Milk-V Oasis Sounds Like An Interesting RISC-V Board With 16 Cores, Up To 64GB LPDDR5

[drastic]

Super exciting machine, some preliminary research shows the gpu averaging about 0.5tflops for fp32 and 1tflop for fp16, these cores are claimed to be quite fast, the gpu is quite the nice gpu, and plenty of IO, this really seems like a "daily driveable" class hardware for an insane price. this can be a massive win for riscv. would love to try emulation on this machine

I agree that the estimated price is the most important aspect of this announcement.

I always get lost in Tflops, and I personally care much more for integer performance. Can someone here produce a rough GeekBench 5 single-threaded score, just for orientation? I agree that it is a big win for RISC-V, and I would love to see RISC-V ecosytem progress further.

Hopefully, whn this board is released, we might get a first geberal-purpose RISC-V desktop PC.

[drastic]

Ok, so SiFive P670 is a 4-issue, out of order, 2.5 GHz. I guess that would be closer to 700 Points in GeekBench 5 single-threaded. That's crazy fast for a Risc-V. But, someone must confirm this performance number that I have just guessed.

This thingy should be able to easily run an entire standard Linux desktop, like Ubuntu.

[drastic]

Since the CPU is obviously more than good enough for running a Linux desktop, I was trying to figure out whether the GPU is good enough.

The integrated IMG AXT-16-512 from IMG A-Series (Albiorix) was "released" towards the end of 2019. It supports OpenGL 3.0, OpenGL ES 3.x, and Vulkan 1.3 (OpenGL 3 is good enough for a Linux desktop).

The GPU performance is hard to estimate, but I put it at around 10 FPS in Heaven 4.0 Extreme, which is similar to the AMD Vega 3 that is currently shipping in many laptops. That confirms that this board is inteded to run a desktop, and it is capable of doing so.

While the GPU performance is sufficient, there is no PCI-E expansion slot (edit: I am wrong here, there is a PCIe slot on the board), so the GPU performance could be crucial for sucess of this product in the segment of cheap small desktops. In that light, a faster GPU would have been better (instead of the gazillion integrated CPU cores), and the IMG AXT-32-1024 GPU would have been a better choice, in my opinion. So, multithreaded performance is prioritized in this product over GPU performance.

For anyone who needs a small low-power desktop computer with a lot of CPU cores, this should be a great product.

[Quackdoc]

I wonder if there'd be nicely working OpenCL / Vulkan support for it.
Could be nice for home automation or something where one wants some compute power for the control functions and better I/O than RPI.

these devices have closed source drivers, mesa driver is still not ready

I agree that the estimated price is the most important aspect of this announcement.

I always get lost in Tflops, and I personally care much more for integer performance. Can someone here produce a rough GeekBench 5 single-threaded score, just for orientation? I would guess around 200 points, but the speed doesn't really matter, what matters is that people have a cheap board to develop on. I agree that it is a big win for RISC-V, and I would love to see RISC-V ecosytem progress further.

Hopefully, in about 10 years, we might see a RISC-V desktop PC.

IMG reports 2tops on their marketing stuf, however if you are looking into doing integer compute, its a fairly safe assumption you are doing some sort of compute workload? in that case the NPU should greatly interest you instead, since you get a 8 Core SiFive X280 + Sophgo TPU for up to 20 tops.

Since the CPU is obviously more than good enough for running a Linux desktop, I was trying to figure out whether the GPU is good enough.

The integrated IMG AXT-16-512 from IMG A-Series (Albiorix) was "released" towards the end of 2019. It supports OpenGL 3.0, OpenGL ES 3.x, and Vulkan 1.3 (OpenGL 3 is good enough for a Linux desktop).

The GPU performance is hard to estimate, but I put it at around 10 FPS in Heaven 4.0 Extreme, which is similar to the AMD Vega 3 that is currently shipping in many laptops. That confirms that this board is inteded to run a desktop, and it is capable of doing so.

While the GPU performance is sufficient, there is no PCI-E expansion slot, so the GPU performance could be crucial for sucess of this product in the segment of cheap small desktops. In that light, a faster GPU would have been better (instead of the gazillion integrated CPU cores), and the IMG AXT-32-1024 GPU would have been a better choice, in my opinion. So, multithreaded performance is prioritized in this product over GPU performance.

For anyone who needs a small low-power desktop computer with a lot of CPU cores, this should be a great product.

not too hard to estimate, it gets around 0.5 fp32 Tflops and and 1Tflop of fp16, more then enough for desktop and gaming workloads, judging by flops alone, which is flawed ofc since drivers do matter. but it is roughly the equivilent of modern apus from AMD that use 2CUs for gpu. Also yes there is a PCIe slot, it has PCIe Gen 3 x8 slot on the machine

[Sonadow]

Hard pass on this and any RISC-V system until I see a clear and definite RISC-V standard platform implementation that will allow me to take any up-to-date kernel from kernel.org, build it with nothing more than the .config file used by a desktop Linux distribution, and successfully boot up and use said distribution with that new kernel like what I have been doing with x64 systems for the past 11 years.

I will never run desktop Linux on any hardware that doesn't allow me to always use the latest available mainline kernel, especially when Linux is the only system where the kernel is more important than the OS userland because every single damn driver has to be baked into it.

This is the kind of restriction you'd expect on Apple hardware and macOS, not on hardware that aims to be an "open commodity". If this is the case then there is absolutely no difference between RISC-V and the giant crappy mess that is ARM today where one board only works with one vendor-supplied kernel version for only one Android or desktop Linux distribution.

Even ChromeOS promises at least 8 years of official ChromeOS support for Chromebooks and Chromeboxes and Apple maintains support for at least four or five releases of macOS for every machine it ships. As it is M1 has been supported by macOS 11 to macOS 14 and it sure doesn't look like Apple intends to stop macOS support for the M1 any time soon.

[drastic]

Also yes there is a PCIe slot, it has PCIe Gen 3 x8 slot on the machine

Yep, there is actually one PCIe x8 Gen3 in an x16 slot. That makes it a pretty rounded desktop machine, since you can plug anything you want into that PCIe x8, including a more powerfull GPU with open-sourced drivers. Heck, you could even run modern games, if there was some good x86-64 emulator for it.

Can't think of any serious objections, so now I want to buy one. Perhaps some time after the drivers for that SoC are stabilized.

[uxmkt]

And there was I thinking that American English, the language of innovation, freedom, and simplification, would have taken Noah Webster's lead and removed unecesary friperies like double leters, in a similar way to those in traveler, counselor

As Eddie Izzard already noted, AE at least got "thru" (through) right.

[Quackdoc]

Yep, there is actually one PCIe x8 Gen3 in an x16 slot. That makes it a pretty rounded desktop machine, since you can plug anything you want into that PCIe x8, including a more powerfull GPU with open-sourced drivers. Heck, you could even run modern games, if there was some good x86-64 emulator for it.

Can't think of any serious objections, so now I want to buy one. Perhaps some time after the drivers for that SoC are stabilized.

Im seriously thinking about early adopting one even if only to be a tester for various things. the price is just too good

[SteamPunker]

not iven kloz

Okay, I should have emphasized that it's at least more phonetic than British English. It's somewhat of an improvement. (And I continue to stand by that.)

As you and others here rightly pointed out, English is indeed not a phonetic language overall.

[dracc]

Neat. I tried to pre-order but I keep getting an error during checkout whether I use Google Pay or Shop Pay. "There was an issue processing your payment. Try again or use a different payment method." I'll try again later.

They've since updated the pre-order price to be $5 because the $0.1 payment was troublesome to many customers.
These $5 will be deducted from your total when you purchase the Oasis.

Yes, I've placed my (first ever) pre-order.