Frenetic Pony
Veteran
So... https://en.globes.co.il/en/article-uk-chip-giant-arm-developing-gpu-in-israel-1001486761
Pretty much all we've got right now
Pretty much all we've got right now
ARM Speculation Thread
- Thread starter Frenetic Pony
- Start date
-
- Tags
- gpu
raytracingfan
Newcomer
I don't think there was ever a good opportunity. Before 2016 any attempt by ARM to acquire Imagination likely would have been pre-empted by Apple outbidding ARM. During 2016 ARM was handling two acquisitions and its own acquisition by SoftBank. And the next year Canyon Bridge acquired Imagination.
Ok , then maybe Samsung should have tried to acquire the company as they are in direct competition with Apple and are now licensing tech from AMD.
Or maybe AMD should have acquired it, as Imagination to my understanding has good IP w.r.t to ray tracing , though actual h/w is almost non-existent on this front.
All in all , I found the eventual owner of Imagination to be quite weird as it was none of the estd. tech companies.
raytracingfan
Newcomer
I don't think AMD considered ray tracing to be a priority before they were informed about DXR, which was probably too late for them to grab Imagination before Canyon Bridge. Some would even argue they didn't take ray tracing seriously before the past few years. Imagination's acquisition by a Chinese private equity firm is no weirder than ARM's acquisition by a Japanese investment holding company.
They weren't "informed about DXR", they're active part of DirectX development just like NVIDIA, Qualcomm and probably many others. Of course someone had the idea first, but all of them become aware of it as soon they tell MS about their idea.
DavidGraham
Veteran
Arm is canceling a license that allowed Qualcomm to use Arm intellectual property to design chips.
They're mad because licensees make so much better cores than they can do themselves. I don't doubt for a second that if MediaTek started designing their own cores they would beat Arms cores silly in few years too. Or NVIDIA for that matter.
Never understood why Qualcomm stopped making their own earlier.
Both NVIDIA and Samsung designed their own cores and gave up because ARM caught up and were as good or better. ARM does a very good job most of the time for the specific markets that they focus on; most architectural licensees bet that they could design better cores by targeting higher-end performance at higher area/power than ARM, which was true some/most of the time, but as performance targets gradually increased over time ARM just ended up matching/beating those proprietary cores before the next generation of internal cores was ready anyway (remember ARM has 3+ CPU design teams: Cambridge, France, Austin), resulting in not a huge benefit at very high cost/risk.
BIG.little had something to do with Qualcomm giving up iirc, where they didn't have (/want to have) the resources to do both kinds of cores internally and were not (technically? legally?) able to mix & match their internal cores and ARM-designed cores.
I think both the transitions from in-order to out-of-order and 32-bit to 64-bit also had something to do with what happened in those years. ARM didn't have any in-house experience designing out-of-order cores before the Cortex-A9 and while a very impressive core for its time, I am sure there were much more experienced OoOE CPU architects from other companies that could be poached to work on in-house OoOE high-end ARM cores back then (while ARM has a lot of internal expertise as well by now). And NVIDIA bet on the Transmeta-like approach as an alternative to OoOE (which mostly failed) while there were also attempts to do better in-order cores or better implementations of existing in-order cores (e.g. Intrinsity which was acquired by Apple but I unfortunately/obviously cannot comment on the role they played later if any) instead of moving to OoOE as quickly.
On the other hand, Apple beat ARM to 64-bit/ARMv8 (plenty of articles talking about why that was) and ARM beat all of the in-house proprietary ARM cores to 64-bit/ARMv8 by a *lot* which was a disadvantage for Qualcomm versus other non-proprietary ARMv8 SoC manufacturers iirc. So no ARMv8 and no BIG.little and realising in-house cores were not necessarily an advantage if you didn't manage to be ahead by enough was surely a big part of why Qualcomm stopped when they did, but I don't know what else happened internally at that time, there might have been other factors (attrition?) at play here, it wasn't entirely clear to me what the truth was at the time despite paying attention, I think it's very hard to know for anyone who wasn't working inside the mobile CPU industry at the time.
As for the licensing nonsense... I don't know what to say except that ARM is an amazing company, but their stock price is absolutely ridiculously overpriced. NVIDIA is basically an undervalued value stock compared to the ridiculous bubble that is ARM. The *only* way their stock price makes *any* sense is if they either manage to change their licensing model to make a LOT more money per core, *or* they end up selling their own AI chips as some of the recent rumours implied (how that works and how much is that ARM vs Softbank vs Graphcore is really unclear to me, and does not inspire much confidence tbh). Their current valuation is not realistic for an semiconductor IP company irrespective of how much the market and/or AI grows. I hope they lose because I think this is actually very bad for the health of the ARM ecosystem even if it did work out for them and resulted in much higher profits, although giving RISC-V a chance to carve a bigger niche would not necessarily be a bad thing in the grand scheme of things...
Last edited:
Similar threads
