However, at $799 + 4-cores or 8-cores, the main purpose of this board is for "toy" uses. There's no way that the 4-core IBM is going to be anywhere as fast as a 16c/32t Threadripper 1950x (going for ~$450 these days btw), but it would be the cheapest machine to build for the Power9 system.
So that alone is important. The "Basic" box would be for developer machines, while the "production" machine would be 22-core, or maybe 12-core SMT8 (the "thicker" IBM design where 8-SMT is possible).
Consider this: what if you were developing for Summit? Do you want to rent supercomputer time when you're writing your code? Or do you test it out on a cheaper 4-core, $799 motherboard machine? https://www.olcf.ornl.gov/summit/
For some applications, the compute performance is secondary to that, but price always plays a role. A moderately powerful workstation that runs 100% audited code is very desirable for some uses.
I mean it really sucks that we cry for non-X86 desktops, and when someone finally delivers, we complain about price. It's just little hard to justify for something I'd basically buy as a "toy".
Compare to an Intel 16 thread CPU + motherboard with ECC RAM and the ability to address more than 64GB of RAM. You're well into Xeon parts probably in the $300-400 range just for the CPU, then to get a decent motherboard is probably another $200. So you're paying a 2x premium for this Raptor Computing offering, that's really not THAT bad of pricing!
This pricing is an order of magnitude better than some of the previous Raptor Computing POWER systems which will hopefully open up their market to a wider audience and help drive down prices in the future.
Sure it costs something, though.
More information on FlexVer: https://www.raptorengineering.com/TALOS/documentation/flexve...
> FlexVer™ is a new, owner-controlled security technology designed to safeguard critical data and applications in the event of software or hardware tampering. FlexVer™ allows a system to be provisioned in a trusted physical environment, then deployed to an untrustworthy physical location while retaining system integrity.
OCuLink can turn into M.2, U.2, or a variety of SATA connectors.
Having such a powerful (or at least performant), free and risc-y machine has been a dream of mine for quite some time now... If I can only find a good argument to my wife as why we need one :P
Oh nice! Almost 2x the speed of i9 for 7-Zip compression and ahead of everyone on LLVM compilation speed and Rust prime benchmarks.
> 2 Broadcom Gigabit Ethernet ports > 1 Isolated BMC Gigabit Ethernet port
What does 'isolated' mean in the context? AFAIK ethernet connections are always isolated using transformers, so I don't suppose they are talking about galvanic isolation here.
Can anyone elaborate?
Having the port shared can save in cabling, but you trade that off for trusting some firmware to do the right thing.
Isolated ports mean that you can have one cable that goes to the BMC and only the BMC, and one that goes to the host and only the host. i.e. you place trust for isolation in the fact that two cables go to two different places, and don't involve another bit of firmware.
I think this research was where I heard about this:
So if you want something with published scores on a wide variety of benchmarks that are included in CPU2017 get an Epyc or a Xeon.
Usually the description has a CPU entry something like "Intel Xeon Gold 6148". The IBM entries have things like "3.4 - 3.8 GHz, 40 core, SLES". So if you search any of the lists for "power9" you don't find anything.
Is the power9 only competitive for integer related workloads? E950 finds results on the Integer throughput list, but not the FP throughput list.
Seems like Power9 does really well on interpreters.
the BMC chip (ast2500) has support for being a GPU for the host processor. It's a good old pretty dumb 2D framebuffer. Good for running an OS installer on.
Because (currently, we're working on it) boot time is more than a couple of seconds before we could bring up a discrete GPU, this connector will get you boot progress (and let you see an error) that occurs before we could fire up a PCIe GPU.
Most systems assume a minimal level of 3D acceleration support for desktop use. Watching video, browsing the web, etc... could suck down a lot more CPU than you're used to.
NBD if you're using it as a server, but problematic as a daily driver desktop.
[1] https://wiki.raptorcs.com/w/images/8/89/POWER9_um_OpenPOWER_...
Edit: It was a Lian-Li PC-06SX and not an 05!
> Expected to ship late Q1 2019
such low core count is making the product not that attractive - I can go with µATX MB + Xeon if I want computing power or a decent 6/8-core ARM board for smaller form factor. to just try a different ISA, I'd probably choose RISC-V to maximize my potential investment return.
Power9 just has relatively weak vector units: only 128-bit. (Even AMD Threadripper has support for 256-bit. Power9 128-bit is executed per super-slice, so its really 2x64-bit, to compare against AMD's 2x128-bit support) Otherwise, its throughput is downright insane. 6-issues per clock (Skylake only has 6 if its in the uOp cache: otherwise the decoder only pushes 4 uOps), 10MB of L3 per pair of cores. That's 20MB of L3 on the 4-core, and 110MB of L3 cache on the 22-core.
-----
As far as price/performance goes, the 18-core / 72-thread chip seems best. But the 4-core / 16-thread or 8-core / 32-thread chips aren't bad, as far as I can tell.
The main issue is that most compute-heavy code these days use those vector units (2x128 on AMD, 2x256 on Intel). But if you got a big GPU for SIMD / vectorized ops, then that disadvantage kinda goes away. GPUs are way fatter than Intel or AMD.
The other issue is that AMD is offering cores at stupidly low prices. AMD Threadripper and EPYC are the machines to look at for raw CPU power these days. But I'd argue that the 18-core Power9 ($1,050) is at a good price/performance level. If you can find any application that benefits from the stupid-high 90MB of L3 cache on the 18-core, then you're more or less golden.
I'd bet that databases love that L3 cache.
https://www.solid-run.com/marvell-armada-family/macchiatobin...
[1] one 12kb blob: https://github.com/MarvellEmbeddedProcessors/binaries-marvel... [2] https://github.com/MarvellEmbeddedProcessors/edk2-open-platf...