<davidlt[m]>
Jisheng Zhang seems to be from Alibaba.
<davidlt[m]>
If Alibaba will be working on it, it could kill VF2 :)
<davidlt[m]>
This is just basics so far, enough to boot and get UART.
<davidlt[m]>
Still cannot believe it :)
<thefossguy>
Is this the SoC from the first riscv laptop?
<davidlt[m]>
Yeah, but I don't think there are proof that it exist :)
<davidlt[m]>
Apart that crazy announcement with tons of clickbait.
<thefossguy>
I thought that was just an announcement that "hey we are working on this"
<davidlt[m]>
Don't really care. That was had NFT, Web3, Crypto, etc. all the clickbait tags in it.
<davidlt[m]>
(all that are not related to hardware).
<thefossguy>
Yeah, felt bizarre to me too
<thefossguy>
Does the SoC exist standalone though?
<thefossguy>
For integration in some real products?
<thefossguy>
Or is this like Shakti?
<davidlt[m]>
Probably just a milestone.
<davidlt[m]>
Most RISCV SoC are not designed for specific mass market product.
<davidlt[m]>
But are just a learning milestone and to slowly break into the market (well, find it).
<thefossguy>
Okay
<davidlt[m]>
Yeah, basically, that means that making JH7110 is lost of money.
<davidlt[m]>
loss
<thefossguy>
Intern talk: well now, dont be sad, you gained experience
<davidlt[m]>
Depending on how much Baidu invested, StarFive might have their main market target already.
<davidlt[m]>
A few initial SoCs might not have a particular large market and thus are technically e-waste.
<thefossguy>
I didn’t know Baidu invested. Just checked, its very recent news.
<davidlt[m]>
Yeah.
<davidlt[m]>
If this is a large investment, Baidu will define StarFive priorities.
<davidlt[m]>
It would be nice to see some competition against Alibaba.
<davidlt[m]>
So far, StarFive basically list all possible markets as their interest.
<davidlt[m]>
You only do that for a few years as early startup.
<davidlt[m]>
At some point, they need to narrow down their focus to make money.
<davidlt[m]>
These SoCs are basically to acquire skills, processes, marketing, etc.
<davidlt[m]>
So Baidu might be their moment.
<davidlt[m]>
But we will not see any fruits of that some for years probably.
<thefossguy>
I hope they go in HPC but not further into AI
<thefossguy>
Because AI is such a vendor locked down segment
<thefossguy>
But AI is also profitable for this exact reason
<thefossguy>
Personally, what Tenstorrent did with tensix cores is very genius
<davidlt[m]>
To be determined. It sounds cool, but will it survive we will know for some time :)
<davidlt[m]>
Most companies don't care about some vendor 1st, or even 2nd gen stuff. You typically start gaining traction from 3rd gen. The roadmap and delivering on it is more important.
<thefossguy>
Well, it Jim Keller. His name ought to hold some weight.
<davidlt[m]>
Just look how long it took for AMD to start going again in the server market.
<davidlt[m]>
Not really.
<davidlt[m]>
He worked in pretty much every large company by now.
<thefossguy>
Yeah, first and second gens are just public beta
<davidlt[m]>
What's interesting (I would say) is that Tenstorrent used existing IP to start.
<davidlt[m]>
They have 3 generations (I think) products based on 3rd party IP.
<thefossguy>
I think that’s needed initially? Otherwise the ROI would take a long time
<thefossguy>
Tensix is first party IIRC
<davidlt[m]>
I think their last one is usign SiFive X280.
<thefossguy>
Yeah
<davidlt[m]>
Yeah, that's kinda of prototype phase.
<thefossguy>
With the BF data type
<davidlt[m]>
It takes half a decade basically from idea to a product in a customers hand.
<davidlt[m]>
And again, no one really cares about 1st, or 2nd gen, and only considers you from ~3rd one.
<davidlt[m]>
So basically you need to burn money for a long time before starting to make a profit.
<davidlt[m]>
It's different for HPC. Government can drop you 500 million USD to build a chip, or similar.
<davidlt[m]>
In that case you need to make profit on 1st gen.
<davidlt[m]>
Otherwise you need significant killer app, or be competitive and deliver on your roadmap.
<thefossguy>
You also need to deliver on first gen ;)
<thefossguy>
*deliver well
<davidlt[m]>
So if you are 1.3x better than Nvidia no one will care about you :)
<davidlt[m]>
Delivering + roadmap.
<davidlt[m]>
Because companies are working on their own products, and that depends on 3rd company roadmap.
<davidlt[m]>
Like Apple depends on TSMC delivering their latest manufacturing process for a new iPhone.
<davidlt[m]>
If you mess up, or/and mess up quite often you might go out of the picture.
<thefossguy>
Yeah. Hardware is extremely cut throat.
<davidlt[m]>
fuwei: who is Jisheng Zhang <jszhang@kernel.org> ? Does he work at Alibaba/T-HEAD, or just a person interested in TH1520 upstream support?
<fuwei>
I guess ,NO
<fuwei>
maybe the people in community who wanna to do it
<davidlt[m]>
Anyways, still cool to see someone doing it.
<davidlt[m]>
The patches are minimal enough that this probably will land in v6.5.
<davidlt[m]>
I just hope there will be more parts too.
<davidlt[m]>
I looked at SDK and majority of IP blocks are well known. There is very little 1st party IP.
<davidlt[m]>
If configuration of those IP blocks and revisions are supported it should be easy to enable them sooner than later too.
<davidlt[m]>
GPIO is snps,dw-apb-gpio.
<davidlt[m]>
QPSI is snps,dw-apb-ssi-quad
<davidlt[m]>
SPI is snps,dw-apb-ssi
<davidlt[m]>
Watchdog is snps,dw-wdt
<davidlt[m]>
Ah yes, the interesting part was RTC because it's compatible with apm,xgene-rtc :)
<davidlt[m]>
DMA is snps,axi-dma-1.01a
<davidlt[m]>
eMMC is snps,dwcmshc-sdhci
<davidlt[m]>
same for the SD card.
<davidlt[m]>
I2C is snps,designware-i2c
<davidlt[m]>
TRNG is inside-secure,safexcel-eip76, but I don't know this one.
<davidlt[m]>
Networking might be a problem as it's thead,light-dwmac. So there is something specific for the SoC.
<davidlt[m]>
USB is also problematic as it's thead,dwc3. Again some kind of deviation from standard IP block, or quirks, or something.
<fuwei>
I am looking in to it , but I have ported it to 6.2, happy to see something in upstream
<davidlt[m]>
fuwei: are you interesting in working in upstreaming some bits? I bet you already have the board.
<fuwei>
yes, I am trying to sorting out the patchset
<fuwei>
I have evb, licheepi 4a and beaglev
<fuwei>
I have contacted with him , will call him for a sync today
<fuwei>
alexfanqi:
<fuwei>
davidlt[m]:
<davidlt[m]>
fuwei: cool
<davidlt[m]>
If possible focus on networking, USB and eMMC/SD.
<fuwei>
sure, I am also working on uboot
<davidlt[m]>
Yeah, that's also needed. The fastest storage on this would be USB.
<davidlt[m]>
The benefit of having USB means you can plug pretty much anything (storage, networking, etc.)
kalev has joined #fedora-riscv
<thefossguy>
<davidlt[m]> "GPIO is snps,dw-apb-gpio." <- Is snps Synopsis?
<davidlt[m]>
Yes
<davidlt[m]>
I think, I have a stupid script to scan all builds for breakage.
<davidlt[m]>
But that's after lunch, starving already.
<thefossguy>
Isn’t it 14:30-ish there?
<davidlt[m]>
It is.
<davidlt[m]>
That's why I am starving :)
<thefossguy>
Get a habit, it’s literally harmful for your body
<thefossguy>
It’s already affecting you if you have acidity now
<davidlt[m]>
Let's see if I can setup the test on a remote server instead of enjoying that fan noise at home
zsun has joined #fedora-riscv
zsun has quit [Quit: Leaving.]
<davidlt[m]>
Okay, let's do 30 minutes test and see how many packages we can check.
zsun has joined #fedora-riscv
esv_ is now known as esv
<davidlt[m]>
I think, this will be enough. ~30 min and ~4000 builds checked.
<davidlt[m]>
I have 44 RPMs already affected by TEXTRELs bug.
zsun has quit [Quit: Leaving.]
<davidlt[m]>
50+% of builds scanned.
<davidlt[m]>
Almost done.
<davidlt[m]>
172 RPMs that have TEXTRELs broken, not bad.
<davidlt[m]>
fuwei: did you get any news about 64-core dev kit?
<dtometzki>
<davidlt[m]> "fuwei: who is Jisheng Zhang <..." <- In .mailmap you will see synaptics.com