How A Raspberry Pi 4 Performs Against Intel's Latest Celeron, Pentium CPUs

[wizard69]

Interesting that the pi was faster at exporting to PDF. For the rest it looks to be about 1/4 the speed of the Intel chip. I wonder if the price is matched at that level for general use.

Don't forget a vastly slower storage and limited RAM.

[wizard69]

Unfortunately the microarchitecture of the RPi4 is still a bit of a joke.

I've had a lengthy (and peppered with the occasional insults, of course) discussion earlier on phoronix forums.

It's not just the clock speed. The IPC of the RPi4 is mediocre.

The RPi4 chip (Cortex A72) doesn't have a proper Load Store Queue, to allow it to issue loads out-of-order with respect to earlier stores (so all prior stores need to complete before issuing a load). This would require machinery to detect and correct potential misspeculations, of course.

It can only decode 3 instructions per cycle, and issue 5.

Actually the venerable Alpha 21264 from 1996 had better microarchitecture. Granted, that was a trailblazer chip, but still ... 1996!

So, this is not about nm. It's about a toy microarchitecture.

Sure is is a toy or more accurately a cell phone architecture but still it has come a long way.

This is one reason why I'm very interested in what Apple delivers as far as a performance ARM processor goes. Hopefully we can see what ARM can do as a mainstream processor then.

Frankly I'm a bit disappointed that Micheal didn't spend more effort to configure the intel machine like the Pi. That is minimal RAM and slow SSD. On the other hand I really doubt if the PI even comes close to the power usage of the Intel beast.

[ed31337]

During the summer, it's just too damned hot to use anything other than my Raspberry Pi for computing. I don't care how much more performant an Intel x86 might be, the Raspberry Pi manages to play video, browse the web, and let me write code without breaking a sweat. The Pi not only saved me a few bucks in purchase price for a new computer, but on-going it is saving me a ton in electricity for running the A/C.

It also seems to be a lot more reliable. My old Intel quad core laptop will occassionally spontaneously reboot, while my RPi4/4GB has pretty much never done any crashing, aside from Out of Memory situations (which isn't really a crash per se, but Linux gets so slow when it runs out of memory that it might as well be locked up).

[CommunityMember]

There are custom ARM designs from Amazon

Also don't forget the Ampere 80 core processor (designed for the hyperscale cloud providers). It was supposed to ship around now in volume (but, again, only to the hyperscalers, and I would not be surprised the schedule has been "adjusted").

[schmidtbag]

ARM is an architecture that is supposed to scale up into the high-performance too.

According to what? Just because there are scaled-up models, doesn't mean the core architecture is supposed to. Bear in mind, the context here is scaling up in clock speed, not instructions.

The Raspi and embedded devices in general are designed to be efficient, but there are high-end ARM CPUs that really need a real heatsink.

I'm aware, but they're practically a wholly different product, as are Apple's CPUs. So it doesn't make sense to compare a low-end SBC like that, which is what's going on here.

That guy is talking about SPECIFIC CPU core designs from ARM (the ones that are licensed to third parties) being inferior, which is at least somewhat believable at face value.

If performance is what you seek, yes. But these aren't the products you buy for high performance.

You are just assuming a whole architecture/ISA is the same just because it has the same interface for software, which is complete bullshit.

You are assuming that I don't know that, yet you ignore the fact that we're talking about a $35 computer here. To compare it to high-end stuff is complete bullshit.

[CommunityMember]

There are custom ARM designs from Amazon

Amazon bought Annapurna Labs around five years ago to get their ARM tech, and their current (branded) Graviton2 processors (a variant on the Neoverse N1) are quite appropriate for many scale out use cases, and the pricing is very very good for those use cases (well, AWS very good pricing). I only wish AWS would offer graviton processors in their lightsail offering so the entry to experiment effort was low enough so more people would consider them (spinning up an EC2 instance is not especially hard, but if you are starting from little AWS knowledge, it seems like a mountain to climb). I am hopeful that will eventually happen (the AWS people I talked to at the last ARM TechCon seemed receptive).

[CommunityMember]

Unfortunately the microarchitecture of the RPi4 is still a bit of a joke.

The broadcom chips that the raspberry pi foundation selected were always very specific cost optimized designs, typically not even available to other vendors. Price, price, price, and price, was a core consideration (and it is generally believed that broadcom offered a sweatheart pricing deal such that broadcom may have made little profit on the chips). The RPi foundation never started with a plan to offer general purpose computing, they offered a maker item, to teach computing, and building simple experimental devices. Sure, they pivoted a bit once they got successful, but even they never quite expected for the RPi to be a widely used general purpose desktop (there were, and still are, better solutions for that, even in the ARM space), even while understanding that for their target audience they had to offer a GUI. I am glad the RPi exists, but it is somewhat unfortunate that its mindshare crushed some alternative solutions that were more flexible.

[starshipeleven]

ARM isn't built to compete with desktop performance.

This is your own idea.
There is higher end stuff from Marvell and NXP and whatever that is using high-performance CPU designs licensed from ARM, there are Apple ARM designs where they are designing their own cores from scratch and are actually pretty decent for desktop/laptop. There are custom ARM designs from Amazon, again nothing like ARM's CPU designs.
NVIDIA's jetson and whatnot boards too.

That guy is talking about SPECIFIC CPU core designs from ARM (the ones that are licensed to third parties) being inferior, which is at least somewhat believable at face value.

You are just assuming a whole architecture/ISA is the same just because it has the same interface for software, which is complete bullshit.

[starshipeleven]

ARM isn't built to be performant. It's built to be efficient.

Is it really?
ARM is an architecture that is supposed to scale up into the high-performance too.

The Raspi and embedded devices in general are designed to be efficient, but there are high-end ARM CPUs that really need a real heatsink.

[schmidtbag]

I agree that there isn't really a one size fits all. For instance, there are really low power/low cost microcontrollers (e.g. PICs). Still, it's a disappointment that ARM implementations from ARM itself are doing poorly (that's how I view things).

They're doing poorly because you're judging it for something it wasn't meant to do. As the saying goes, "judge a fish by its ability to climb a tree and it will think it's stupid". ARM isn't built to compete with desktop performance.

The main reason Apple ARMs work so well is because they hired Jim Keller and a few other people to build a high quality microarchitecture, which gives really good IPC, and good performance per Watt as well. Look, IPC matters a lot, you really can't ignore it.

Yes, I basically said that myself but in fewer words.
IPC can matter a lot. It isn't the only thing that matters.

Let me repeat something I said in my first post - RPi4 (Cortex A72) doesn't even have a proper load-store queue, to issue loads speculatively wrt prior stores. What are your thoughts on that?

I don't know enough about that to make a worthwhile comment on it, which is why I didn't comment on it in the first place. What I do know is it clearly isn't a necessity to make ARM usable on a day to day basis, of course, assuming you're ok with the level of performance (which plenty of people are). I feel like if it was such an obvious thing to add, they would have done so. After all, the NEON instructions for example aren't exactly a simple addition.

I'm not making a statement about Intel here (you don't need to repeat yourself). I'm not saying at all that Intel's doing ok - I'm saying that RPi 4 is performing poorly. And Intel does have actual low-power chips as well (Y series) - probably OEM only (harder to benchmark), and expensive as hell (yes, cost matters too ... no doubt about it).

Poorly compared to what? You have to make comparisons, and the context of the article included Intel. ARM is a hell of a lot better compared to most, if not all other RISC architectures (POWER is faster but it is a lot more power hungry too). Apple's CPU might be better, but it's not going to be cheap; this Broadcom chip offers some fantastic performance[-per-watt] for the price. Any CPU can be made better if you just cram more instructions in it, but then it becomes expensive and inefficient for more basic tasks. Like I said multiple times already: you're expecting this CPU to be something it's not. It does what it was built to do very well.

Well, I'd really be interested in Michael doing some perf per watt numbers. I'd also be interested in instruction counts and IPC data for the benchmarks. Just sayin ...

Me too. Though, kinda the point of these CPUs is they don't have a lot of instructions. My server uses A53 cores (not Broadcom) and this is all it shows for features in cpuinfo:
half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt lpae evtstrm aes pmull sha1 sha2 crc32
Sure isn't much to look at, eh? But it's fast enough for what I need it to do.