Hacker News Reader

Comment by lx3459683 | Hacker News Reader

Is this the beginning of the end for x86?

Maybe. Traditionally it's hard for companies to compete with much cheaper but slightly less performant rivals. That's the whole idea behind "Disruptive Innovation." When x86 came out it was the cheap knockoff that everybody could afford and it eventually ended up eating it's way into workstations and servers as its scale let it pour more engineering resources in. ARM may be achieving that sort of scale advantage over x86 but its fragmented nature makes the story more complex.

Probably not. Arm is power efficient but still x86 performance is ahead. At some point you need a balance between battery life and performance.

"No wireless. Less space than a nomad. Lame."

Right, but Intel CPUs are used in more than laptops.

I don't know if Intel's days are numbered or not, but I don't think this alone would be enough.

iPods could gain market share from marketing and consumer focused industrial design. ARM cores don't have the same way out.

Woah, fatal_error.

Does "Intel Inside" sound familiar? That's because it worked amazingly well, and people chose Intel over AMD back in the day just because they "felt" better about it. Some even chose it because they didn't want to get AMD shamed ("Aw poor person; they could only a afford an AMD.").

Not saying Intel CPU lineup wasn't better than AMD's product, (especially at the height of the marketing program when laptops used to have "Intel Inside" stickers), but a lot of people didn't even bother comparing. They just chose Intel-based computers.

And that's the story of how so many people I know ended up buying a PC with a crappy Celeron* (I know, I know, the new ones are better).

Sure, CPUs are harder than iPods to market to consumers. Most haven't a clue what ARM Cortex-A7 is. But, that could be the issue. They need to step up their brand identity. They also should probably tell people that ARM doesn't actually make hardware for consumers. They just design the architecture and license that design to silicon manufactures.

Most people knew they were buying an Intel. Most people don't know they bought an ARM. They could do better.

*: I assume some bought them because the salesman at Best Buy said "dude, Intel's are what you want. AMDs are slow."

Intel has ~60x the revenue of ARM and is already a known brand with the public. If it gets into a marketing slugging match, Intel wins hands down.

I'm not sure that type of marketing will happen again though. Since then, computers have gone further down the path of being an appliance. What's inside matters less than what it can do. If Apple releases a MacBook with 1-5 days of battery life, no one will care what's inside.

I agree, I don't think it will. It just doesn't make sense.

ARM is owned by Softbank, which has a net worth of ~2x Intel.

And many many more commitments. They can't dedicate the same resources that Intel can.

It is rather amazing that just reading "Intel Inside" I heard the little chimes from the TV commercials and saw it in my head.

If everything goes right for ARM, next year's CPU line should be single-core competitive against Intel's laptop line; the year after, ARM could be ahead.

Meanwhile, AMD's CPUs are winning on price/performance and pure performance for multi-core in the server market, and it looks like they're going to be competitive with Intel on basically every desktop area.

Intel has a marketing advantage. That's pretty much it.

Just built a 64 core workstation based on AMD Epyc cpus, seeing how fast my simulation workloads run on it brings a smile to my face.

I can appreciate the fun to be had with so many core cpu, but did you actually measure the performance? According to Passmark, both single and multithread performance of EPYC is very poor [1][2]. Passmark database is years in building and very informative, but I think for EPYC that is an erroneous result. Could you run Passmark benchmark on your rig to get another data point public?

[1] https://www.cpubenchmark.net/cpu.php?cpu=AMD+EPYC+7501&id=31...

[2] Compare that with performance and price of something like E5-2670 from 2012: https://www.cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-2670+...

Will give that a try tonight. Have to warn you that right now 1 of my DIMM was bad, so the system is running with only 15 DIMMS (which is an unsupported config) so the results might be suboptimal until I receive a new DIMM from the seller.

That's great, looking forward to the results. But, there's no rush. If you'd like to measure today anyway, please just make sure the result doesn't get reported to Passmark database so as not to spoil the small dataset with a biased result. Good luck with the replacement, hope it will stay solid from now on.

Just realised that Passmark does not have a Linux benchmark program, so wont be able to run it.

Ah, that's a shame. I'd love to know how much faster the Epyc is than the E5-2670. Could you try to run sysbench to get some numbers? Here are mine:

# sysbench --test=cpu run --max-requests=20000 Test execution summary: total time: 25.9014s total number of events: 20000 total time taken by event execution: 25.8983 per-request statistics: min: 1.27ms avg: 1.29ms max: 3.19ms approx. 95 percentile: 1.29ms

# sysbench --test=cpu run --max-requests=20000 --num-threads=16 Test execution summary: total time: 1.6859s total number of events: 20000 total time taken by event execution: 26.9264 per-request statistics: min: 1.26ms avg: 1.35ms max: 3.59ms approx. 95 percentile: 1.51ms

[dman@epyc ~]$ sysbench --test=cpu run --max-requests=20000 WARNING: the --test option is deprecated. You can pass a script name or path on the command line without any options. WARNING: --max-requests is deprecated, use --events instead sysbench 1.0.14 (using bundled LuaJIT 2.1.0-beta2)

Running the test with following options: Number of threads: 1 Initializing random number generator from current time

Prime numbers limit: 10000

Initializing worker threads...

Threads started!

CPU speed: events per second: 1461.82

General statistics: total time: 10.0004s total number of events: 14621

Latency (ms): min: 0.67 avg: 0.68 max: 1.76 95th percentile: 0.69 sum: 9997.72

Threads fairness: events (avg/stddev): 14621.0000/0.00 execution time (avg/stddev): 9.9977/0.00

[dman@epyc ~]$ sysbench --test=cpu run --max-requests=20000 --num-threads=128 WARNING: the --test option is deprecated. You can pass a script name or path on the command line without any options. WARNING: --num-threads is deprecated, use --threads instead WARNING: --max-requests is deprecated, use --events instead sysbench 1.0.14 (using bundled LuaJIT 2.1.0-beta2)

Running the test with following options: Number of threads: 128 Initializing random number generator from current time

Prime numbers limit: 10000

Initializing worker threads...

Threads started!

CPU speed: events per second: 47980.46

General statistics: total time: 0.4152s total number of events: 20000

Latency (ms): min: 0.68 avg: 2.06 max: 111.24 95th percentile: 3.36 sum: 41275.73

Threads fairness: events (avg/stddev): 156.2500/118.37 execution time (avg/stddev): 0.3225/0.06

Let me know if you want me to run any other benchmarks.

Wow, that's impressive. Thanks for posting this. I have a mere 8C/16T 2700X.

Thanks. I just ran v1.0.14 the same way and got 793 events/s singlethread and 10264 events/s multithread (16) on the E5-2670. So you've got single thread almost 2x faster and multithread almost 5x faster. In singlethread, that's much better than what I expected of low frequency Epycs. You've got a nice snappy machine there:) Interestingly, the performance per thread in multithread is 641 on E5-2670, while only 374 on Epyc. Probably there's some massive thermal throttling going on Epyc. With that fast cores, one should get at least 1491x64=93504 without throttling.

I think the problem size is too small. Dialling up the max requests and setting thread count to 64 yields the per thread in multi thread to 1104. (I am guessing the the scheduler will take a while to bring all 64 threads up, so on small problems one might not see the full benefit of available parallelism, but this is just an armchair hypothesis).

[dman@epyc ~]$ sysbench --test=cpu run --max-requests=2000000 --num-threads=64 WARNING: the --test option is deprecated. You can pass a script name or path on the command line without any options. WARNING: --num-threads is deprecated, use --threads instead WARNING: --max-requests is deprecated, use --events instead sysbench 1.0.14 (using bundled LuaJIT 2.1.0-beta2)

Running the test with following options: Number of threads: 64 Initializing random number generator from current time

Prime numbers limit: 10000

Initializing worker threads...

Threads started!

CPU speed: events per second: 70704.31

General statistics: total time: 10.0014s total number of events: 707263

Latency (ms): min: 0.68 avg: 0.90 max: 25.23 95th percentile: 1.50 sum: 638330.33

Threads fairness: events (avg/stddev): 11050.9844/1635.68 execution time (avg/stddev): 9.9739/0.02

That's interesting, same increase in requests makes little difference on my machine, I get 10432 for multithread (16). I noticed above you have used 64 threads on 64-core system - does that give you better result than using 128 threads? It shouldn't, SMT should give you some increase in speed. At least on my 8-core system, using 16 threads gives 10432, while using 8 threads gives only 7311.

My info may be out of date, but I think Intel also has a production advantage. AMD doesn't have the capacity to, for example, own 45% of the market and leave Intel with 55%, as I understand it.

Desktop/laptop is a shrinking market, and HEDT/Workstations have always been extremely tiny -- it's nowhere close to where Intel makes most of their money. Consumer Ryzen 7 uptake numbers or ARM laptops are basically meaningless as far as predictions about "The death of Intel" go. It's nothing. Datacenters are what matter.

This is a good move for ARM perhaps, since their licensees have (repeatedly, at least _for now_) failed to move into the server market where x86 reigns supreme, so if they can take a shrinking market off Intel's hands and make some inroads there, hey, whatever works. People actually don't care about processors, because for consumers, price is king. So if they can deliver cheaper Chromebooks or whatever, people are happy. And ARM already dominates the lower end market. But large players don't work that way.

EPYC has better pricing per-core (I say this as a very happy 1950X owner) but you're kidding yourself if large scale vendors who buy thousands of SKUs per year/quarter do anything but buy in bulk, on multi-year contracts, with extensive sales negotiations. They are far ahead as far as vendor validation/stability goes (my 1950X motherboard still has BIOS/IOMMU glitches that I'm waiting on updates for, this stuff just takes time). For the biggest customers, Intel customizes their SKUs directly to their requirements. That's a significant amount of integration with their partners that AMD is not going to match overnight. And even if they take away some of Intel's total-monopoly status in the DC, say 25%, which is a metric shitload, they've still got a hell of a lot of technology (in their foundries) to back themselves up, as well as a massive warchest. I wouldn't be surprised if Xeon margins were above 50%. You really think they can't drop some of that off and immediately tilt that ratio back around, while having tens of billions on hand for R&D anyway?

You live in a castle of sand if you think they're actually going anywhere anytime in like, the next 5-7 years. And I have many bridges to sell you, if you think "marketing" is their only advantage in this fight -- as opposed to their foundries, deep integration, near-total monopoly status in the only market that matters, their massive warchest, and huge R&D setup.

I honestly wonder if Intel actually wants people think that ARM Chromebooks are a threat to them or whatever. It means they can keep deluding themselves while Xeon sales and margins continue to skyrocket for cloud providers while everyone else chases pennies (except AMD, who are actually trying, and are absolutely not guaranteed to dominate by that alone)...

(I do hope they start feeling the pressure, of course. I'd love cheaper Xeons, personally. :)

intel is still single thread performance king and will be for the foreseeable future. It's the most important performance metric when considering a desktop processor for most workloads.

While arm is catching up, there is no gaurentee that it will actually be competitive one day, not to mention beat it. Intel is still a beast and spends more in R&D than what amd took in last year. It would be foolhardy to write intel off.

You seem to have missed the part of the article* where the Cortex-A76 matches a core i5-7300 in performance, with much lower TDP. That is planned to come... oh wait, it's already available.

* it's the second paragraph

Literally exactly what Arm is talking about in this roadmap.

You can't really look at power and performance in isolation, given that you can trade one for the other by modifying their clock speeds. It would be better to say that from the milliwatt range to 20W socket draw or so ARM is more performance and current x86 designs are more performant above 20W.

I doubt it. But competition is sorely needed.

I used an ARM Chromebook with Linux installed for about 2 years. YMMV, but VLC seemed to not work anywhere near as well as it does on the x86 (in terms of playback and codec availability).

Good battery life though.

Doubtful. Performance generally corresponds to active area and, hence, power.

So, for ARM to become as performant as x86 means they wind up burning the same area and power.

That having been said, breaking a monoculture would be welcome, especially given how cavalier Intel is about security.

(You will note I didn't say "In light of" with respect to Intel and security. IBM, DEC, etc. have been preaching about the fact that x86 has lousy security for 30+ years. It's just that nobody cared until x86 became a mainframe ... err ... cloud.)

> Performance generally corresponds to active area and, hence, power

Then how come ARM won on mobile?

Atom came too late and it wasn't enough to beat ARM. All the other low power players stagnated.

Dramatically lower performance expectations.

Low power Intel is not as good as low power ARM. "High power" ARM competitive with desktop doesn't quite exist yet, although the iPhones are very close (and better than a desktop of a few years ago)

Because an enterprising engineer managed to hand code an assembly language implementation of cellular baseband on ARM thus entrenching ARM as the low cost implementation on a cell phone. After that, ARM expanded outward to run the GUI.

In addition, the batteries of the time demanded a specific power envelope. Not many chips had this envelope ... pretty much only ARM , MIPS, and a handful of also rans that you've never heard of.

Once ARM got going, network effects took over. There is no reason you couldn't implement a cell phone on a MIPS core, for example, at this point except for network effects.

Another thing is that at the time only other ISA designer that really cared about this market was Hitachi with it's SuperH (and then founded Renesas with the sole intent of becoming meaningful ARM competitor, which obviously didn't happen).

You could get usable low power microcontrollers and almost-SoCs with ARM cores in late 90's. While there were SoCs with other 32b RISC cores they typically were intended for mains powered high performance applications ranging from DVD players to network equipment. See how large part of Freescale's PowerPC SoC lineup are without much exaggerattion "Cisco 2500 on a chip" (obviosly with ppc core instead of m68k and with wonderfully complex DMA-engine/protocol decoder/whatever-thing)

Wasn't ARM also the de-facto standard for PDAs? Which was a deliberate market profiling after Apple came into the fray and wanted a CPU for their Newton (and helped spin off the ARM we know today from Acorn).

I would imagine it comes down to margins. In a competitive market, customers aggressively try to drive down prices. Most silicon companies are too smart to play that game. You fight to establish dominance up front. If you don't win the market, you're unlikely to every become profitable, so you cut your losses and exit.

x86 has overhead in small low power systems; it didn't help that the initial intel low power products were low-effort garbage.

> So, for ARM to become as performant as x86 means they wind up burning the same area and power.

Yup. The thing is that ARM can indeed match Intel on scaling up as you noted. Intel is struggling with x86 in scaling down, though. Easy to see which one is in a better position overall.

Intel's x86 architecture reset and housecleaning is due to land in 4 or 5 years and it should be really interesting then.

yes

Reply on news.ycombinator.com