They are responsible through licensing for huge number of cheap chips on the market. It really doesn't matter they don't sell it themselves. If they stop licensing it would have the same effect as if they werse selling them directly and they stopped selling.
ARM is hardly a "good thing". These things rise and fall almost purely by chance since the differences between ISAs are menial at best. If the stars had aligned differently this could very well have been a SuperH world. Or a MIPS world.
None of those 3 even tried to have a product for applications with low cost and low power consumption.
The only competition was from Motorola/IBM POWER (and MIPS in some niche products), but neither Motorola nor IBM ever tried to offer low cost products.
During 2000 to 2005, I have ported some products with embedded computers from POWER to ARM.
The porting was not very easy, because the ARM CPUs had much lower performance, but they were much cheaper, so the conversion was done anyway.
The ARM advantage has always been in its business model, which created a lot of competing CPU vendors, willing to give you the best price, not in any technical advantages of its CPU architecture.
The NVIDIA acquisition certainly threatens this business model, despite all contrary claims of the NVIDIA management.
Alpha and PA-RISC died before the smartphone revolution (end 90s vs mid 00s). x86-32 killed them, before AMD64 (x86-64) took off.
Laptops were all x86-32/64. There has been an UltraSPARC laptop (one AFAIK), and Apple switched away from POWER because of power inefficiency. Also before the smartphone revolution.
Intel was busy with Moblin and PowerTOP back in the 00s. This before iPhone.
MIPS and RISC-V also exist. Without ARM it would've happened but slower/later.
This is ridiculous. ARM was already pretty much well-settled-in in the PDA way before the smartphone, so it definitely didn't "single-handedly made it happen". And during the early PDA era, when they got settled in that market, it most certainly did not win due to "low power draw". Back in the day where SH(-3) PDAs where still a thing, the battery life was measured in days out of 2 AAA cells (even for ARM devices). That was destroyed due to forced backlight usage in most screen techs rather than any change in ISA.
Today MIPS is still used in network devices, SH is still used in things like optical drives, SPARC is still used on some servers, x86 is still used on many servers and desktops and mobile devices, and so on and so forth. Do you think that there is something technical that warrants the choice of these architectures for their usecases ? Do you think that there is something in SH that makes it a better ISA particularly for CD-ROM drives ? Or that there is something in SH that made it a better ISA for some videogame console generations, then something changed and the better ISA was PPC, then ARM, then x86 again ?
It's mostly arbitrary marketing decisions, which manufacturer happened to be doing well that day, and some "historical reasons"/network effects which define these choices. There's very little difference at this ISA/architectural level, and a lot in the actual design and manufacturing level, where ARM is not doing poorly but hardly shines. Apple, Intel, AMD, Qualcomm, even nVidia all do better than ARM there.
If we the wind had blown even slightly differently some day 20 years ago, we may very well be using anything else, even say Itanium in our smartphones. And before you complain that Itanium/EPIC are hardly suited for low power, remember that Transmeta made their money selling _low-power_ VLIW processors emulating x86 instructions. And actually they were lower power than their x86 competition those days. Perhaps Torvalds would still be there rather than making Linux...
Nvidia had at least two generations of Android tablets. Intel had several as well. These aren't smartphones, but they did have options for cellular modems, and there's really not much difference. I've got two tablets at home, one with a 3g modem and sim card slot (a family friend got it free from at&t). Whine I'm not aware of AMD making processors for any mobile devices, there's also nothing stopping them.
No, you can't buy either such device today, but reason they lost out had nothing to do with them being inappropriate for such use.
There were many other CPUs with the same or lower power draw.
The advantage of ARM has always been the cost. For given requirements of performance and power draw, you could find much cheaper ARM CPUs than the existing alternatives.
This happened because there were many competing ARM CPU vendors.
Motorola could give you a PowerPC CPU with the same or better performance and power draw, but they were not willing to lower their prices, falsely believing that there are no alternatives.
Therefore everybody abandoned the other CPU architectures and switched to the cheaper ARM solutions.
I'm sure many companies could do low power performant and cheap chips when high resolution small screens, lithium batteries and large capacitive touch surfaces matured enough to be harvested.
The thing is only Arm did actually designed appropriate chips at the right time and licensed those designs in the best way possible. Arm was IBM PC of the smartphone era and I don't understand how people might not appreciate that.
I hope you’re not calling ROCm a workable alternative…
it’s really not close. I love that they are trying, I just honestly feel they need to be trying with 10 times as much funding/resources behind them.
Or do you mean Apple’s Metal Compute? useful only with inconvenient (at scale) Apple hardware.
If workable CUDA alternatives were really here, I feel people would be talking about them more given how large the hate is for Nvidia on multi fronts from multiple different user groups.
That’s fair, for me it never worked with any of the hardware or software I wanted it to and I’m not enough of a GPU expert to really fix that myself, so for me and many others I found with similar problems it definitely has yet to become a workable solution, but I was probably a bit harsh if I came across like I thought it didn’t work at all, I’ve seen it demoed and I know it works, but for such a restrictive set of hardware and software that Ive never seen it in use myself outside of demonstrations.
As for Vulkan, I tend to think of CUDA at the driver level given how closely coupled it is to Nvidia’s hardware. So I wasn’t really thinking about broader cross compatible APIs like Vulkan which tend to get implemented on top of the drivers providing low level access like CUDA gives… but with the exception of Apple (because we know they will never change their mind) I’m hoping the broader industry gets behind Vulkan compute shaders enough we finally get something that delivers on the promises made when the OpenCL effort began. I want clean understandable abstractions over the top of all the different SIMD and MIMD capabilities we have these days it doesn’t have to magically compile my code down to FPGA hardware but how about actually being able to get the best matrix multiplication performance out of both my CPU or GPU or both if I have a CPU with the right kind of integrated GPU that they can efficiently enough share memory and both work together without slowing each other down when working on an embarrassingly parallel matrix multiplication task. Is this really too much to ask of software in an an era that gives us near magical JIT performance in multiple languages, and manages to build cross compile/transpilation tools that can convert entire assembly code bases into JavaScript and all the other nice things we have as software developers… it never felt like too much to me but the fate of OpenCL appears to disagree.
I never understood the dominance of CUDA. I did GPGPU more than half a decade ago, and I feel like stuff has hardly changed. There are thread groups, group shared variables, global, and group shared atomics etc. Underneath, all GPUs run the same-ish wide SIMD architecture, so I'm pretty sure most stuff is semi-performance portable.
DirectCompute, OpenCL and CUDA and probably all the others expose this same programming model. Why is CUDA so dominant?
nvidia decided they wanted to be the market leader, and spent $$$$ developing tools, training materials, and libraries (like cuDNN) which they gave away for free.
They also hire software engineers to integrate CUDA into open source projects. On the one hand, those projects can now take advantage of advances in hardware. On the other, users will have to buy NVIDIA to use it, so how open is it? Also NVIDIA will not provide any help with CI, so those costs are borne by the project – or the code isn't tested.
JAX/ XLA seem quite popular, especially since you can use them in Colab on TPUs. IIRC, there are people who have managed get jax compiled with ROCm support.
Tensorflow lets you deploy to quite a few backends.
oneAPI support in GPUArrays.jl seems to be coming along.
It's likely that there are many new designs being spun up everyday by various people, but that doesn't mean that we want to remove competition from the market.
Considering they've even gotten away with breaking the rules (custom ISA extensions), I imagine nothing ARM does could realistically cause Apple to be unable to continue manufacturing ARM-compatible chips.
I would imagine that Apple (which put money into Acorn and ARM JVs over thirty years ago and I think was the first third-party company to do so, by a very wide margin) has a much older license than anybody else.
It's not correct to say they just license the insn set. They license ready to use cpu and gpu core designs (like the Cortex family in most Android phones).
It depends on the customer. Apple and Nvidia indeed license just the ISA and build their own CPU implementation from scratch. At the other extreme, you can make a complete chip just from ARM IP - they have way more than CPU and GPU IP, they can sell you pretty much everything else, too. At the middle are chip makers taking eg ARM CPU IP but no other ARM IP and making other IP themselves or licensing from other vendors.
Well the argument is that ARM is cheap and good for consumers because it's cheap. I wanted to make it clear that you can't go to ARM and buy <cheap arm chip>. You have to go to Qualcomm or Apple or Broadcom and buy <not quite so cheap arm chip>. So that kinda impacts the argument.
Whatever Nvidia wants to do with licensing they have to combat market forces already in play. There's a market for somewhat less expensive arm chips, for one, so they'd lose that revenue. Two, if they want to royally fuck arm shops and become the only ARM vendor, well, good luck... you just killed the ISA you bought for 40 billion...
Depends. Apple does not sell chips. Apple sells a vertically integrated platform that has many components from cloud services to physical devices. The price of one component in the physical device is not that important.
