15 ms·
x86 is dead, long live x86
- harpratap 3y agoWe did a migration from GCP's Intel based E2 instances to AMD's T2D instances and saw huge 30% savings in overall compute! It is similar amount of savings folks got from switching to AWS Graviton instances, so looks like AMD might keep the x86 ISA still alive
- jeffbee 3y agoE2 is just extremely old. E2 are a mixed fleet that contain CPUs as old as Haswell. Haswell launched over 10 years ago. It makes sense that you get a better bang for your buck from using something that isn't gravely obsolete. You should also keep that in mind when benchmarking E2, since it's a grab-bag of CPUs you need to control for the actual CPU type, either by specifying the minimum one you require or assuming the worst.
- harpratap 3y agoThat is true. But in cloud our hands our tied, we cannot really switch from one one generation to another so easily. GCP has so far never launched a new chip while keeping the price same or lower. We did same cost/perf analysis on newer generation chips like Ice Lake from Intel and even Milan from AMD in the form of N2D, but both are quiet expensive for the performance uplift they provide. The unique thing about T2D is that the price is competitive with 10 year old E2, which has been the case only for ARM based CPUs like Graviton and Ampere Altra (T2A from GCP)
- jeffbee 3y agoI get what you mean. For the record, here's a GCP engineer holding forth on price/performance ratios who also concludes that T2D is optimal. https://medium.com/google-cloud/google-cloud-platform-cpu-performance-in-the-eyes-of-spec-cpu-2017-part-3-a88642f70ed#:~:text=One%20picture%20worth%20a%20thousand%20words%20%E2%80%94%20Price%20to%20Performance https://medium.com/google-cloud/google-cloud-platform-cpu-pe... ETA: Since you are using Go and targeting a specific modern CPU, you may also get a measurable benefit from setting GOAMD64=v3, so the go compiler generates code using AVX2, BMI2, LZCNT, etc.
- harpratap 3y agoThanks, that's a really insightful article! > Since you are using Go and targeting a specific modern CPU, you may also get a measurable benefit from setting GOAMD64=v3 That's actually a long pending open issue in our backlog :)
- mobilio 3y agoI've recently switch from VPS with Intel to physical server with latest AMD Zen4. Single thread performance blow my mind with scores like 4000. Without change a single line of code = performance was 10x than before.
- bombcar 3y agoIt's really important to benchmark this stuff, because depending on the workload, a cheap VPS is the way to go (you're almost always idle, say) or getting some of the latest and greatest hardware under your desk can quickly pay for itself.
- jsheard 3y ago4000 in what benchmark? I don't doubt that bare-metal Zen4 is fast but that number doesn't mean anything by itself.
- mobilio 3y agoI'm checking PassMark: https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+7600&id=5172 https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+7600&id...
- aidenn0 3y agoYeah, I read that and just saw Vegeta saying "it's over 8000"
- jeffmcjunkin 3y agoAhem, that's _9000_.
- retskrad 3y agox86 lives on desktop now. Windows, Intel, AMD and Nvidia have more or less thrown in the white flag and given the laptop market to Apple.
- jimmaswell 3y agoBased on what? I would never buy an Apple laptop and no one I know has one either. The Windows "gaming laptop" I got a few years ago for game dev is perfectly adequate and has features missing from Macbooks: - ethernet port - hdmi port - multiple usb A and C ports - solid keyboard Same for my work laptop, again Windows/x86 and no one I know with a work laptop is supplied a Mac either.
- ezfe 3y agoMacBook Pros have HDMI ports and many USB-C ports. I also don't know any issue with MacBook keyboards except the 2016-2019 model years. I work for a large financial services company and receive a MacBook Pro for my work. My work laptop battery can handle all-day IntelliJ & web browsing without charging, I've never heard the fan and the laptop stays cool. It has a Geekbench multicore score of 13,649. My personal laptop also lasts that long and stays cool, and doesn't even have a fan.
- edgyquant 3y agoWe live in different worlds. All engineers I know have Macs and the ones that don’t have Linux. The one guy on my team who uses windows is a constant problem as he has to find work arounds for every process we have/service we build.
- redwall_hp 3y agoYep. I work in an entire office of engineers with only Macs (mostly ARM ones now) and I only have a Mac at home, besides a very old ThinkPad running Ubuntu. (Aside from FFXIV on MacOS, I've moved any gaming to PS5 and Switch.) I haven't actively used Windows since 2008. It's such an obnoxious OS to use, and I have never seen a Windows laptop with quality in the same zip code as a Mac. There's always a terrible touch pad, keyboard flex, plasticky trash fit and finish, mediocre display or something that just ruins the whole thing.
- netr0ute 3y agoWhy do these kinds of articles essentially refuse to admit RISC-V exists?
- xcdzvyn 3y agoBecause it essentially doesn't? At least when I tried looking a few months ago, it was really hard to find any commercially available RISC-V SoCs, let alone high performance ones. Sure there's little hobbyist boards going for $200>, but that's about it.
- jsheard 3y agoProbably because the article is looking at silicon that's actually available today, and big-iron RISC-V silicon akin to Xeon/EPYC/Graviton/Ampere is more of a hypothetical at this point.
- harpratap 3y agoCloud world is really slow. Imagine writing about Zen3 Milan in Aug 2023 when AMD has already announced Zen4c Bergamo. Actually we were "ahead of the curve" since we got access to T2D before it was made publicly available in Tokyo region, and even then it took several months to get enough capacity in Tokyo to fully migrate our production Kubernetes cluster. I really wish we could test out RISC-V SoCs from the likes of tenstorrent, but it's a long journey
- re-thc 3y ago> Cloud world is really slow. Imagine writing about Zen3 Milan in Aug 2023 when AMD has already announced Zen4c Bergamo. AWS has Zen 4 in preview. Azure and Oracle have Zen 4 available. It's only Google cloud that's been behind this release. The cloud world isn't slow.
- harpratap 3y agoIt takes very long time for going from preview to actual production usage for anyone. We had T2D preview access more than a year ago, it took several months to get enough stock in Tokyo region (US always gets preference in such cases whenever a new machine type comes out). GCP already has Zen4 in preview for some US customers. Also, us being one of the largest GCP customers in Japan made things even slower
- slfnflctd 3y agoWhen I think about how long chips like the 6502 have still been in active use (almost 50 years now), it is hard to conceive of a world where there isn't a significant presence of x86 activity for the rest of my life. The majority of 'the market' may go elsewhere, but for a gazillion reasons, x86 will not be disappearing for quite a while. At this point it would honestly surprise me if we didn't at least have high quality emulation available until the end of the human race as we know it. Sure, we've probably lost most of the software ever written on it, but a whole lot of interesting artifacts from a key transition point for our species still remain locked up in this architecture.
- babypuncher 3y agoAs we learned from Terminator 2, the machines that eventually rise up to eradicate humanity will still be running some kind of 6502 derivative.
- riffic 3y agoA thing's future longevity can sometimes be predicted by how long it's already been around.
- _a_a_a_ 3y agoBut not humans, regrettably.
- User23 3y ago99 year old humans are considerably more likely to live to be 100 than 50 year old humans.
- Dylan16807 3y ago"future longevity" in this case means how many years are left, not how many years total.
- wongarsu 3y ago
- amusingimpala75 3y agoI would just like to point out that that is not how the saying goes. When Queen Elizabeth died it would have been “the queen is dead, long live the king”, as the dead thing is the predecessor and the living thing is the one that follows it
- harpratap 3y agoThe dead thing is Intel (x86) and the successor is AMD (x86) as opposed to ARM in our case. Isn't it correct?
- re-thc 3y ago> Isn't it correct? No. Intel isn't dead. They may be behind (for now) but they're definitely catching up and have in a way on the desktop. It's not certain that Intel will die and AMD will for sure win. Competition is great.
- saghm 3y agoIf we're going to be that pedantic, we might as well just say that because Intel and AMD aren't living entities so they can't be "dead". Or we could just accept that the original meaning of the phrase has evolved and is flexible enough to be used in situations where it's not absolutely literal.
- Teever 3y agoI just closed another post on HN because the top thread descended into some bullshit semantics pissing contest so I come here and see the same. Is it just me or has anyone else noticed an increase in arguing over the meaning of words on HN lately?
- lcnPylGDnU4H9OF 3y agoI'd guess it's a timing thing. This site likes to prioritize more highly the "current" discussions which are generating upvotes as long as they're not also generating a ton of downvotes.
- WeylandYutani 3y ago"The macOS (OS X) version of this game does not work on macOS Catalina (version 10.15) or later due to the removal of support for 32-bit-only apps" And that's why x86 is good.
- rideontime 3y agoI'm failing to see the connection, could you please elaborate?
- 7jjjjjjj 3y ago[dead]
- singhrac 3y agoAs a separate data point, I briefly switched one of our servers from an r6a.4xlarge (AMD Epyc) to a r6i.4xlarge (Intel Xeon) and saw a 30% speedup in our number-heavy compute task. I would love to find out why (MKL or AVX512? Do I need to recompile numpy?), but for the time being it pays to stay on Xeon. We eventually switched to m-instances since that fits our compute/memory usage better when we’re at limits.
- re-thc 3y agoIf it's AVX512, you'll be excited to know that m7a, currently in preview has great support for this. AMD Epyc 4th gen i.e. genoa is a lot faster.
- harpratap 3y agoYes, I point this out in the article too. Which CPU will perform better is heavily dependent on your workloads, so I refrain from relying 100% on synthetic benchmarks and directly ran canaries in production instead. It's definitely possible Ice Lake is superior for your workload than Milan
- satanicpanic 3y ago[flagged]
- m4r1k 3y agoIf you wanna have additional insights, head over my Medium ports with SPEC CPU 2017 results. https://medium.com/google-cloud/google-cloud-platform-cpu-performance-in-the-eyes-of-spec-cpu-2017-part-3-a88642f70ed https://medium.com/google-cloud/google-cloud-platform-cpu-pe...
- tester756 3y agox86 would be dead if ARM ISA offered like at least 20% perf boost or energy savings. But since ISA doesn't imply perf. characteristics itself, then x86 will be alive. The hard part in changing hardware's stuff is getting software to adjust.