I agree with those celebrating the availability of this mATX board. There are not enough of off-the-shelf non-x86 socketed motherboards for proper grunty computing out there. Whether it's for AmigaOS or not (and let's face it, it's not), it's still a cool development.

Heck, the A-Eon AmigaOne hardware doesn't even have a socketed CPU; Removing all doubt on whether or not we're playing with embedded silicon.

I would be more excited if there were socketed ARM CPU/APU/SoC/SoP motherboards. Though not socketed, several years back there was this thing:
Gigabyte MP30-AR0
Should I get excited because it is technically possible to have AROS running on it?

For linux, and FOSS I could see this being an interesting product. Though for emulation it is kind of overkill. It'd be like celebrating the arrival of the Canon EOS R because it was possible to emulate a polaroid picture.

_________________
All the way, with 68k

@matthey

>A Raspberry Pi 3 using 4 cores will outperform the POWER9 using one thread in some cases.

CPU wise: price performance of the Pi 3 is superior to Power 9, but why are we comparing 1 thread to 4 cores ?

Last edited by bennymee on 08-Mar-2019 at 10:10 AM.

@matthey

Quote:

POWER9 using virtualization will have less performance so maybe you should benchmark before calling the Raspberry Pi 3 performance a joke.

Don't put words in my mouth.

I mentioned nothing about RPi performance. It is a great little device. I use it with VNC as a web browser, but unless people are willing to pay more for the OS than the hardware I don't believe a port would be possible. Of course you could start a bounty.

_________________
Tinkering with computers.

Quote:

agami wrote: Heck, the A-Eon AmigaOne hardware doesn't even have a socketed CPU; Removing all doubt on whether or not we're playing with embedded silicon. I would be more excited if there were socketed ARM CPU/APU/SoC/SoP motherboards. Though not socketed, several years back there was this thing: Gigabyte MP30-AR0

That GIGABYTE motherboard is a work of art (Nicest I've seen since the Natami MX motherboards which were made more impressive by the fact they were proto boards with an FPGA CPU). Are you sure the X-Gene1 SoC is not socketed?

Applied Micro (AMCC) bought/licenced the IBM PPC 4xx CPU line (mostly used for embedded). The Acube Sam440 and Sam460 motherboards used AMCC PPC 440 cores and the Apollo Core competed against and outperformed a synthesizable PPC 440 core in FPGA in an IBM contest so there is Amiga history here. AMCC was bought out and ended up with Ampere Computing. The X-Gene name was changed to eMAG but the designs are similar and even the X-Men themed micro-architecture names were kept (Storm, Shadowcat, Skylark, Quicksilver). It sure looks like the eMAG SoC is socketed as a chip is shown in a picture at the following link.

https://fuse.wikichip.org/news/1663/ampere-ships-first-gen-arm-server-processors/

It is possible the CPU socket is compatible but there were some major I/O changes. If the X-Gene SoC uses the Storm micro-arch then it probably is fabbed at 40nm which is the same as the "outdated" Raspberry Pi 3 SoC. More modern ARM chips are no doubt higher performance (even some ARM designs on a newer process). The modern Qualcomm Centriq 2400 is socketed as the chip can be seen in pictures from the following article.

https://www.anandtech.com/show/11737/analyzing-falkors-microarchitecture-a-deep-dive-into-qualcomms-centriq-2400-for-windows-server-and-linux

ARM Holdings is completely fabless so does not produce any chips to sale. This makes it more difficult to try to standardize a socket, I/O pins and timings. I expect these ARM architecture license holders each have incompatible sockets but it is common for server processors to be socketed. They are expensive enough that the extra cost of a more expensive package likely isn't a concern and, as I recall, they have better heat dissipation.

Quote:

bennymee wrote: CPU wise: price performance of the Pi 3 is superior to Power 9, but why are we comparing 1 thread to 4 cores ?

There is no PPC Amiga OS which uses SMP so they would only support a single thread on the POWER9. The Amiga could only use somewhere between 1/4 and 1/16 of the performance of a 4 core POWER9.

@Barana

One thing I noticed is that the speed looks a bit slow. 3.2Ghz? Or 3.45Ghz for the bigger one. Stil looks a bit dated. I think it should be at 4.2 at least by now. How do they expect to compete against a Ryzen/i7?

I am surprised the current fastest supercomputer in the world is a POWER9. Though the current list is old. Over a year. That's like ten years in dog terms. Only 3.07Ghz? For a supercomputer? Nah they need to go faster!

I reckon 4.2 will be SMP before it is 64. If it even exists.

Last edited by Hypex on 09-Mar-2019 at 03:40 PM.

@Signal

Quote:

I own a POWER9 and have no delusions of ever having AOS/MOS running on this beast. All I can say is, it would be nice.

Considering that, the hardware this is very similar to the first AmigaOne. It is a POWER based server board. With matching expensive ECC registered server RAM. Can fit into an ATX case. Cost a lot still. Sounds familiar?

OS4 was ported to a similar board the first time. They can do it again. I don't think the current boards are sustainable but no one listens to me as usual. Too little for too much. Amiga people seem to hate them while always comparing with a PC. But they sleep with the enemy and have for years so they just don't understand. They don't get it. On price and perfomance they can't even compare an AmigaOne Xxxx with an XE.

Right now they are running OS4 on dual core and soon to be four core boards in a crippled state. So the same could surely be done. It's not a cheap solution. But if you are going to pay more money than the average bear. Then you might as well get some kick ass kit for your banging bucks.

Last edited by Hypex on 09-Mar-2019 at 03:43 PM.

@matthey

Quote:

POWER9 3.8GHz (1 thread) 4090 3140ARM Cortex-A53 1.5GHz (4 cores) 2630 5940

Seriously? POWER9 can be beaten by Cortex A53?

The Power9 is in desktop class. I had better hopes than this.

If they can't get their act together by now I can't see a Power10 around the corner. Or perhaps it will be a PowerX, crossed off the list.

@Hypex

Multicore benchmarks can lead to sometimes interesting results...

Single core comparison in the same benchmark (compressing MIPS, decompressing MIPS):

PowerPC 970FX 1.8 GHz (1 thread) 750 1330
PowerPC 970MP 2.5 GHz (1 thread) 1230 2050
POWER9 3.8GHz (1 thread) 4090 3140

ARM Cortex-A53 1.5GHz (1 thread 32 bit ARM32) 880 1600

Core i7 4770 (Haswell) 3.4 GHz (1 thread) 4000 4000


It is obvious Power9 offers splendid single core performance n this benchmark.

Quote:

pavlor wrote: Multicore benchmarks can lead to sometimes interesting results...

Yes. Highly parallel workloads allow the cores to work together so the performance of each core can nearly be added together while sequential workloads are the same performance as only one core is used.

Quote:

Single core comparison in the same benchmark (compressing MIPS, decompressing MIPS): PowerPC 970FX 1.8 GHz (1 thread) 750 1330 PowerPC 970MP 2.5 GHz (1 thread) 1230 2050 POWER9 3.8GHz (1 thread) 4090 3140 ARM Cortex-A53 1.5GHz (1 thread 32 bit ARM32) 880 1600 Core i7 4770 (Haswell) 3.4 GHz (1 thread) 4000 4000 It is obvious Power9 offers splendid single core performance n this benchmark.

The single core single thread performance is significantly better on the x86_64 CPUs. It would be better to compare to the same fab process. The Core i7 Haswell you chose is 22nm where the POWER9 is 14nm. Broadwell and Skylake are 14nm. All the following are single thread performance at 14nm.

POWER9 3.8GHz 4090 3140
Intel i7-6900K (Broadwell) 4GHz 4900 4800
Intel Xeon E5-2699 v4 (Broadwell) 3.6GHz 5100 3900
Intel i7-6700 (Skylake) 4GHz 4640 4640
Intel i7-7820X (Skylake) 4.3GHz 5300 4600

POWER9 gets spanked here but let's compare single core (total) performance with multi-threading where the POWER9 should be good.

POWER9 3.8GHz (1 core 1 thread) 4090 3140
POWER9 3.8GHz (1 core 2 threads) 5540 4560 (35% and 45% improvement)

Intel Xeon E5-2699 v4 (Broadwell) 3.6GHz (1 core 1 thread) 5100 3900
Intel Xeon E5-2699 v4 (Broadwell) 3.6GHz (1 core 2 threads) 7350 5170 (44% and 32% improvement)

Intel i7-7820X (Skylake) 4.3GHz (1 core 1 thread) 5300 4600
Intel i7-7820X (Skylake) 4.3GHz (1 core 2 threads) 8270 6150 (56% and 33% improvement)

We can start to see a POWER9 advantage in decompression which is a highly parallel workload but the less parallel compression allows the single core single thread performance of the x86_64 cores to dominate in performance. Some servers can keep many threads busy with unrelated tasks allowing the more efficient fine grained multi-threading to have an advantage in single core performance and energy efficiency. It looks to me like POWER9 has an advantage but it is specialized.

In general, single core single thread performance is the holy grail. It allows work to be done at a faster pace before idling which is more energy efficient or the frequency can be reduced to reduce power requirements. ARM understands this for embedded.

Quote:

Performance and power dissipation are closely linked. By using more efficient instructions (ARM instruction set), it may be possible to clock the design at a lower speed to meet the specified functionality and, therefore, significantly reduce the power dissipated in the design. Alternatively, by completing the required operations earlier then putting the system into a low-power mode, it may be possible to reduce the overall system power consumption.

https://www.eefocus.com/data/07-06/27_1181634654/File/1181636102.pdf

Ironically, the ARM strategies in the server market have been to add cores but it doesn't help to have more inefficient cores. ARM cores usually have lower idle power but that isn't so good for most servers. For all the instruction sets and DynamIQ/big.LITTLE (couples a high performance core with low power core) configurations they have, they (and other top chip designers) have failed to design a core with competitive single core single threaded performance in the server and desktop markets. At least POWER9 found a small niche and has some server market share.

@matthey

Quote:

POWER9 gets spanked here

I see your interpretation of these numbers is quite creative.

Power9 is in the same performance league (in this benchmark) as recent x86_64 CPUs. Performance increase over current PowerPC AmigaOne designs (see PowerPC 970FX result for comparison), compatibility with current software base and price/performance ratio make Power9 an interesting target for PowerPC based Amiga-like OSs.

We both know Amiga users are willing to pay 2000 USD for telecom-SoC based AmigaOne. With Power9, they would get mainstream performance for the same price.

@pavlor


WELL DONE, Pavlor.
I couldn't and didn't say it as well as you, good job.

If you were a golfer, That's the best hole in one I've seen, Ever.

Quote:

by pavlor on 10-Mar-2019 18:41:12 @matthey Quote: POWER9 gets spanked here I see your interpretation of these numbers is quite creative. Power9 is in the same performance league (in this benchmark) as recent x86_64 CPUs. Performance increase over current PowerPC AmigaOne designs (see PowerPC 970FX result for comparison), compatibility with current software base and price/performance ratio make Power9 an interesting target for PowerPC based Amiga-like OSs. We both know Amiga users are willing to pay 2000 USD for telecom-SoC based AmigaOne. With Power9, they would get mainstream performance for the same price.

Last edited by Barana on 10-Mar-2019 at 10:39 AM.

_________________
Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway.

I serve King Jesus.
What/who do you serve?

@matthey

I am talking about the benchmark claiming the Pi 3 is superior to the Power 9, you are comparing 1 Power 9 thread to 4 Pi 3 cores, that is my point.


A copy from 7-zip page:

Power 9, 1 thread, Mips decompressing = 4090
Pi 3, 1 thread, Mips decompressing = 860

Only Aros can use, in some degree, more then 1 core/thread.


That the Power 9 is not fully utilitzed by AmigaOS 4 is not really an issue, even with 1 core it is faster then anything else.