[modeless]

Seems silly to me. CPU designs are important, but these companies have more than enough engineers to make competent designs even with some people leaving. There's another factor that completely dominates. It's all about the fabs. Intel lost the performance lead, was it because of their designs? No, it's because they lost the lead in fabs. AMD passed Intel, was it because of their designs? No, it's because they use TSMC's fabs and TSMC passed Intel. Apple blew everyone away with M1, was it because of their designs? No, it's because they paid TSMC boatloads of money for exclusivity on their latest fabs. Apple M2 disappoints on CPU performance, is it because of their designs? No, it's because TSMC's next fab isn't ready yet so they're still using the same fabs as M1.

These days I care more about which TSMC process node my chips came from than which company designed them. I need a new computer but I'm waiting until next year because there will be a wave of new CPUs and GPUs coming out with much better performance. Better designs? Maybe a little, but it's really because they're all moving to TSMC N4.

I really hope Pat Gelsinger can save Intel's fab business because we really need another company that can compete in fabs and Samsung isn't doing too hot either.

[inkyoto]

> No, it's because they lost the lead in fabs. AMD passed Intel, was it because of their designs? No, it's because they use TSMC's fabs and TSMC passed Intel. Apple blew everyone away with M1, was it because of their designs? No, it's because they paid TSMC boatloads of money for exclusivity on their latest fabs.

The fixation on the fab process is bewildering. Yes, it does help, but it is also an optimisation step that is decoupled from and that bears no relevance on the chip design. Yes, the smaller node size also brings the increased density along and an increased number of things that can be whacked into the same sized piece of silicon, but it will not magically improve the overall system performance or result in the linear architecture scalability.

The article is specifically calling out a potentially decreased ROB size in M2 cores, and ARMv9 also potentially not arriving until M3 which are crucial to the speed or software performance. There is absolutely nothing the fab process can do to make SVE2 and matrix instructions automagically appear in lithographic chip designs – those are the «silicon» design time decisions. As we have recently been seeing more and more practical, mainstream use cases of the advanced use of the SIMD instructions at the C/C++/Rust runtime level that bring an order of magnitude level performance gains, having the SVE2 implementation at the ISA level is becoming somewhat critical.

[modeless]

Stuff like adding SVE2 can be great for specific applications but it's really marginal when looking at whole system performance. What's not marginal are the improvements in power efficiency and room for more cache that come with new process nodes. These chips are power constrained in almost everything they do, because of heat dissipation or battery life or both. Less power and more cache benefits everything automatically, not just the very few things that actually start using new SIMD instructions or other new hardware blocks each year.

[inkyoto]

> Stuff like SVE2 is really marginal when looking at whole system performance.

It is not. A recent paper (https://arxiv.org/pdf/2205.05982.pdf) from Google engineering has compared performance of a vectorised (SIMD) vs non-vectorised implementation of the quick sort in the Highway library as well as the performance difference of the AVX-512 vs NEON/SVE1 implementations. By switching to the SIMD processing alone, the 9-19x speedup has been reported, depending on the SIMD unit size (32/64/128-bit numbers have been sampled and measured up). Even the smallest of the two, the 9x perfomance gain factor, is far from being marginal.

On the SIMD unit size of things, the performance difference between AVX-512 (the average of 1120 Mb/sec has been measured) and NEON implementation (the 478 Mb/sec throughput on average) is 2.4x smaller for NEON/SVE1 largely due to the smaller width of the units of processing. Again, the 2.4x factor is not in the marginal territory.

> What's not marginal is the improvements in power efficiency that come with new process nodes.

And that is an optimisation step, albeit a very important one. However, it will not make a quick sort implementation run 2.4x faster alone.

[modeless]

You completely ignored the "whole system performance" part of my statement. What percentage of your CPU time is spent running SIMD-optimized implementations of Quicksort? Now apply Amdahl's law.

[inkyoto]

«Whole system performance» is a meaningless term as it is a function of many, usually poorly controlled, input variables, and your whole system is different from my whole system. If my VPN tunnel allows me to have faster transfer speeds simply by virtue of having ISA assisted optimisations in the cryptographic library it uses, the net result will be very noticeable to me but perhaps not for you unless you also have to use the same VPN client.

Even the web browser you are using right now to comment on HN likely makes use of the very same Highway library (Chrome and Firefox certainly do, unsure about Safari) the speedup gains have been reported for. The «overall» browser performance will also improve as the result due to it receiving gains transparently, by simply dropping an optimised implementation into the browser build.

[modeless]

The idea that my overall browser performance might improve in a noticeable way because it is switching from NEON optimized Quicksort to SVE2 optimized Quicksort is simply laughable. On the other hand, switching to a processor fabbed on a better process node could easily have a noticeable impact on overall browser performance, or battery life while browsing, or both.

[brigade]

Soo... basically a 2x speedup in going from 4x128b to 2x512b ALUs, after discounting the frequency difference. But realistically, Intel's client configurations are 3x256b, which is only 25-40% faster in that paper.

(I suspect any application doing enough quicksort that the 2x speedup is significant, would be even happier going slightly off-core to a coprocessor more specialized in vector processing, like Hwacha. There's plenty of space between "tightly-coupled CPU SIMD" and "GPU" that I think makes more sense than needing to implement 512-bit registers in little cores.)

[inkyoto]

> Soo... basically a 2x speedup in going from 4x128b to 2x512b ALUs, after discounting the frequency difference. But realistically, Intel's client configurations are 3x256b, which is only 25-40% faster in that paper.

2.4x difference was, in fact, reported, however I still find it somewhat difficult to interpret the reported results. The processing unit size difference alone and the number of LU's can't account for such a big difference in transfer speeds as the M1 Max that was used in the assessment has a very wide memory bus (256 bit wide for a performance core cluster or 512 bit wide for the entire SoC) as well as unusually large L1-D cache and a large L2 cache, with both caches having deep TLB's. The test set they used could also fully fit into the L2 cache. I have asked the Google engineer a question in a separate thread about what else could influence the observed performance difference but have not received a satisfactory explanation.

[janwas]

I am that engineer; happy to clarify/go into more detail. First, just to ensure we're on the same page: the speeds we quote are sort goodput, i.e. amount of sorted data produced per time. Memory bus is only incidental to keeping the vector units fed.

The key bottleneck is partitioning. AVX-512 does really well there because it has dedicated compressstore instructions, and it's actually even faster to partition a vector via vperm* (because we only need to do that once, whereas two compressstore are required to partition). So AVX-512 reaches >25 GB/s partition throughput per core; it's instead limited by the memory bandwidth each core can access (around 11 GB/s if a single core is active, less when all are competing for the total "128 GB/s").

By contrast, NEON for example in the M1 has 128-bit vectors. Its "4 vector units" (even if they can actually execute all instructions concurrently, which is not clear to me and unlikely - Intel can also only execute some instructions on certain ports) are definitely not as good as actual 512 bit vectors, because partitioning only has a left and right side, and we don't have enough ILP for each of those to keep 2 vector units busy. Hence NEON reaches 11 GB/s partition throughput. It would seem like this matches Skylake, but no: once a subarray fits into cache, Skylake is freed from the memory bottleneck and is at least twice as fast there (which is a sizable fraction of the total sort time).

Does this help explain the results?

> The test set they used could also fully fit into the L2 cache.

This seems unlikely because we're sorting 8 MB and my understanding is that cores (unless L2==LLC) generally have private, partitioned L2 caches, so 3 MB in the case of M1. Is that incorrect?

[brigade]

If something is bound by memory/cache bandwidth, then increasing ALU width wouldn't help in the first place.

[janwas]

This is an interesting question. Has there been much uptake of such accelerators/coprocessors? One concern is that by the time the HW is ready, SW wants to do something different, perhaps fusing some other step with the sort. Another is deployability: everyone has SIMD/vectors on board, but even GPUs aren't quite everywhere nor so easy to scale out.

Also, there are now several RISC-V CPUs with 512-bit vectors, and it seems fair to call them little cores especially compared to x86 and M1/M2. Perhaps 512-bit is more feasible (and sensible) than is widely believed?

[janwas]

> adding SVE2 can be great for specific applications but it's really marginal when looking at whole system performance. What's not marginal are the improvements in power efficiency

Depends on the applications, I suppose. But did you know that (at least on OoO x86), the energy cost of scheduling an instruction dwarfs that of the actual computation? That is why SIMD, including SVE2, can be so important - it amortizes that cost over several elements. Let's spend (more of) our energy budget on actual work.

Is it really just "very few things that actually start using new SIMD"? I'm not a huge fan of autovectorization, but even that is able to vectorize some fraction of STL algorithms. And there are several widely used libraries, including image/video codecs and encryption, that use SIMD and wouldn't be feasible otherwise.

[brigade]

Android flagships are shipping with SVE2 as of this year, which I actually didn't realize until like two weeks ago because there's been nearly zero buzz about it. What's SVE2 being used for over NEON as of now?

[inkyoto]

Low level runtime optimisation that yields substantial performance gains in the user facing or system level software, ranging from cryptography through to data processing algorithms and very high throughput JSON parsing.

Take OpenSSL as an isolated example. By simply fiddling with the C compiler flags to allow it to use NEON on M1, the sha256 calculation speed-up is 4x for 128 and 256 block sizes, with performance gains quickly tapering off for larger block sizes and resutling in a modest 10% increase only. And that performance increase happens without the involvement of hash functions having been manually optimised for NEON/SVE1.

SVE2 with its variable vector size support could improve performance for larger unit sizes. Perhaps it is the time to spin up a Graviton3 instance and poke around with clang/gcc to see how actually good or faster the SVE2 is.

[brigade]

Yeah that's NEON. And there's instructions that literally calculate SHA256 so generalizing that is moot. My point was first, what real benchmarks are there of SVE2's benefits over NEON with mainstream CPUs that M2 would compete against? Unlike AVX-512, NEON was already pretty rich, so the new instructions have rather specialized usefulness.

Because outside of servers where little cores don't exist, 256b ALUs in big cores mean 256b registers in little cores, and Cortex-A510 is way smaller than Gracemont. And then you're giving Samsung another opportunity to screw up big.LITTLE...

And even the server CPUs with SVE are 2x256b except A64FX which is HPC exclusive, so no better than 4x128b.

[adrian_b]

SVE2 does not increase the maximum speed. That depends only on the width and number of the ALUs, on the number of cores and on the clock frequency.

The purpose of SVE2 is to simplify the writing of the software that exploits the data parallelism, both when that is done manually and when that is done automatically by an autovectorizing compiler.

With SVE2 it should become much easier to deal with data structures where the sizes and the alignments are not multiples of the ALU width and it will also no longer be necessary to write many alternative code paths, to take advantage of any future better CPUs, like when optimizing for Intel SSE/AVX/AVX2/AVX-512.

There are still a majority of programs that do not utilize as frequently as possible the existing SIMD units. With SVE2, their number should diminish.

[ezconnect]

Fab process is very important. The fast design is nothing if you can't build it.

[Sakos]

Key talent is incredibly important. If you lose enough senior engineers, it doesn't matter how talented the rest are. You've lost so much institutional knowledge that is either extremely difficult or impossible to regain. And Apple is notoriously under-staffed for a lot of their projects. With the staffing losses to Nuvia, I wouldn't be surprised if they lost enough key talent that it's going to take them a long time to recover and be able to deliver significant performance improvements again. That's what happens when you treat software developers/engineers like commodities.

[iosystem]

I thought AMD succeeded because they managed to get Jim Keller and other great engineers? Unsure why you're placing your hope on a CEO.

[wmf]

No one person deserves the credit for a processor and if Intel had delivered on their roadmap (Ice Lake in 2017 and Alder Lake in 2019) AMD would be dead today.

[BackBlast]

AMD survived bulldozer. They survived not because of the quality of their chips, but because there exists a sizable x86 market that is literally anything except Intel.

[hn_go_brrrrr]

Why does that market exist?

[BackBlast]

I would expect a large part of it to come from personal dislike in dealing with Intel by people in the position to pick parts and make the sourcing decisions.

[ChuckNorris89]

Because, before the mobile explosion, we had the PC industry explosion which created a huge demand for X86 PCs and Macs in every home.

And even after the mobile revolution shrank the demand for X86 PCs, the cloud revolution further entrched X86 in the cloud.