5 ms·

It's all just a bit ridiculous because it's an SBC SoC with basically those kinds of specs -- onboard flash for storage, etc. but dressed up like it's a desktop

It's all just a bit ridiculous because it's an SBC SoC with basically those kinds of specs -- onboard flash for storage, etc. but dressed up like it's a desktop computer. "nVME SSD via USB" -- ehhh. No thanks. I want an ARM workstation. This isn't it. It's a step above a Raspberry Pi, but not much of one.

sweetjuly 3y ago
> I want an ARM workstation. This isn't it. The RK3588 is actually a pretty decently spec'd SoC. It's a dual cluster with 4xA76 (P-cluster) and 4xA55 (E-cluster). Neoverse-N1 is nearly identical to an A76 and it what most cloud vendors offer and it's performance is generally decent. It's only four P cores but it's no slouch by any means.
- fmajid 3y ago
  It’s half the speed of a 4 year old iPhone 11. So yes, a total slouch, like all the non-phone ARM SOCs, only useful in low-power embedded applications.
  - sweetjuly 3y ago
    Sure, but the iPhone 11 [1] is also 17% faster in single core Geekbench 5 scores than the Intel i9-9880H [2] that Apple was using in their top of the line MacBook Pros back then too, so "a four year old iPhone" is not exactly some old and unusably slow processor. Apple's mobile SoCs were (and largely still are) in a performance class of their own. Half as fast single core on the RK3588's P-cores is still perfectly usable for workstation activities, in fact that's about what you'd get on an i5-8210Y MacBook Air from 2018 [3] which I'm sure a few people here are using as their daily driver. [1] https://browser.geekbench.com/v5/cpu/21459716 https://browser.geekbench.com/v5/cpu/21459716 [2] https://browser.geekbench.com/v5/cpu/21459261 https://browser.geekbench.com/v5/cpu/21459261 [3] https://browser.geekbench.com/v5/cpu/21415175 https://browser.geekbench.com/v5/cpu/21415175
- cmrdporcupine 3y ago
  The issue is the lack of real PC-quality I/O interfaces.
  - sweetjuly 3y ago
    The SoC does have 4xPCIe 3 lanes and 1xPCIe 2 lanes. The lack of IO is purely a board design issue, you can open KiCad and fix this problem right now.
    - cmrdporcupine 3y ago
      All the more reason why this product is so garbage. They could have made what they're advertising (a PC) by doing that... but instead they just slapped a SBC barely better than an RPi in a case and marked the price way up. Disappointing.
fmajid 3y ago
Try this, but it’s very expensive at nearly $3K: https://www.jeffgeerling.com/blog/2023/ampere-altra-max-windows-gpus-and-gaming https://www.jeffgeerling.com/blog/2023/ampere-altra-max-wind...
- PlutoIsAPlanet 3y ago
  This kind of generic OS support should be mandatory on all ARM SBCs, otherwise its just adding to more e-waste.
  - fmajid 3y ago
    I believe this is because ARM doesn’t have a standardized hardware interface like x86 does so every chipset vendor does it’s own thing, mostly closed source for no good reason.

sweetjuly 3y ago

> I want an ARM workstation. This isn't it. The RK3588 is actually a pretty decently spec'd SoC. It's a dual cluster with 4xA76 (P-cluster) and 4xA55 (E-cluster). Neoverse-N1 is nearly identical to an A76 and it what most cloud vendors offer and it's performance is generally decent. It's only four P cores but it's no slouch by any means.

fmajid 3y ago
It’s half the speed of a 4 year old iPhone 11. So yes, a total slouch, like all the non-phone ARM SOCs, only useful in low-power embedded applications.
- sweetjuly 3y ago
  Sure, but the iPhone 11 [1] is also 17% faster in single core Geekbench 5 scores than the Intel i9-9880H [2] that Apple was using in their top of the line MacBook Pros back then too, so "a four year old iPhone" is not exactly some old and unusably slow processor. Apple's mobile SoCs were (and largely still are) in a performance class of their own. Half as fast single core on the RK3588's P-cores is still perfectly usable for workstation activities, in fact that's about what you'd get on an i5-8210Y MacBook Air from 2018 [3] which I'm sure a few people here are using as their daily driver. [1] https://browser.geekbench.com/v5/cpu/21459716 https://browser.geekbench.com/v5/cpu/21459716 [2] https://browser.geekbench.com/v5/cpu/21459261 https://browser.geekbench.com/v5/cpu/21459261 [3] https://browser.geekbench.com/v5/cpu/21415175 https://browser.geekbench.com/v5/cpu/21415175
cmrdporcupine 3y ago
The issue is the lack of real PC-quality I/O interfaces.
- sweetjuly 3y ago
  The SoC does have 4xPCIe 3 lanes and 1xPCIe 2 lanes. The lack of IO is purely a board design issue, you can open KiCad and fix this problem right now.
  - cmrdporcupine 3y ago
    All the more reason why this product is so garbage. They could have made what they're advertising (a PC) by doing that... but instead they just slapped a SBC barely better than an RPi in a case and marked the price way up. Disappointing.

fmajid 3y ago

Try this, but it’s very expensive at nearly $3K: https://www.jeffgeerling.com/blog/2023/ampere-altra-max-windows-gpus-and-gaming https://www.jeffgeerling.com/blog/2023/ampere-altra-max-wind...

PlutoIsAPlanet 3y ago
This kind of generic OS support should be mandatory on all ARM SBCs, otherwise its just adding to more e-waste.
- fmajid 3y ago
  I believe this is because ARM doesn’t have a standardized hardware interface like x86 does so every chipset vendor does it’s own thing, mostly closed source for no good reason.