| 1 Mar 2024 |
|  @adam:robins.wtf joined the room. | 13:32:38 |
|  Pratham Patel joined the room. | 13:34:50 |
| Pratham Patel | In reply to @fgaz:matrix.org
Mic92: What about using some of these while we wait for a RISC-V machine that's suitable for the rack? I'm not the right person to answer but davidlt in #riscv:fedoraproject.org has a few reasons why these cores from Alibaba are not up-to-the spec and are best used for test builds instead of actual builders. | 13:35:52 |
 K900 | AFAIK those cores are also dog slow | 13:37:42 |
| Pratham Patel | As are the ones in the VF2 too | 13:39:44 |
| Pratham Patel | compiler optimizations are not upstream yet afaik | 13:39:59 |
 Mic92 | Are they are any cores, that are not slow for riscv64? | 13:40:56 |
| Pratham Patel | They compare to the ARM Cortex A55 cores if my memory serves me right | 13:41:01 |
 fgaz | As far as I know they are slower than the sifive cores in compilation workloads, but it's not like this is a permanent solution | 13:41:01 |
| fgaz | * As far as I know they are slower than the sifive cores in compilation workloads, but it's not like this is a permanent solution anyway | 13:41:19 |
| Mic92 | Maybe for testing cross-compiled riscv packages? | 13:41:52 |
| Pratham Patel | In reply to @joerg:thalheim.io
Are they are any cores, that are not slow for riscv64? upcoming SG2380 has 12x P670 which are equivalent to ARM's Cortex A78 cores (so faster than the RK3588's A76) | 13:41:55 |
| Pratham Patel | In reply to @thefossguy:matrix.org
upcoming SG2380 has 12x P670 which are equivalent to ARM's Cortex A78 cores (so faster than the RK3588's A76) and 4x more P670 which are somehow binned at a slower freqency | 13:42:32 |
| Mic92 | Isn't sifive now also pivoting into a weird direction? | 13:42:33 |
| Pratham Patel | *frequency | 13:42:37 |
| Mic92 | i.e. firing a lot of folks? | 13:42:40 |
| Pratham Patel | Moving a lot of talent to India, as one does | 13:43:01 |
| Pratham Patel | And serving big customers, as one does | 13:43:28 |
| Pratham Patel | Look at what they did to Horse Creek. It was once as sought after as the HiFive Unmatched successor, but years later, we only see a tiny demo at the Ubuntu Summit (might have the event wrong but it was demonstrated by some Ubuntu/Canonical folks) in ~Nov 2023. No pricing yet. | 13:44:56 |
| Pratham Patel | Milk-V's Oasis will easily beat it in almost every single metric now | 13:45:18 |
| K900 | Well 16 cores is a lot more than 4 cores either way | 13:46:11 |
| K900 | And builders mostly do want more cores | 13:46:16 |
| Pratham Patel | with up-to 96 GB RAM and 16x (maybe 16x physical but 8x electrical) PCIe 4.0 | 13:47:31 |
| Pratham Patel | it's gonna be a monster | 13:47:39 |
| Pratham Patel | starts at $120 USD; haven't told how much RAM that will have; and yes, the RAM is soldered | 13:48:10 |
| fgaz | is mITX going to fit in the rack though? | 13:48:47 |
| Mic92 | can be mITX bee grouped into 4 boards in rack unit? | 13:49:11 |
| Pratham Patel | In reply to @thefossguy:matrix.org
with up-to 96 GB RAM and 16x (maybe 16x physical but 8x electrical) PCIe 4.0 also 5x memory channels IIRC, so best to get the 96GB model to get max memory bandwidth, I am guessing they will order the same SKU of RAM to save cost and the $120 model is gonna have only one channel populated | 13:49:27 |
| Mic92 | * can mITX be grouped into 4 boards in rack unit? | 13:49:48 |
| Mic92 | * can mITX be grouped in 4 boards per rack unit? | 13:50:03 |
