Hacker News Reader

Comment by bcatanzaro | Hacker News Reader

Importantly, it appears some of their GPU comes from Imagination Technology. https://www.theinformation.com/articles/chinas-little-nvidia...

View on HN

That company's GPU was used in a few of Intel's early Atom SoCs, and infamously lacked documentation and source code, unlike other Intel GPUs of the time. It was also dismal in performance and efficiency, lacking a blitter and other acceleration for 2D unlike nearly all PC GPUs since the original 8514.

Ironically the source and docs were leaked many years ago, causing a lot of commotion in the OSS community; and more relevantly to this article, a Chinese underground forums community managed to make a decent usable driver for Windows XP, something neither Intel nor Imagination had.

Imagination IP was also the core of apple's initial mobile GPU, which apple initially modified slightly for their products, before eventually replacing them after years of access. They were not terrible designs.

Poulsbo, that brings back memories. https://www.phoronix.com/scan.php?page=news_item&px=MTMyODA

Imagination technologies reminds me of Luc Verhaegen's attempt at creating opensource drivers for PowerVR GPUs... If it weren't for the official effort by imagination technologies to create an opensource driver for Vulkan I would have considered imagination technologies to be one of the worst GPU companies.

Do you have a link to more information on that XP driver?

It's been quite a few years now, I tried looking for it again with no success (Google's increasingly worse results don't help either) and I'd have to dig out the Atom box to check what the files were called...

There were plenty of these tiny Atom PCs sold cheaply and they ended up in lots of semi-embedded applications, which might be why the community was so motivated to make a driver, and possibly even the initial source leak was because of that.

> lacking a blitter

That's fucking hilarious! :D

How can they make such a design decision and think it's okay?

"Blitter? Nah. Nobody needs that anyway!" ??

It was a design intended for 3D acceleration, and they probably thought the additional hardware blitter wasn't necessary in a space and power-constrained mobile application where these cores mostly ended up.

What rationale did the UK government use when approving Imagination being sold to a private equity firm controlled by the Chinese government?

Supposedly because at the time of the acquisition, "Canyon Bridge was licensed and regulated by US law." but "Since then it has moved its headquarters to the Cayman Islands and as such is no longer a US-controlled entity." [1]. Although that sounds pretty flimsy. I suspect, given it happened in 2017, it was judged politically unwise to block a Chinese state investor when the UK was looking to increase global trade post-Brexit.

Eventually it seems the UK gov woke up and blocked a Chinese attempt to take over the board and move the company to China[2]. Now it seems there's a plan to IPO again in London or Nasdaq[3] so the Chinese owners can exit.

[1] https://www.bbc.co.uk/news/business-52275201 [2] https://www.ft.com/content/654a6d68-ef26-47b2-9da5-9029c570f... [3] https://www.ft.com/content/e0c48d30-866d-4efd-9d78-21e49d366...

As we've seen with Russian money and influence over London... the UK government couldn't care less where their IP goes.

IMG wasn't considered important enough to be a strategic asset unlike ARM. Remember IMG was on the verge of bankruptcy or collapse. Apple, representing 70% of their revenue will somehow no longer be paying them a single penny. They had lots of baggage that were not part of Graphics IP from Wireless, MIPS, to IoT.

ARM had Mali. So on the surface UK isn't really losing any strategic asset.

And you have to remember, this was in 2016 / 2017. The world still doesn't know a damn thing about China. ( or may be they do but they turned a blind eye to it )

None whatsoever - they don't care.

Oh trust me, they do care if your company is important enough.

The sale of a company I'm deeply invested in has been dragging on through reviews from the UK MOD for most of a year.

This place had revenues of 125m freedom bucks last year, its hardly a big player.

They do:

https://www.thisismoney.co.uk/money/markets/article-8196445/...

That was after there was a public backlash.

There's another Chinese GPU coming based on that design:

https://videocardz.com/newz/innosilicon-graphics-cards-based...

Even if relies on TSMC and IP-blocks purchased from other companies - it is still cool that a 2-year old startup can create graphics card that appears to compete with what giants like NVDIA and AMD are offering.

there is no price, there is no power usage, the rated fp32 is roughly on par with a video card that is 6 years old now.

unless it's crazy efficient (at 12nm I doubt it), I wouldnt say it's competing with the giants.

If available and priced attractively I would very much buy this card. I never buy top of the line graphics cards but the equivalent of 1060-1070 works just fine for my needs. And I bet that there are enough people all over the world in the same position. I think it is a competition assuming they do not fold.

Also it is a startup. If they do not go down for one or another reason nothing precludes them from going higher end given a time and favorable market.

My guess from this press release is it'll be heavily focused on the China domestic market, since there are a lot of MIC 2025 related subsidies/incentives for that. Note the list of compatibility with various domestic architectures and software.

>If available and priced attractively

i'm thinking the same. the game that i play now is either wow or hearthstone which isn't very graphic intensive. i think this card if price cheaply can be use for casual gaming and workstation (maybe?)

Something people forget is that it is a dedicated graphics card. Unless someone is going to implement high bandwidth system memory like Apple did, that Chinese graphics card will run circles around most integrated GPUs. The competition is horrendously expensive because of the high demand for GPUs. There is enough room for a niche GPU manufacturer nowadays.

It would run most of the games just fine. Sure heavy games would not be able to run at the highest settings but then again not that many people would care. And the ones who do can always splurge some megabucks on high end nVidia / AMD.

Maybe it'll have Linux drivers.

Asking the important questions !

But maybe it can be sold in Russia.

Doubt it. It's either manufactured in Taiwan or by a Taiwanese compant in China.

It doesn't matter. You could always still buy everything in Iran regardless of what's sanctioned. Engineering firms were actually a bit upset because even though they wanted to buy licenses for e.g. STAAD Pro, they were forced to use pirated copies.

At the end of the day the sanctions will do nothing but increases prices. The goods themselves won't stop flowing.

I thought cards supporting AV1 are new-ish?

The entire GPU IP is licensed from Imagination. This is like Google making their Tensor phone chip and claiming it's novel when in reality everything is licensed from someone else and all they had to do was integrate it together.

Except it’s three or four generations behind in performance. The article compares it to a 1060 from Nvidia.

According to Steam Survey, the 1060 is the most popular GPU in use today.

But is the 1060 the most purchased gpu? What matters is sales, not users.

Nonsense. Sales = Users buying.

When the vast majority of people use a 1060, then a new card with roughly that performance, at an attractive price point, will sell. The reason is simply that, apparently, a 1060 is enough.

You're being misled by all the voices who only ever talk about the fastest and most modern stuff. Reality is that the vast majority of people aren't anywhere near the high-end. That's the market.

to a GeForce GTX 1070 in fact, 2 generations behind the current RTX 3070. All 3 generations are compared in this video https://www.youtube.com/watch?v=t3Iq6pFUdvE

A 1070 though was slower or on par to a 980ti which was released 7 years ago.

Volta (10x0 series), Ampere, Turing (16x0 & 20x0), then Lovelace (30x0).

10x0 was Pascal, Volta was Quadro only (Titan V), 16x0 and 20x0 are Turing, 30x0 is Ampere and Lovelace is Quadro only at the moment.

Half of that is because it's on a 12nm node.

It compares it to a 1070 which is on par with a 2060.

> can create graphics card that appears to compete with what giants like NVDIA and AMD are offering.

uh ... no. six teraflops is a card from 2017.

But the market is how it is. I'm on an 1050Ti. Even this would be an upgrade if price/power consumption is right.

I have strong doubts about power consumption though.

well, if you want to say "competes with 2017 nvidia," that i could agree with

but i basically guarantee that they keep falling further and further behind.

2022 basically doesn't exist unless you're willing to overpay to have the latest and greatest :)

You can get NVidia cards at retail prices right now

My GPU that was midrange in 2012 could use an upgrade.

That seems to be about an AI chip. This is a GPU.

Reply on news.ycombinator.com