> As I understand it, the GPU world is full of patents and dominated by just a few parties.

This would be a impediment typically, but considering they're blackballed already, how beholden are they to patents? Block imports? Think that's being done already. Requiring US companies to not do business with them? That's being done already.

At this point, they just have to ride out the US/China trade war and then when things are being negotiated, I imagine a "turn a blind eye" to Huawei is going to be given. They're gimped in some ways, but in others they've probably never been more free.

Actually, what's to stop Huawei from just copying and incorporating whatever they need into their IC designs regardless of patents? I can imagine foreign foundaries refusing orders from Huawei to avoid manufacturing IP infringing designs, but are there foundaries in China that can produce the chips necessary while ignoring foreign patents?

As an aside to that, how do foundaries know whether their client designs infringe on patents? Is there some sort of automated system that goes through the design files they are sent looking for areas of silicon that may contain infringing IP blocks?

>are there foundaries in China that can produce the chips necessary while ignoring foreign patents?

Realistically, no. The only two competitive, bleeding edge foundries are TSMC and Samsung.

SMIC are now rolling out 14nm - that's apparently competitive enough for people to still use Intel CPUs.

There's a very, very big difference between "rolling out 14nm" with unknown yields, performance, etc. Vs Intel's venerable 14nm process.

Intel is Intel, as unhelpful as that statement is. So these guys are starting on 14nm, while others are pushing ahead to 5 and below. Sure it'll work, but I don't find that competitive in an already crowded smartphone market in the next 2 years.

They've been rolling it out since 2016, better wait until their product actually appears.

> Is there some sort of automated system that goes through the design files they are sent looking for areas of silicon that may contain infringing IP blocks?

I'm going to venture a solid guess that no such system exists in America simply because possessing the IP to match potentially infringing IP against would be problematic for the companies it's designed to protect.

Intel: "Hey GlobalFoundries, you're not one of our suppliers but here's the schematic for our latest chip. Please make sure nobody takes it!"

And I can all but GUARANTEE that doesn't exist in China. Unless of course it's to pick up any IP they might have missed when they stole it the first time.

Infringement is detected by display of patented features or potentially reverse engineering relevant portions of an IC. There is no practical way to police infringement using the disparate array of design inputs mapped onto proprietary gate libraries.

Infringement would most likely be detected by reading data sheets and looking for similarities. Armed with that, an IP holder would go through the very expensive reverse engineering process. They're unlikely to do that unless a win in court is likely.

There is no easy way to check for IP infringement here. When the key idea of a design may be 100% lifted illicitly, the schematics may be somehow different, and the mask layout, incomprehensibly different.

You imply Huawei was playing by the "rules" and not already stealing. [1] [2]

[1] https://arstechnica.com/tech-policy/2019/01/us-indicts-huawe...

[2] https://www.bloomberg.com/news/features/2019-02-04/huawei-st...

This seems disingenuous and arbitrary considering the decision to put Iran on a ban list was basically taken unilaterally by the US, all the while Cisco and Bluecoat were selling tech to build the great firewall of China, and a long list of other (US, Israeli, etc.) companies are selling tech to basically every authoritarian, oppressive regime out there without any legal or moral consequences.

As for Huawei, the fact that the bans and sanctions tend to be fluid and negotiable despite being called a national security issue kind of puts the "national" part into question. Someone certainly benefits.

And consider that the media does a great job in painting half of the picture. The other half is filled with the complete absence of articles related to US companies trying to steal secrets. The plausible explanation is that the media is heavily biased (in al honesty it's not like they could publish articles on active operations). In this context does it seem fair to judge having only half the information?

Whats arbitrary about pointing out a string of high profile reports regarding Huawei attempting to steal trade secrets? I don't see the same reports coming out of US companies in regards to Chinese companies.

> Whats arbitrary

The decision to put Iran on a ban list, as I wrote very clearly.

> reports regarding Huawei attempting to steal trade secrets

This is called a "national security" issue but doesn't appear to be treated as such. Which means the threat is not taken seriously or there is no threat.

> I don't see the same reports coming out of US companies in regards to Chinese companies

I'm not sure I understand. The reports about Huawei came from US companies. But in case you were wondering why there are no reports about US companies attempting to steal trade secrets there are 2 possible answers: 1) US companies never spy or 2) US media never talks about it. Which sounds more plausible to you? And if it's 2) I'll ask again: is it fair to form an opinion having only half of the picture?

"Tappy" again, plus a story from Bloomberg, which has shattered its reputation when it comes to technology/security reporting (SuperMicro, among other false articles whipping up anti-Chinese hysteria).

Ok, how about NPR:

https://www.npr.org/2019/06/08/730898429/huawei-accused-of-t...

Or ZDNET:

https://www.zdnet.com/article/huawei-regularly-tried-to-stea...

or CNBC:

https://www.cnbc.com/2019/05/22/huawei-executive-accusetd-of...

The NPR segment is just repeating what Bloomberg said.

The ZDNET story is about a Huawei engineer ... wait for it ... asking someone at Foxconn about what components Apple uses in its smart watches.

The CNBC story is about Huawei and a tech company founded by one of their former employees suing one another, each alleging the other stole trade secrets. These sorts of disputes happen all the time.

You'll notice that nowadays, there's tons of reporting on every minor IP dispute or claim involving Huawei. At the same time, and I'm sure this is completely coincidental, the United States government is trying to paint Huawei as a "bad actor," so that it can be used as a hostage in the trade war.

Curious where I implied Huawei didn't before (merely stated the enforcement for patents, or "rules" as you put it, no longer apply because the negatives are already in place against them).

True.

If we're all just gonna take our marbles and go home, I guess there's nothing really stopping any of us from just making our own marbles.

Still pretty dumb to make your own marbles though. But I get your point there.

GPUs have sort of plateaued except for the raytracing component. There are quite a few vendors, Mali from ARM, Adreno from Qualcomm, PowerVR from Imagination, NVIDIA's GeForce, AMD's Radeon, Intel's integrated stuff, and then there is a few more smaller players (what Matrox, Broadcom, etc.)

I think there is room for an open source straightforward GPU based on the standard principles. Although maybe there is patent licensing between all the parties that I am not aware of.

> GPUs have sort of plateaued except for the raytracing component

What? As far as I understood it, CPUs had plateaued outside of efficiencies, GPU power still follows Moore's. Generational benchmarks show big leaps in every new arch on userbenchmark, whereas CPUs from 7 or 8 years ago are only marginally slower (SC) than modern ones.

>Generational benchmarks show big leaps in every new arch on userbenchmark,

That is because GPU scales very well with Transistor count, throw in more transistor and memory bandwidth and you are good. On the CPU you could have doubled the transistor budget and get marginal improvement in IPC.

To expand on that, this is my understanding: process shrinks used to lead to faster clocks AND more transistors, but a wall was hit and since then they only lead to more transistors. GPU's can use those extra transistors because their workload is embarrassingly parallel (thousands of parallel things), but CPU's can't because their workload is mostly sequential (developers have to explicitly write code for every thread to do something useful, and find this hard to do). You can see the clock rates peaking with the Pentium 4, crashing when that design proved a dead end, and then plateauing: https://i.stack.imgur.com/z94Of.png

IPC improvements are still happening, which is why Intel has been on 14nm for years and still makes CPU's every year that are a little bit faster, but they are hard to do and you only get incremental improvements.

If someone could figure out a way to write regular software that easily parallelized to hundreds of cores, CPU's could also get the same leaps in performance that GPU's have been enjoying. It's a software problem more than it is a hardware problem.

>If someone could figure out a way to write regular software that easily parallelized to hundreds of cores, CPU's could also get the same leaps in performance that GPU's have been enjoying. It's a software problem more than it is a hardware problem.

Amdahl's law [1] , and while we have demonstrate parallelised code can speed things up, generally speaking Software performance still have lots of low hanging fruit without even going deep into parallel and concurrency world, it is just a matter of cost whether it is worthwhile to improve or fix it.

[1] https://en.wikipedia.org/wiki/Amdahl%27s_law

Why can't a GPU power a computer?

I'm not an expert and might be horribly wrong and welcome edification, but I believe you don't want your GPU powering your computer because

- GPUs don't handle I/O other than writing to video RAM.

- GPUs don't handle interrupts.

- GPUs don't handle branching well.

For the first two, a lot of infrastructure that is part of the chipset and platform would have to be extended to each compute unit of the GPU.

Imagine a database on a GPU. It's not like 1.5K-2k+ cores that are super good at math can read and write your disk or disk array at once.

Same reason we have OLAP and OLTP systems. Same reason it makes sense for some use cases to set up a Hadoop cluster, and for others to have a single but fast CPU: some tasks can be run in parallel, others can't.

See Amdahl's law : https://en.wikipedia.org/wiki/Amdahl%27s_law

You can calculate the value of each screen pixel without knowing the value of the others. To calculate the n value of a fibonnaci sequence you have to calculate the n-1 and n-2 values first, so it doesn't matter how many cores do you have. Gpus are for the former and CPUs for the latest.

Nitpick: I understand your point in general, but it's not actually true that you need to calculate the n-1 and n-2 values first for Fibonacci.

You can use Binet's formula or matrix exponentiation to calculate it without such a dependency.

GPUs are made to execute a limited set of the same operations operation on on a huge amount of data in parallel. This is a totally different workload than your usual computer programs, that commonly have a long and complicated series of commands. Thus just translating your program 1:1 to your GPU computer would make it way slower. Your GPU runs probably around ~1.5GHz, a third of your CPUs. The GPU does have a few thousand cores that run in parallel while our consumer CPUs does have at best a dozen (that are more capable than any single GPU core), but most of those are often already idle, since it's a lot of work to make your software take advantage of them. For some tasks it's worth it, and those take advantage of e.g. you GPU.

The biggest issue is lack of huge caches, and the extreme latency to main memory due to the speed of light.

GPU’s are fine or even better for a wide range of compute tasks see CUDA etc, they just end up being very slow at a subset of common tasks.

Circuits don't operate at speed of light, though. People like to say this over and over again, but electron mobility through a circuit is only around 2/3s the speed of light, IIRC. They may have very tiny mass, but they still have mass.

No, it's the electromagnetic field which propagates at around 2/3c in the conductor. The electron drift velocity is on the order of a fraction of a mm/s.

It is the mass in the electrons that creates the heat.

Can you explain, why CPU's do not scale so well?

Because GPUs always run trivially parallelizable workloads and CPUs usually don't.

I am not an expert but I will take a shot. CPU speed is in part determined by the clock. Faster clock, more data processed per second. That's the simplest understanding. Now for modern CPUs, we use a pipeline architecture. So just like how a compiler like LLVM has a tons of optimizations, the same can be said for CPUs. Your CPU doesn't quite execute instructions one by one like a hardware Turing machine. It consumes a large chunk of instructions, try to figure out what's needed and what's not, and then process it. This is all done the few pico seconds of each cycle. It's less like a car manufacturing line and more like a fruit sorting machine, different operations gets executed simultaneously if the processor can prove that they are not e.g. required to be in order. A CPU with more transistors can easily process more optimizations i.e. bigger pipeline. You can also throw in other stuff like caches to squeeze out even more performance. The other side of Moore's law is economics, essentially cost of computing gets halved, more processing power for half the cost. Same chip, more power. Now for CPUs, clock speed has pretty much plateaued in the last decade. It is increasing, but slowly. At the same time, there are limits to pipelining, and if you get too eager with optimizations, you end up introducing vulnerabilities (since this is HN, I think I don't need to cite examples, there have been plenty). We are approaching the limit of transistor scaling and the clock isn't magically oscillating faster so there is now greater urgency on the architecture side. Current software is unable to take full advantage of multi-core processors or parallelism to the extent of making Moore's law great again. (the last time Intel tried to launch a CPU with emphasis on parallelism didn't really work out, look up Itanium). It shifts a lot of work to the compiler end and makes life unnecessarily difficult for compiler authors.

Now GPUs are designed for running multiple computations simultaneously, like CPUs, there's the fact that most operations cannot be trivially parallized. Matrices and linear algebra can because of obvious reasons — large grid of numbers, easier to process at once than with a for-loop. Most problems are not like this. The next step for computing may be FPGAs, digital circuits that can be reconfigured on-the-fly. Certain new Intel CPUs have this and most processors have a basic version in the form of microcode. Whether this will work out, I do not know. FPGAs is too more compatible with trivially parallized tasks than your usual CPU based workload. Theoretically in future you can perhaps spend 20min compiling Minecraft or Factorio to FPGA bitcode during the installation process, though more likely it is only the processor intensive bits that gets optimized. When you load the game, part of the chip gets reconfigured for it. I believe this will benefit long running processes more than short ones.

Now I am no subject domain expert so somebody with more insider experience should elaborate on this.

Number of transistors/$ is about to reach a plateau soon (if it hasn’t already.)

More powerful GPUs will happen as Moore’s Law continues for a few more years, but the free for all will soon be over.

This is true for the whole semiconductor industry, and the consequences will be very interesting, not necessarily in a good way.

There is plenty of work being done on RISC-V vector computing and GPU equivalents. E.g.:

https://www.phoronix.com/scan.php?page=news_item&px=Libre-RI...

As someone who looked hard into it, RISC-V doesn't really make sense for a GPU shaders, IMO. You'd have to change so many things about how the ISA works that you wouldn't get a benefit to using a common ISA. Stuff like how shaders are batched and dispatched, textures accesses make sense to be their own instructions, you want a larger register file on the scalar side, the memory barrier semantics are much more complex, etc.

RISC-V might make sense inside of the GPU accelerators though. GFX command list processing, DMA engines, and video codecs all make sense to be something like RISC-V.

Most of the GPU workloads these days are not graphics related. They are general compute vector machines. The RISC-V vector extension is based on Cray's ISA and meant to compete in this space.

Nvidia are supposed to be using RISC-V inside future GPUs.

Yes, as a replacement for their internal proprietary logic controllers which operate in the periphery.

That has nothing to do with the RISC-V ISA being the right choice for core functionality like shader cores.

AFAIU their RISC-V variant can also be proprietary, akin to FreeBSD based products such as Junyper or macOS

To replace their falcon cores in the places I talked about in my second paragraph.

I was just trying to point out that Nvidia seems to be a bit more positive about the idea than "it might make sense".

There is an okish Chinese GPU maker. We'll have to see if they actually deliver though. But with more funding they may be competitive.

https://www.extremetech.com/computing/297099-chinese-vendor-...

>Military ties, Chinese listed company

That how you know it's nothing but a marketing campaign to boost their stock price.

Nothing about RISC-V would cause any more difficulty interfacing with GPUs. Any GPU you can license piecemeal right now, can also be connected to a RISC-V complex.

Presumably if Huawei can't license an ARM core before of trade restrictions, they're not going to be able to license a GPU because of similar restrictions.

Maybe Jingjia. Most GPU IP vendors are U.S, Korean, or broadly Western; so I can imagine it'd be difficult. It's not like they're having a more difficult time as a result of the host ISA. That's my main point.

ARM isn't going to license their GPUs to a competitor.

AMD just announced a major license of their GPU tech to Samsung. Nvidia, Qualcomm, and other vendors have complete designs with their own GPUs. Vivante and Imagination would surely be glad to license their tech to smaller players.

From hardware POV GPUs these days are to a close approximation just multicore CPUs with more focus on parallel throughput instead of single core performence, and a few specialized graphics op instructions (like texture lookup) that are unlikely to be badly patent encumbered due to their age.

If they can only sell their phones in China, they will go ahead and make their own GPUs since they can violate US patents all they want in their own market. That will just force them to develop good GPUs.

Huawei is still a largely a networking equipment company, the consumer branch of it is growing very fast, but not essential, as per their CEO.

I was in Beijing at their press conf last January when Richard Yu proudly announced that in 2018 the Consumer Business Group had a record revenue of fifty billion US$, making as much as the infrastructure division for the first time ever. Sure Ren Zhengfei is ready to throw the CBG under the bus if needed (but it won’t happen, as it is sustainable even as a China-only business). Still the consumer division is way past being just a small part of the company.

It's not clear to me that you can't make an open source cpu with a chip like risc-v

Let's be real, the Chinese have been stealing technology right and left for years. Losses number between $225 and $600 billion per year [0]. One in five companies has had its technology stolen [1]. They will just steal the tech. Or maybe they'll decide it's easier to work with AMD (they're already doing it on the CPU side [2]).

This is not popular to say, but it is accurate and my statements are sourced. Feel free to provide rebutting evidence if you disagree.

[0]: https://money.cnn.com/2018/03/23/technology/china-us-trump-t...

[1]: https://www.cnbc.com/2019/02/28/1-in-5-companies-say-china-s...

[2]: https://en.wikipedia.org/wiki/AMD%E2%80%93Chinese_joint_vent...

The methodology in the US study seem to suggest that Chinese are inable to do anything on their own and any technology that US company have that somehow China also got, is ALL counted as stolen.

That's a very fun verdict

Especially since China is the number one scientific country in terms of number of published papers and e.g are the leaders in nuclear fission and AI.

The losses numbers are from a US government investigation. So they might have some bias ... and therefore not really mean anything.

I'm surprised that it's only 1 in 5 companies. Based on prior client experiences with Chinese contractors, I would have expected it to be 5 of 5 foreign companies doing business in China.

I would assume that it's due to a limited number of companies having useful IP to steal. I have had similar experience with several companies, and hear similar stats from those who work with other companies, but that is anecdotal. I would also hazard a guess that many companies don't want to admit it, even if it does happen.

What you say may be accurate, but it's also extremely biased by a paradigm of strict imaginary property... which isn't necessarily useful for understanding how things are going to play out.

You can't steal an open source ISA .... unless you don't follow the license (which in this case is trivial to do)