Review Raspberry PI (ARM A72) vs EPYC for DC study. Interesting results.

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
OK, now this is a very specific test, but as much as they are disparate, I don't think other scenarios would be that different.

So I got an 8 gig Raspberry PI (8 gig, 32 gig HD, A72 v8 4 core 1.5 ghz). Its does the Open Pandenics covid 19 WCG unit, 4 cores times 7.5 hours

My EPYC 7742 does 128 of these same units at 2 ghz in 2.5 hours.

So it would take 32 PI's to do the same work, but 3 times slower. An 32 PIs at 5 watts each is 160 watts, vs the 7742 @250 watts. (motherboard, memory and all) So more power and 3 times longer for the same work is 2 times the power usage. (1200 watts total vs 625)

Cost... 32 PIs at about $60 each (including power supplies and cables, probably more) would be at least $1920.

The EPYC is about $4000+480+580 or $5060. (I paid $3000 less, that retail on ebay)

So the EPYC effectively less because it cost 2 times less electricity for the same work.

This seems WAY different than I am seeing when people talk about ARM and efficiency and power usage. Please tell me where I am mistaken in my math., But don't be a jerk if you find the errors of my ways. The run times are real.. The PI is at 7.5 hours and hasn't actually finished a unit yes, no other tasks running on the PI. And I am looking at several units on the 7742, one is at 97% in 2:21.


Edit: The first unit finished in 7:50, almost 8 hours. The next 2 units are at 8 hours and still running at 96 and 98%.

And my comment on "wins by a large margin" has to do with total power usage, which in data centers, and for me is a big deal.
 
Last edited:

insertcarehere

Senior member
Jan 17, 2013
639
607
136
To catch up with Milan, ARM parts will need to be twice as fast as the A77 while consuming orders of magnitude less power. This is not hyperbole.

The A77 trails the Apple A13 and most current high end x86 chips by 30-40% in single threaded performance. The Apple A13 has 2 big cores that can match last year’s x86 in performance, but they run too hot and consume too much power to scale up. The A14 is expected to only be 5% faster in ST performance than the A13 and will be significantly slower than this year’s x86 chips (Zen 3, Tiger Lake, etc.) despite having a huge node advantage.
Qualcomm and Apple didn't get the memo that ARM cores were supposed to lay down and die vs x86, A77 is already competitive vs Zen 2 (4800u boosting far higher than EPYC Rome no less) and A13 matches TGL, both with significantly constrained thermals and power vs their x86 counterparts, as usual.

111168.png
 
  • Like
Reactions: Tlh97 and Etain05

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Qualcomm and Apple didn't get the memo that ARM cores were supposed to lay down and die vs x86, A77 is already competitive vs Zen 2 (4800u boosting far higher than EPYC Rome no less) and A13 matches TGL, both with significantly constrained thermals and power vs their x86 counterparts, as usual.

111168.png
Oh, a 14 year old single core benchmark, and we are supposed to be impressed ?? And my iphone11 beats a graviton2 and a 3950x ? Yea right....

Its like "he who shall not be named" is back.
 
Last edited:

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Please tell me where I am mistaken in my math.,

As as been said, A72 is old and Pi4 is on 28nm. So even I'm not an ARM fan either, this is for sure not a fair or meaningful comparison.

Does anyone have a suggestion for a faster one that runs any version of linux that I can buy ?

That's the issue. I think somewhere here there is/was a thread about getting access to such embedded boards with more modern ARM CPU. simply said it doesn't exist. Only way to get access is via a smartphone. An A77 on 7nm (eg what is in most modern flagship phones) will run circles around the A72 in the Pi4 while the apple SOCs will be even faster.

However more extensive benches done by phoronix on graviton2 show some serious performance issue especially in compilation workloads or workflows being able to load all cores (interconnect issue most likely). So ARM for sure has a long road ahead. With apple releasing ARM laptops for developers to build apps this will be a big boost to the ARM ecosystem. Remember, that is exactly how x86 success started. x86 is what devs worked on so developing on x86 for x86 server was simply more seamless than to make software for an alpha CPU on x86. With ARM developer devices it will become more seamless to develop for ARM server and ARM consumer devices. It would make sense honestly. I do think Intel needs to be afraid. Very afraid. IMHO the biggest possible bump in the road to "ARM world dominance" is NV buying them. But even if it runs smoothly it will be a decade(s) long process as the server world is far,far more entrenched in x86 than it was in what preceded x86 and especially big enterprise is notoriously slow to invest into migrations (else oracle wouldn't exist anymore)
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Oh, a 14 year old single core benchmark, and we are supposed to be impressed ?? And my iphone11 beats a graviton2 and a 3950x ? Yea right....

Its like "he who shall not be named" is back.

Sometimes you puzzle me really. If someone is posting benchmarks, you complain that the benchmark is old - as if newer versions of SPEC would paint a drastically different picture. However if someone posts total nonsense like eek2121, which is not even worth replying to, you give likes to him.

And yes, an A13 is really that fast - Graviton 2 is using a relatively old Cortex A76 - look it up in the table i posted earlier. Also a Phone using an Snapdragon 865 (Cortex A77) is supposed to be faster than Graviton 2 single core for the same reasons.
Likewise an EPYC 7742 with SPECInt2006 score of around 36 would make the bottom half of the list as far as single core is concerned - despite it being clocked 20% higher than say a Cortex A77 in the Snapdragon 865 (3.4 GHz vs. 2.8 GHz).
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Do you have a cochlear implant like me ? Do you use your phone to control it ? There re no benchmarks for this, so you will have to take my word for it. If you don't like that too bad, go kill your hearing with drugs or something and get a cochlear implant. Then talk to me.

So you're saying your iPhone 11 is slower to control your cochlear implant than your Galaxy S9 and your conclusion is that the S9 is faster? What does controlling a cochlear implant have to do with raw CPU performance?
 

SarahKerrigan

Senior member
Oct 12, 2014
339
468
136
Seriously? We're comparing a 28nm chip with a microarchitecture that's something like five generations old with a currently-shipping 7nm chip with a current uarch? Shall we drag out some K8's, compare their efficiency against a Pi4, and then claim x86 is overhyped?

A current ARM core is going to be something like 2.5x as fast at iso clock, compared to the A72, and they're capable of running efficiently into the 2.5-3.0GHz range. This thread is almost as hilarious as a certain "HIGHEST IPC CORES OF ALL TIME!!!" hype thread.
 
  • Like
Reactions: Lodix

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Seriously? We're comparing a 28nm chip with a microarchitecture that's something like five generations old with a currently-shipping 7nm chip with a current uarch? Shall we drag out some K8's, compare their efficiency against a Pi4, and then claim x86 is overhyped?

A current ARM core is going to be something like 2.5x as fast at iso clock, compared to the A72, and they're capable of running efficiently into the 2.5-3.0GHz range. This thread is almost as hilarious as a certain "HIGHEST IPC CORES OF ALL TIME!!!" hype thread.
Why don't you try reading more than one post, before you react and post.
 
  • Like
Reactions: Tlh97 and Drazick

naukkis

Senior member
Jun 5, 2002
701
568
136
Oh, a 14 year old single core benchmark, and we are supposed to be impressed ?? And my iphone11 beats a graviton2 and a 3950x ? Yea right....

Its like "he who shall not be named" is back.

If brave enough CEO of big company sees those spec results, and instead of denying them believe in those to be valid could make some strong bets on those results. Like $40 billion dollars....
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
@Markfw if you do intend to look at phones, you don't match the Xavier NX or AVX until Qualcomm 855 chips or better. The 855 might be able to match the NX under ideal circumstances, and phones with it start at about $300 on amazon. The 865 looks to be better than the NX in theory, but, it remains to be seen if it can actually sustain that level of performance. While the NX has an older lithography node, it is far better able to discipate heat and has much more capacity to draw power. The leading edge node of the 865 will give it a lot of advantages, especially in performance per watt. Phones with it are still a bit pricey though.

Neitger of them have the GPU resources of the Xavier's. If you employ those, this gets academic really quickly.
 
Apr 30, 2015
131
10
81
It seems that ARM have a Neoverse N1 SDP- 'segment development platform'. It is a board about 6 or 7 inches square. It looks like an extra large Raspberry Pi. It is available to partners. They give a sales office or you can contact them online.

https://community.arm.com/developer/tools-software/oss-platforms/w/docs/458/neoverse-n1-sdp refers.

It appeared in ARM TechCon 2019:


I think that this board could be the basis of a desktop machine for enthusiasts. I would like to use the N2 board with SVE, when it comes out. It would be good if ARM recognised that there is a market for these things. A full software-stack is available.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
It seems that ARM have a Neoverse N1 SDP- 'segment development platform'. It is a board about 6 or 7 inches square. It looks like an extra large Raspberry Pi. It is available to partners. They give a sales office or you can contact them online.

https://community.arm.com/developer/tools-software/oss-platforms/w/docs/458/neoverse-n1-sdp refers.

It appeared in ARM TechCon 2019:


I think that this board could be the basis of a desktop machine for enthusiasts. I would like to use the N2 board with SVE, when it comes out. It would be good if ARM recognised that there is a market for these things. A full software-stack is available.
I contacted them. They said it was a $10,000 part, and extremely limited production, so I was excluded.
 
  • Like
Reactions: Drazick

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
It really is hard to get modern ARM cores outside of a phone. Really, really hard. There is Huawei's cut-down Kunpeng 920 PC board (8c) but it doesn't seem all that performant. Taishan v110 definitely doesn't seem to be on the level of A76, much less A77.

Plenty of people are still waiting for HiKey 980 boards. And the Rockchip 3588:


I really wanted an RK3588 just to play with it. At this point though, ugh.
 
  • Like
Reactions: blckgrffn

CluelessOne

Member
Jun 19, 2015
76
49
91
Oh, a 14 year old single core benchmark, and we are supposed to be impressed ?? And my iphone11 beats a graviton2 and a 3950x ? Yea right....

Its like "he who shall not be named" is back.

I am not sure I get the referred person correctly. I thought the "internet strongman" had never posted here although some members dared him to. Or are you referring to a certain Danish username?
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
It really is hard to get modern ARM cores outside of a phone. Really, really hard. There is Huawei's cut-down Kunpeng 920 PC board (8c) but it doesn't seem all that performant. Taishan v110 definitely doesn't seem to be on the level of A76, much less A77.

Plenty of people are still waiting for HiKey 980 boards. And the Rockchip 3588:


I really wanted an RK3588 just to play with it. At this point though, ugh.

Doesn't this seem just al little ridiculous?

I mean, if we could buy a "developer board" - mITX would be sweet, although I know the focus is even smaller - even just one year back from current that seems like it would make a lot of sense?

Linux seems to run well enough and you can get a ton of stuff easily up and going via Docker, there are lot of applications (btc related stuff comes to mind for me, lightning nodes, etc) and then just network appliances (I want to build a solid Untangle box) that it seems like even one solid, near-current platform would see a lot of traction.

It seems like about the time the market is moving to the next hot thing there should be availability on this, even if not limitless enough to create tens of thousands of units? They make many, many millions of smartphones.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
The biggest issue that I see for the market is that no one produces an ARM processor on any sort of ZIF socket for anything other than the big server iron. They are all BGAs soldered to boards, and those are all tailored to either a very specific niche (Nvidia Jetson Xavier) or are more multi-purpose, but VERY heavily cost cut (RaspPi and the ilk). There very likely is some sort of market for an ITX form factor development system, but, is it enough to fund the sort of development money that it would require? Remember, most ARM development can be done in virtual machines on x86 workstations. Even though that ITX system may be desirable from a flexibility standpoint, it likely won't be any more representative of the target of the software that's developed on it than the VM on an x86 PC.
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Remember, most ARM development can be done in virtual machines on x86 workstations. Even though that ITX system may be desirable from a flexibility standpoint, it likely won't be any more representative of the target of the software that's developed on it than the VM on an x86 PC.

What would the purpose of the VM here? Typically you just cross compile and den deploy on an ARM machine for testing.
Ideally of course you have a sufficient powerful native ARM development platform - many build flows having issues with cross compiling.

Point in case, i am doing all ARM Windows and Android development using cross compilation and Linux mostly native ARM (using Surface Pro X). There is no point where virtualization helps here.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Doesn't this seem just al little ridiculous?

It does, on many different levels. I'm gonna throw out some guesses though:

ARM desktop/workstation/server doesn't have a standardized EFI (that I know of). While Linux seems to be doing a pretty good job of supporting proper booting on nearly every version of server/workstation ARMv8 out there, there's always that risk of ARM hardware not having proper kernel support. See HiKey 970 and those ridiculously-overpriced Snapdragon developer boards. Those were only ever meant for Android.

"high end" ARM hardware needs to ship in fairly significant volumes to produce profits. How many phones do you have to sell to be able to justify producing a phone based on A77, for example? You can't necessarily just take a phone platform, strip off the touchscreen, hook it up to a power brick and sell it as an SBC since it would still lack the ports you would want for a miniature developer board. I guess you could use a USB-C attached daughterboard to provide other ports, but you're still dealing with a board that was only ever meant to run Android natively, so good luck getting Linux to boot on it properly.

Your alternative is to go the route that organizations like Pine take, which is to get the cheapest ARM chips you can (usually a Rockchip offering), design a board around that years after the ARM core the SoC is based on is relevant, and hope you can sell enough SBCs to makers and other weirdos who like that sort of thing. I would hate having to rely on Pine.org for a dev machine. Trying to design something around a performant A77 SoC today - for example, Qualcomm 865 - and bringing it to market in a reasonable time span (let's say Q3 2021) would probably yield a board that could be pretty expensive.

There very likely is some sort of market for an ITX form factor development system, but, is it enough to fund the sort of development money that it would require?

That appears to be a problem. Nobody seems to want to take the risk of designing a one-off SBC for one particular generation of ARMv8. We don't know how many people would buy these things.

edit: I will say this: if you are in China and your org is considering full-fat 64c Kunpeng 920 hardware for your server room or wherever, getting the admittedly-kind-of-expensive 8c Kunpeng 920 as a dev board makes a lot of sense. So kudos to Huawei for offering a cut-down version of their 64c server offering through channels in China that can probably supply such hardware in considerable volume, at per-unit prices below $1000.

Since you can't actually buy Graviton2 for your org, well, you can just use AWS to profile applications for deployment on (drumroll please) AWS. I guess Amazon deserves credit for that too!

Otherwise, it's pretty barren. I think there were some ThunderX2 dev machines that were hard to get and really expensive. ThunderX3 is mostly dead in the water so developing for it is not really desirable. There's those pricey Ampere systems that are using old Xgene hardware. Bleh. A64FX is so incredibly niche that we'll probably never get dev hardware for that. And disappointingly, ARM doesn't seem to have partnered with a hardware manufacturer to promote its own Neoverse platform (N1, V1, or N2) by producing their own dev hardware that could give people a chance to profile applications on bog-standard cores in such a way that could help application deployment on just about any present or future ARM server/workstation offering.
 
Last edited:
  • Like
Reactions: blckgrffn

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
While Linux seems to be doing a pretty good job of supporting proper booting on nearly every version of server/workstation ARMv8 out there, there's always that risk of ARM hardware not having proper kernel support. See HiKey 970 and those ridiculously-overpriced Snapdragon developer boards. Those were only ever meant for Android.
It gets more ridiculous right there. Android's kernel is Linux. So all the Android devices are actually running Linux. The problem is many ARM chips only get specific Linux kernel blobs, with the code never being moved upstream and as such those hardware specific kernels becoming outdated quickly and none of the hardware compatibility being retained in the mainstream kernel proper. Google is constantly trying to fix the update nightmare resulting from that widespread ill-suited "industry practice" e.g. with AOSP Common Kernel and Project Treble (which were very overdue), but in the long run Google may well move away altogether from Android and Linux to Fuchsia with a new kernel called Zircon instead.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
The other issue that we're going to start facing is that, as the actual SoC layouts get even more and more customized for mobile use than before, they will be less and less suitable for use as a standard desktop type of system. Our hopes seem to be moving towards the slowly growing laptop ARM market, like the 8CX, and its follow ons. If someone could arrange for extra 8CX production for dev boards, or even uSFF desktops, you'd at least have somewhere to start with.
 

soresu

Platinum Member
Dec 19, 2014
2,612
1,811
136
but in the long run Google may well move away altogether from Android and Linux to Fuchsia with a new kernel called Zircon instead.
More likely it will remain Android with a Zircon/Fuchsia based foundation instead, and the same with Chrome OS.

Basically the same as happened with the NT kernel transition in Windows.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Getting the better ARM chips outside phones is impossible... consumers generally get the chips a few years late, at this point we are getting the A72 cores... The Hexa RK3399 is about the best we can get and im not even sure that is faster than the RPI 4 CPU, the GPU is, by a lot.

The reason for the delays is simple, there is no demand for such a thing, so the best ARM chips they first go in phone, then to premium notebooks, then to tv boxes, tablets, etc etc.

PINE64 usually have some good devices there, single boards, notebooks, even a cluster board, but RK3399 is the best they have.

Things need to change a lot if they wanna capture more public, my question is... is this even necesary? they have a huge market with phones, and is already hard to keep up there.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Otherwise, it's pretty barren

Thats why ARM Macs are dangerous to x86. Suddenly you have a readily available probably rather powerful ARM dev environment. As said in a previous comment, normal PCs being x86 is what made it successful as it triggered more and more x86 software.
 

piokos

Senior member
Nov 2, 2018
554
206
86
Thats why ARM Macs are dangerous to x86. Suddenly you have a readily available probably rather powerful ARM dev environment. As said in a previous comment, normal PCs being x86 is what made it successful as it triggered more and more x86 software.
That has to be one of the weirdest arguments in this thread. ARM Macs being dangerous to x86, because they will be better for developing software for the - non-existent - ARM servers.

And of course this is not true - the simplest counterargument being: people have been using x86 PCs to develop ARM software (for iPhones, Androids, anything embedded) for 3 decades.
 
  • Like
Reactions: Tlh97 and blckgrffn