocking higher at lower supply voltages become the more expensive variants.

That could be a viable explanation - perhaps larger memories generally work but are not guaranteed, or even tested in production.
On the other hand, what's there to test really - just a few more address lanes? That's just my far fetched speculation.

I work on a Dell terminal with Celeron J4105 (just like H2), and users online also reported it works with 32GB RAM, even though Dell specs claim only 8GB is supported.

[tmihai20]

The RAM question also popped up when H2 was released. These low power SKUs from Intel should support 64 GB of RAM, but not all RAM is equal for these SKUs. There were a lot of modules that did not work properly on H2, even if they did not hit the maximum amount. I expect that those should also work on H3/H3+.

[p01mt3r]

Come home!!! Intel CPU and 60w power supply??!! I don't need a PC to let run Microsoft shit ... Much better arm and not hungry power consumption

[dehade]

Since the h3/h3+ is back, does Hardkernel plan to release new bios updates for the h2/h2+ so that TPM2 and a windows 11 update will be supported.

[arthur99]

Congratulations, it looks like a real nice update. What will be the longevity of BIOS updates for this version?

We've released a minor BIOS update yesterday. https://wiki.odroid.com/odroid-h3/hardw ... os_release
Anyway, I think we can keep updating the BIOS two more years based on two years of contract with the BIOS vendor.

Please keep in mind that firmware as part of the firmware package is a matter of security and vulnerabilities need to be fixed (unless you are selling this for gaming only as a toy). In my country there is a 2 year warranty by law starting with date of purchase, so a distributor such as reichelt.de needs another 2 years support from you as vendor after end of manufacturing and selling the last item from stock. But besides the warranty, it would be best to support a long lifetime of the product by providing updates, where one can expect running 5 years and more until serious hardware defects occur.
Considering this, I don't understand why a BIOS vendor is limiting a contract to such a short period. Or it's just that you didn't ask?

So: What are the options for buying an enhanced product variant with 5 years of security updates, or alternately a service contract available for at least 5 years?

arthur99

[profixit]

Since the h3/h3+ is back, does Hardkernel plan to release new bios updates for the h2/h2+ so that TPM2 and a windows 11 update will be supported.

Due to a restricted BIIOS licensing contract per each platforms, we can not do that. Sorry about that.
If you really want to run Windows 11 on the H2/H2+, consider How to Bypass Windows 11 TPM the Official Microsoft Way section in the following link.
https://www.tomshardware.com/how-to/byp ... equirement

[odroid]

Or it's just that you didn't ask?

1 year of development support + two years of maintenance is a contract term between the BIOS vendor and a board manufacturer.
Except for that, as far as I know that there is no other contract with other terms in particular.
Because of the NDA with the vendor, we can't go into more detail about it publicly.

[odroid]

Thanks to all of you for your interest.
Since this topic already have more than 40 replies, please post your ideas and questions on the H3 sub-forum with a specific topics.
I'm locking this thread.