≡
[05:22:19] <adjtm_> marex-cloud, well, olimex sells the A64 SOC for 5.20€ single unit, it's really great for the price
[09:53:54] <jo0nas> agreed, the chip is not top notch - but price and availability of OSHW devices is what makes me prioritize attention to it
[09:54:18] <marex-cloud> Ha
[09:54:36] <marex-cloud> Yes, it's certainly not top notch
[09:55:54] <jo0nas> for top notch, I guess look at RISC-V and consider starting your own production :-P
[09:56:56] <jo0nas> ...and please then OSHWA-certify your products for the rest of us who are fan of transparency but cannot afford to make it ourselves
[10:22:03] <marex-cloud> jo0nas: I don't think the high performance RV cores would be very open
[10:22:12] <marex-cloud> You'd have to implement your own
[10:22:25] <marex-cloud> And then there's all the peripherals
[10:22:28] <marex-cloud> And gpu
[10:39:38] <adjtm_> marex-cloud, the open source riscv cores are decently capable for a device like teres-I (A64 cores are not much faster and the GPU is not usable with free software)
[10:40:10] <adjtm_> by far the worst part would be the peripherals: ddr, usb...
[10:42:38] <adjtm_> but the A64 SOC was the main reason why I didn't consider buying a teres-I, on the other hand I'm very happy with olimex's A20 SBC
[10:47:49] <adjtm_> I don't understand why allwinner made several cheap quad cortex-a53 SOCs: A64, H5, H6... but they should have made one of them hardware compatible with A20 (as they did with A20 as a replacement for A10)
[10:48:21] <diego71> a53 is still faster of a cortex a7
[10:50:18] <marex-cloud> adjtm_: considering that most of the ddr4 controllers are running some sort of iffy firmware nowadays, it seems ddr4 might be a problem
[10:50:38] <marex-cloud> adjtm_: last time I checked Lima, it was quite okay
[10:51:35] <marex-cloud> Regarding the cheap chips, because there's demand and they're easy to make
[10:53:03] <marex-cloud> Btw if you ever compared a53 and a57 performance wise, you'd see there's a huge difference
[10:53:44] <adjtm_> diego71, yes, it's faster than cortex-a7 and cortex-a8, and in some workloads could be more equivalent to cortex-a9
[10:53:55] <adjtm_> cortex-a53 has been a huge success for arm
[10:55:48] <adjtm_> marex-cloud, I agree about cortex-a57, even inexpensive SOCs with them like rk3399 are a lot faster
[10:57:00] <adjtm_> I don't argue against creating cheap SOCs, but why allwinner made so many very similar ones but not continuing with the A10/A20 line (board compatible and with sata and ethernet)
[11:02:09] <marex-cloud> adjtm_: because no one wanted SATA and Ethernet in cheap tablets I guess
[11:02:38] <marex-cloud> adjtm_: and probably they needed more power rails than the A20 package could provide
[11:12:00] <adjtm_> that makes sense, going from cortex-a8 to cortex-a7 (less power), from 1GHz to 960MHz and from 55nm to 40nm allowed them to make a dual core with GMAC from a single core with 100Mb MAC
[11:12:47] <adjtm_> marex-cloud, I don't know if they have moved to something like 28nm for H5, H64 and A64
[11:13:41] <adjtm_> they also upgraded the Mali400 in A10 to Mali400-MP2 in A20
[11:15:25] <adjtm_> they could have used cortex-a35 for lower power
[11:30:16] <marex-cloud> adjtm_: well, they could've, but they didn't :)
[11:30:34] <marex-cloud> They probably had a reason, most likely licensing costs of the core
IRC #olimex 2019-08-19
[01:21:37] <marex-cloud> jo0nas: oh, hum, the chip really isn't great[05:22:19] <adjtm_> marex-cloud, well, olimex sells the A64 SOC for 5.20€ single unit, it's really great for the price
[09:53:54] <jo0nas> agreed, the chip is not top notch - but price and availability of OSHW devices is what makes me prioritize attention to it
[09:54:18] <marex-cloud> Ha
[09:54:36] <marex-cloud> Yes, it's certainly not top notch
[09:55:54] <jo0nas> for top notch, I guess look at RISC-V and consider starting your own production :-P
[09:56:56] <jo0nas> ...and please then OSHWA-certify your products for the rest of us who are fan of transparency but cannot afford to make it ourselves
[10:22:03] <marex-cloud> jo0nas: I don't think the high performance RV cores would be very open
[10:22:12] <marex-cloud> You'd have to implement your own
[10:22:25] <marex-cloud> And then there's all the peripherals
[10:22:28] <marex-cloud> And gpu
[10:39:38] <adjtm_> marex-cloud, the open source riscv cores are decently capable for a device like teres-I (A64 cores are not much faster and the GPU is not usable with free software)
[10:40:10] <adjtm_> by far the worst part would be the peripherals: ddr, usb...
[10:42:38] <adjtm_> but the A64 SOC was the main reason why I didn't consider buying a teres-I, on the other hand I'm very happy with olimex's A20 SBC
[10:47:49] <adjtm_> I don't understand why allwinner made several cheap quad cortex-a53 SOCs: A64, H5, H6... but they should have made one of them hardware compatible with A20 (as they did with A20 as a replacement for A10)
[10:48:21] <diego71> a53 is still faster of a cortex a7
[10:50:18] <marex-cloud> adjtm_: considering that most of the ddr4 controllers are running some sort of iffy firmware nowadays, it seems ddr4 might be a problem
[10:50:38] <marex-cloud> adjtm_: last time I checked Lima, it was quite okay
[10:51:35] <marex-cloud> Regarding the cheap chips, because there's demand and they're easy to make
[10:53:03] <marex-cloud> Btw if you ever compared a53 and a57 performance wise, you'd see there's a huge difference
[10:53:44] <adjtm_> diego71, yes, it's faster than cortex-a7 and cortex-a8, and in some workloads could be more equivalent to cortex-a9
[10:53:55] <adjtm_> cortex-a53 has been a huge success for arm
[10:55:48] <adjtm_> marex-cloud, I agree about cortex-a57, even inexpensive SOCs with them like rk3399 are a lot faster
[10:57:00] <adjtm_> I don't argue against creating cheap SOCs, but why allwinner made so many very similar ones but not continuing with the A10/A20 line (board compatible and with sata and ethernet)
[11:02:09] <marex-cloud> adjtm_: because no one wanted SATA and Ethernet in cheap tablets I guess
[11:02:38] <marex-cloud> adjtm_: and probably they needed more power rails than the A20 package could provide
[11:12:00] <adjtm_> that makes sense, going from cortex-a8 to cortex-a7 (less power), from 1GHz to 960MHz and from 55nm to 40nm allowed them to make a dual core with GMAC from a single core with 100Mb MAC
[11:12:47] <adjtm_> marex-cloud, I don't know if they have moved to something like 28nm for H5, H64 and A64
[11:13:41] <adjtm_> they also upgraded the Mali400 in A10 to Mali400-MP2 in A20
[11:15:25] <adjtm_> they could have used cortex-a35 for lower power
[11:30:16] <marex-cloud> adjtm_: well, they could've, but they didn't :)
[11:30:34] <marex-cloud> They probably had a reason, most likely licensing costs of the core
