Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

techquila.co.in

Post reply on HN
Print view View original

121–130 of 160 posts

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#121

Post by chx »

Tuna-Fish wrote: post #75
chx wrote: post #58

Earlier quoted context omitted.

OP said "QS" these are grey market chips, QS means Qualification Sample, Intel sends them to various manufacturers to test with motherboards and then they leak into the grey market. They are not retail so frequency and other characteristics might differ -- but even their stability might not be as high as a retail one. Although QS is usually better at this, it's ES (Engineering Sample) which can be very dicey.

Black market, not grey market. QS are loaned on terms that explicitly prohibit resale. If you buy them, you are purchasing stolen goods.

You might be right -- https://www.intel.com/content/www/us/en/support/articles/000... says

> [...] are the sole property of Intel [...] Are not for sale or resale.

So, whoever sells ES/QS chips is selling the property of someone else without authorization to do so.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#122

Post by loeg »

rolleiflex wrote: post #98
nottorp wrote: post #95

Earlier quoted context omitted.

Pray tell, where are the docs for AMD hackintoshes? Running on Intel atm but the time for an upgrade is coming close. Do you have power management (and sleep) working on amd?

Start from https://amd-osx.com/ . > Do you have power management (and sleep) working on amd? Yes (null-power-mgmt.kext) and yes. Though mind that this is a desktop, so power mgmt. / sleep needs are much less demanding than a laptop. Specifically, the kext works by disabling power management from within OS X. However I have a kill-a-watt and the the power from the plug varies based on the CPU load, so it seems to be b…

As long as the OS uses C-states when idle (and it does) you get the vast majority of Zen power savings, even without touching P-states (clock). E.g., https://www.youtube.com/watch?v=bzxn1hS7Nq4#t=6m21s

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#123

Post by emanuensis »

eisa01 wrote: post #13
gigatexal wrote: post #2

If you need AVX/AVX2/AVX512 then go intel. Otherwise this AMD chip would be my choice for most applications.

What software is typically able to take advantage of these instructions? My use case is solving large LP/MIP problems for power markets, do those algorithms benefit?

Those languages where the tensor/array is the fundamental unit: such as the APL family. J and K have optimized builds using AVX.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#124

Post by rbanffy »

mhh__ wrote: post #68
nickpeterson wrote: post #30

Earlier quoted context omitted.

I doubt that, think of the amount of r&d they pour into arm for mobile. ARM will be their next platform for Macs.

Probably correct but making mobile CPUs is still a level below making the monsters our desktops are blessed with(Think of all the different SIMD etc. extensions the average Intel CPU) has these days

> Think of all the different SIMD etc. extensions the average Intel CPU

Most programmers don't care whether a CPU has AVX512 support. Most programmers will use a library or an OS service that'll abstract that away and pick a certain code path appropriate to the running CPU.

I don't think a human being can get a full understanding of a modern x86 without major brain surgery that is not yet available.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#125

Post by rbanffy »

mda wrote: post #17
gigatexal wrote: post #2

If you need AVX/AVX2/AVX512 then go intel. Otherwise this AMD chip would be my choice for most applications.

No, bad idea. Test or check reviews for your common workloads first, then choose. New Zen cores quite powerful and there are many gotchas with AVX execution. In most cases I can see AMD would be better overall or offer good value.

Indeed. With any microarchitecture (or ISA) change (and Intel -> AMD is a huge architectural one) you need to test your workloads. You'll find all sort of odd and surprising performance differences caused by throttling, the way caches behave, the way libraries choose code paths, various inter-core latencies, and so on.

At least this is a move that doesn't require a full recompile.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#126

Post by mhh__ »

llampx wrote: post #117
mhh__ wrote: post #91

Earlier quoted context omitted.

Why am I being downvoted for exactly quoting apple marketing from the past?

Because it doesn't apply anymore.

^ said apple have always been like this, which is false as demonstrated by their advertising for what I believe was the OG Mac?

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#127

Post by celrod »

bitL wrote: post #10
reitzensteinm wrote: post #6

Earlier quoted context omitted.

I'd say it's a bit early to be making statements like that; Zen 2 has native 256 bit vector units, and might yet do OK with AVX2 code vs Intel processors. AVX512 though, even if you don't use the full width registers (eg to avoid throttling), the new instructions are very useful and not at all present on Zen 2. I wouldn't be surprised if Zen 3 kept 256 bit vector units, but supported the AVX512 instruction set.

Moreover, AVX2 operates at full frequency on Zen 2. AVX-512 heavily downclocks CPU; not sure about AVX2 loads on Intel right now (some people complained about AVX2 throttling on Skylake X).

I thought I read somewhere that Zen 2 will downclock based on heat, rather than simple instruction set heuristics. I can't find a source though, so take that as hearsay for now.

Intel CPUs do downclock for both avx and avx-512. Many motherboards let you configure that in the bios.

I bought both a 9940X and a (delidded) 7980XE for avx-512 intensive workloads. I use a water cooler with a 360mm radiator. The bios for these (overclockable) chips contains "AVX Offset" and "AVX512 Offset" parameters. I haven't really tested avx loads, but the avx512 downclock is necesary. They run at 70-80C when running all cores at 3.5GHz in avx512 heavy loads. I don't want to push the temperatures further than that. I don't think there's any practical way to avoid having to downclock. It does gives a sizeable speed boost overall (for those workloads).

Re: this thread I'd bet those avx512 workloads will be faster as avx2 workloads running on two 64 core CPUs with avx2 than one 56 core CPU with avx512, all else equal. But it sounds like, instead of all else being equal, things like IPC favor Zen2.

EDIT: If folks happen to be interested, I compared a bunch of different "^" (aka, "pow") functions in Julia, running on my 9940X here: https://discourse.julialang.org/t/slow-arbitrary-base-expone...

Someone else shared results with their Ryzen 2950X here: https://discourse.julialang.org/t/workstation-advice-for-mos...

The vectorized versions were those with "sleef" or "xsimd" in their name. They tended to be 1.5 to 3 times faster, while the nonvectorized versions were 1.25 to 1.35 times faster on the 9940X.

Some of my other code is likely to show a much bigger difference. For example, many small matrix multiplication operations get to take advantage of avx512's masks to vectorize efficiently, as well as the fact avx512 has 32 instead of 16 floating point registers to hold larger matrix blocks in registers, increasing the vfma to vmov ratio.

I suspect the 3.2x difference in speed in the "jsleefpowcob!" benchmark is because of the register counts. I suspect with avx512 the compiler was able to avoid register spills, while with avx2 it had to reload a lot of data on each loop iteration.

The biggest problem with avx512 IMO is that compilers seem bad at taking advantage of it (eg, they never use masks) unless you babysit them / write code with vectorization constantly in mind.

gcc for example will not use 512 bit vectors by default. You must explicitly specify "-mprefer-vector-width=512". My tests (mostly just the Polyhedron Fortran benchmarks, as a set of numerical code) seemed to confirm that gcc (gfortran) was doing the right thing.

Meaning unless you intend to go low level and use it yourself (which can be a rewarding hobby!), or have workloads where optimized libraries exist, you won't see any benefit from avx512.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#128

Post by ianai »

rbanffy wrote: post #119
api wrote: post #103

Earlier quoted context omitted.

I just ninja edited to add that. Hah. Anyway that could happen but keep in mind that a new ISA (for server workloads) is a much harder sell than a new chip brand. An ARM64 would need to be priced much lower (at least in $/perf/watt terms) than this already mega aggressive AMD offering. Personally I think Epyc this cheap really harms ARM's chances in the data center. If Apple goes ARM64 for Mac it could indirectly hel…

> a new ISA (for server workloads) is a much harder sell than a new chip brand It really depends. As long as performance is there and the apps work as expected, I'll deploy my workloads on whatever runs them for the best cost per transaction. Additionally, we don't have a say on what architecture our hosted services are on - if AWS decides my RDS databases are to move to ARM64, or Google decides my CloudSQL will run…

ARM does allow for customized silicon. But right now I’ve heard the architecture is slower than x86 at division because it doesn’t have devoted registers to the operation.

It’s going to be a huge factor in the DC. Never before have people been able to customize silicon. And now (ish) they can for figures that make sense to more than just the usual suspects.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#129

Post by api »

rbanffy wrote: post #119
api wrote: post #103

Earlier quoted context omitted.

I just ninja edited to add that. Hah. Anyway that could happen but keep in mind that a new ISA (for server workloads) is a much harder sell than a new chip brand. An ARM64 would need to be priced much lower (at least in $/perf/watt terms) than this already mega aggressive AMD offering. Personally I think Epyc this cheap really harms ARM's chances in the data center. If Apple goes ARM64 for Mac it could indirectly hel…

> a new ISA (for server workloads) is a much harder sell than a new chip brand It really depends. As long as performance is there and the apps work as expected, I'll deploy my workloads on whatever runs them for the best cost per transaction. Additionally, we don't have a say on what architecture our hosted services are on - if AWS decides my RDS databases are to move to ARM64, or Google decides my CloudSQL will run…

You are in the minority. Most IT and devops people are massively change averse and tend to blame all problems on the most recent change. They would rather pay more for a known quantity. This is even more the case for non technical management.

Re: 7nm AMD EPYC “Rome” CPU with 64C/128T to Cost $8K (56 Core Intel Xeon: $25K-50K)

#130

Post by ComputerGuru »

rolleiflex wrote: post #56
bitL wrote: post #28

Earlier quoted context omitted.

Do you have drivers for everything? Do you think it would work on a Threadripper with many NVidia GPUs as well?

No, generally if you're doing the Hackintosh the modern way, you pick your components beforehand from those supported by Mac OS natively (which are excellent components you should already be picking regardless, with the exception of GPUs), and then the only thing you need is the SMC driver I mentioned. I didn't build this machine for Hackintosh, so I have a few drivers (4-5 I think? It tends to need fewer and fewer e…

> My understanding is that it's a standoff between NVidia and Apple largely due to NVidia's poor behaviour the last time NVidia was included in Macbook Pros around 2012.

Yes, I have one of those famed 15” rMBPs that many devs say “over my dead body” when you suggest upgrading to a newer model. Zero GPU complaints after the initial bugs were ironed out (and the bad batches replaced).

Note that AMD offered virtually no competition to nVidia in either the mobile or the workstation configurations (power or efficiency) up until at least the availability of the R9 Fury a couple of years ago.

Post reply on HN