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

[Markfw]

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.

 

[JoeRambo]

A72 is a design from ~2015 and not on leading edge process like AMD either. So its like comparing fresh Apples to outdated fried fruit.

Make no mistake, those ARM monolithic monsters will steal business from AMD/Intel. Once per core performance is decent enough, it is game over.

 

[Thala]

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.,

Math is most likely not far off, but you are comparing an A72 on a 28nm process node - which should you scream in your face, what is wrong with your comparison.

 

[NTMBK]

7nm processor more efficient than a 28nm processor? Say it ain't so!

 

[piokos]

I don't even understand what was the idea behind this comparison (other than "AMD is so awesome").

You're comparing price and power consumption of a single system vs a cluster made of 32 complete systems.
You're powering (and paying for) redundant elements. Under full load RPi4B can pull slightly over 6W, but just half of that is used by the CPU.

And I know you understand this perfectly well, because you love mentioning how price-effective 64-core EPYCs are.

As was already mentioned: A72 is not the most advanced design, made on a fairly old node.
Frankly, if you got that RPi4 just to write this post, it would be cheaper and more accurate to just use an AWS Graviton instance.

 

[dacostafilipe]

Your test is a great example of the issue I have with all the ARM Server talk at the moment.

Even if some companies are able to produce extremely fast ARM Cores, it does not make everything ARM fast.

People use Apple to showcase how ARM improved over the years, but it has less to do with the ARM instruction set and more with the gigantic amount of resources Apple invested in their custom cores.

/rant

As for you tests @Markfw, I don't think you can compare the scores like this because of the nature of both platforms, but others do it all the time with A13 or Graviton2 scores ...

 

[piokos]

Your test is a great example of the issue I have with all the ARM Server talk at the moment.

Even if some companies are able to produce extremely fast ARM Cores, it does not make everything ARM fast.

Because ARM today is just a totally different ecosystem

RPi4 in question uses a basic chip with cores designed in 2016, made on a node that was lagging behind x86 at that point. And it's totally fine for the target client group.
It's like if Dell launched a laptop with Sandy Bridge in 2019. And even if it was well made, had all the features people need and was perfectly usable for daily tasks, enthusiasts would probably implode.

If ARM becomes a mainstream performance architecture, it will be managed just like x86 is today. We'll get yearly updates and PCs/servers will generally come with the latest and greatest implementations.
Amazon is already on this schedule. They'll probably update Graviton every 1-2 years.

 

[dacostafilipe]

If ARM becomes a mainstream performance architecture, it will be managed just like x86 is today. We'll get yearly updates and PCs/servers will generally come with the latest and greatest implementations.

ARM does not sell CPUs.

Stop talking about ARM in general and get some real products to prove your point.

Amazon is already on this schedule. They'll probably update Graviton every 1-2 years.

Graviton2 is not there yet in terms of raw performance, but they are close and could match Intel/AMD in the next iteration.

Sadly, you can't buy those CPUs outside of Amazone, so it's not really a competition. Not every company has the money a Amazone/Google/Apple/... has to create their own custom CPUs.

Other companies (ex Nuvia) are talking about similar Server CPUs, so let's wait.

 

[Thala]

Your test is a great example of the issue I have with all the ARM Server talk at the moment.

Even if some companies are able to produce extremely fast ARM Cores, it does not make everything ARM fast.

Are you implying that someone is going to use 5 year old A72 cores on a 28nm node to build servers to compete with AMD? If not, the test is not a great example at all.

 

[Markfw]

So, its a v8, most recent version ? its only about 3 years older than Rome ? Does anyone have a suggestion for a faster one that runs any version of linux that I can buy ?

This started as me thinking that PIs would be more efficient than EPYC, so I got one to play with. I was amazed at how slow it was, since everyone was talking up ARM.

 

[Thala]

So, its a v8, most recent version ?

Cortex A72 in RPi4 is 6 generations behind the latest ARM architectures in addition to 4-5 process nodes, compared to what is going into servers. A72->A73->A75->A76->A77->A78->AX...
The recently announced Neoverse V1 platform (up to 96 cores per socket) is going to use Cortex AX and the Neoverse N2 (up to 128 cores per socket) is going to use Cortex A79.

 

[Markfw]

Cortex A72 in RPi4 is 6 generations behind the latest ARM architectures in addition to 4-5 process nodes, compared to what is going into servers. A72->A73->A75->A76->A77->A78->AX...
The recently announced Neoverse V1 platform (up to 96 cores per socket) is going to use Cortex AX and the Neoverse N2 (up to 128 cores per socket) is going to use Cortex A79.

So what you are saying is that the only chip what one can buy to test the same application on, is what I have, but that its so old, it should not be compared. And we can't test any of the new chips using anything but the likes of geekbench ? So I take the word of geekbench that a platform that does not exist yet will be faster than a 2-3 year old chip like Rome ?

This is the problem I have with ARM.

 

[Hitman928]

There was really only 1 poster in this forum who thought A72 could hang in any way with AMD's latest CPUs, but he doesn't post here any more so I don't think you'll find much audience here for this comparison.

If you can get your hands on a Neoverse 1 based chip or something A78 based, that would be more interesting but I don't know where to find these CPUs in a comparable platform on the retail market.

 

[Markfw]

There was really only 1 poster in this forum who thought A72 could hang in any way with AMD's latest CPUs, but he doesn't post here any more so I don't think you'll find much audience here for this comparison.

If you can get your hands on a Neoverse 1 based chip or something A78 based, that would be more interesting but I don't know where to find these CPUs in a comparable platform on the retail market.

The day ANY ARM chip can be used to compare a real application (probably linux based) to AMD or Intel is the day I start believing that ARM is a contender.

 

[Thala]

So what you are saying is that the only chip what one can buy to test the same application on, is what I have, but that its so old, it should not be compared. And we can't test any of the new chips using anything but the likes of geekbench ? So I take the word of geekbench that a platform that does not exist yet will be faster than a 2-3 year old chip like Rome ?

This is the problem I have with ARM.

You can buy phones with up to Cortex A77 since quite some time now. Not sure what problem you have...

 

[Markfw]

You can buy phones with up to Cortex A77 since quite some time now. Not sure what problem you have...

And what application can a phone run to compare with AMD/Intel ? Oh, and I have an iphone 11, I thought that was supposed to be fast, but I am annoyed with it every day. I liked my Galaxy 9 much better. I mean speed wise.

 

[Doug S]

So what you are saying is that the only chip what one can buy to test the same application on, is what I have, but that its so old, it should not be compared. And we can't test any of the new chips using anything but the likes of geekbench ? So I take the word of geekbench that a platform that does not exist yet will be faster than a 2-3 year old chip like Rome ?

So when Apple introduces the new Macs and you have a recent ARM to compare with, if it beats EPYC in your tests will you complain that it is irrelevant because you can't buy Apple cores in a server?

 

[JoeRambo]

You can buy phones with up to Cortex A77 since quite some time now. Not sure what problem you have...

Actually the problem is very clear to everyone. It is impossible to buy a competent (read mainstream, not some joke effort from Q vendor) Arm workstation with Windows or Linux. Once Apple is out with ARM Macs, that problem will go away.

 

[Zucker2k]

I've always seen this ARM vs x86 comparison as a technical, developmental comparison because of the platform differences. Some people treat it literally as a direct product comparison, hence making the task a lot more difficult. Having said that, there's a convergence point in the near future and that's why these comparisons are interesting, to say the least.

 

[Hitman928]

So when Apple introduces the new Macs and you have a recent ARM to compare with, if it beats EPYC in your tests will you complain that it is irrelevant because you can't buy Apple cores in a server?

Mark really only cares about perf and perf/w for his DC projects (except maybe for his personal use machine but that is just one machine compared to many DC machines). If he could buy an ARM machine at a reasonable price that had clearly superior perf/W than any x86 machine, he would buy it in an instant. Don't let his gruffness distract from the fact that he really just hops on whatever train gets him the best results for his DC projects.

 

[dacostafilipe]

So when Apple introduces the new Macs and you have a recent ARM to compare with, if it beats EPYC in your tests will you complain that it is irrelevant because you can't buy Apple cores in a server?

If you need to build a server and Apple does not sell you the CPUs, why would you pay attention to Apple?

 

[piokos]

ARM does not sell CPUs.

Stop talking about ARM in general and get some real products to prove your point.

?
Where exactly did I say ARM sells CPUs?

Graviton2 is not there yet in terms of raw performance, but they are close and could match Intel/AMD in the next iteration.

I don't see why this would be important here. I mentioned Graviton as an example of a product line properly managed and kept up to date by the manufacturer (just like Apple and Qualcomm do). This is in contrast to most ARM chips in production today, many of which are really old. And even the new models could be made out of 5 or 10 year-old cores.

A proper product line for consumers would have to be well organized into a familiar scheme of generations and segments.

Sadly, you can't buy those CPUs outside of Amazone, so it's not really a competition. Not every company has the money a Amazone/Google/Apple/... has to create their own custom CPUs.

I thought we were talking about using processors (say, for computing), not owning them.
OP compares how they perform in a benchmark, not how they smell or whether they match his window blinds.

Other companies (ex Nuvia) are talking about similar Server CPUs, so let's wait.

Of course everyone is focusing on server CPUs. That's the future.

Actually the problem is very clear to everyone. It is impossible to buy a competent (read mainstream, not some joke effort from Q vendor) Arm workstation with Windows or Linux. Once Apple is out with ARM Macs, that problem will go away.

And what if it turns out that Apple only makes ARM chips up to high-end mobile level (so MacBooks, Mac Minis)? No workstations?

 

[KompuKare]

The next time AT's reviews or benchmarks Amazon's Graviton2 or similar, maybe we could nudge Andrei Frumusanu to run something more interesting than Spec figures?

 

[dullard]

Mark really only cares about perf and perf/w for his DC projects (except maybe for his personal use machine but that is just one machine compared to many DC machines). If he could buy an ARM machine at a reasonable price that had clearly superior perf/W than any x86 machine, he would buy it in an instant. Don't let his gruffness distract from the fact that he really just hops on whatever train gets him the best results for his DC projects.

The problem is that Mark then applies his DC project specific conclusions to everyone and every application. If it isn't the best for his DC products, then we are wrong no matter what our usage is.

 

[piokos]

The next time AT's reviews or benchmarks Amazon's Graviton2 or similar, maybe we could nudge Andrei Frumusanu to run something more interesting than Spec figures?

Like what?

If these tests are meant to compare CPUs for enterprise scenarios, SPEC is the golden standard.