Computing power

Hangar Flying
1

So, today I ran this old gem on my phone, on ScummVM.

Screenshot_20191118-195621_ScummVM
Screenshot_20191118-195621_ScummVM1013×493 268 KB

And I wondered just how much faster my phone is, compared to the speed of the PC I played this game on, in… 1990 I guess.

And WTF. I knew today’s stuff is a lot faster, but seeing the numbers is insane. It doesn’t feel right, like, if someone just was off by an order of magnitude or two. It is just unreal.

So let’s see.
MIPS = Millions operations per second
FLOPS = Floating point operations per second. Usually in MFLOPS or GFLOPS (M = millions, G = billions and so on)

My PC back then was an 80286 with a stunning 12 MHz and a MIPS claim of… two point six. 2.6 MIPS.
I can’t say a FLOPS number since it couldn’t do floating point operations at all. Damn. I don’t know enough about theoretical computer science to make an approximation if one would just use big numbers and shift them around.
So, I went and looked at the weakest comparable processor I could find that could do floating point operations, a (much faster) 80486/25. That one is at least four times as fast (in fact its MIPS claim is a stunning fifteen (15), so that’s six times faster) Its MFLOPS number is… wait for it… One. Literally. 1. Uno. That’s it. Cool. So let’s take this.

Edit: corrected some numbers.
Now, my Samsung Galaxy S9 has an Adreno 630 GPU (710 MHz)
Conservatively speaking the CPU alone should have a raw computing power of… somewhere between 200 and 300. GFLOPS. No, not MFLOPS. GFLOPS. With G.

The GPU/CPU (integrated) comes in at (conservatively computed according to my source) at least 720 GFLOPS.

So even if they don’t work perfectly together that puts my phone at something like 700 GFLOPS of raw computing power.

1 GFLOPS is 1000 MFLOPS.

So that difference between my old PC and my phone is… man, that’s… pulls out calculator … no wait… That’s a factor of…
700,000??? Seven-hundred-thousand??? (And that’s compared to the 80486, and not the 80286)
…so I don’t even exaggerate when I say that my phone is more than a million times faster than my old PC??

At which point did we enter the future? This is clearly the future.

(And please check my math, I did this mainly in my head so I might be horribly wrong)

Edit: did I mention that this is a phone? A PHONE!
It isn’t even a PC! A modern graphics card for 500 bucks would be fast enough to make the top500 super computer list of 2008 (not a joke, an RTX2070 can do 14 TFLOPS or so).

Have a good day, all you people of the future! :slight_smile:

8 Likes
2

So it’s… Over 9,000?

4 Likes
3

Yeah, it is ridiculous.
I just added in a small paragraph about a modern GPU. It blows the mind. Hard.

I am tempted to post a meme of John McEnroe yelling “You cannot be serious!”.

1 Like
4

Phelps-9000

1 Like
5

I think there is an important lesson here in making do with what you have - or in our current state, not making use of it. Just because we have all this computing power doesn’t mean a program shouldn’t be designed to use as little as possible.

Your old programs will never suffer from too much CPU power!

3 Likes
6

Well… I played Wing Commander 1 on a 80486 and man that game was way too fast!! :smiley:

But yeah. Running some basic GUI like on Android would already kill any PC from ten years ago. It is ridiculous.

abstraction

7 Likes
7

Stunning.
Your mention of top 500 supercomputers brought me to this page, and then I went and looked up what the best supercomputer in the world was when I was born. It was literally 100 times worse than an RTX2080Ti. I tend to think I am ‘not old’, bu these things give me a taste of what you older guys who played the first flight sims must feel like. Where did the years go?

2 Likes
8

Ooohh that looks like fun, great idea!

Mine is this one.

It is almost exactly 1000 times slower than my phone, LOL.

2 Likes
9

Cray -1 for me. I can’t find testing data for the Pixel XL so no clue what the order of magnitude would be.

1 Like
10

Early on in my career I wanted to work for SSI, the company Steve Chen formed after he left Cray. I was working at a company that made super mini-computers and was fascinated by the technology.
When I told my wife this she asked where it was. I said, Eau Claire, Wisconsin to which she said, “Cya…”
We stayed in south Florida for a while :slight_smile:

2 Likes
11

I found a benchmark, it puts the Pixel XL into a similar range as the Samsung S9.
Edit: found conflicting numbers for both the Snapdragon 821 (pixel) and the Snapdragon 845 (s9), but both are above 600 GFLOPS and below 800 GFLOPS.

Edit: it also seems like I have misread some numbers I used for my original post. The true numbers for my phone are probably a bit lower, more like 700 GFLOPS instead of 900, but still impressive.

Edit: and of course this is still ongoing.
There are already mobile phone processors in the works that easily exceed 1 TFLOPS.

1 Like
12

Just for laughs I tried to figure out what was the first PC I had that exceeded 1 GFLOPS.
It was the Pentium III, 1000 MHz that I got back in 2000, with 2 GFLOPS.
Its predecessor was a Pentium 400 from 1997 that could do around 400 MFLOPS.

Then I tried to figure out which PC of mine was the last one that gets beaten by my current phone.
It looks like a decent PC from 2010 (especially with a then current gaming GPU, as GPUs are much better in FLOPS) can beat my phone, but back then I still had my GeForce 9800 GT and that one loses.
So for me the answer is: the one from 2008 that I had until I bought a new one in 2013 (I didn’t play a lot of games that needed a fast PC during that time).

13

That means in 10 years our phones can run MSFS 2020.

Expensive standalone VR headsets in 7 years/ish.

Maybe Facebook is right.

1 Like
14

Unfortunately Moore’s law is starting to fail, at least for single cores. We don’t see these massive performance gains any more like we used to.

15

We didn’t notice as we were playing games on our phones and watching cats jumping into the box :wink:

Or doing just that on our phones is not the future yet :slight_smile:

3 Likes
16

My first computer, circa 1986. A Heathkit - with the Gunship keyboard overlay and a Falcon F-16 manual on the side -

Sample_125
Sample_125729×528 102 KB

Hey @Dark_Star - that is our desk!

9 Likes
17

Hey y’all!

Anyone remember this thread from… looks at the year 2019 …the distant past?

I am looking to build/buy a new PC soon so I am back at reading numbers, which reminded me of this thread.

My phone now (a Samsung Galaxy S23) seems to be able to do 2200 GFLOPS on the GPU alone. Weirdly I couldn’t even find a precise number for the CPUs (it is complex, you have to combine those multiple different cores) but people put it a bit below 1000 GFLOPS.

The combined GFLOPS number I have seen claimed is somewhere between 3000 and 3500. Let’s take the low number, 3000 GFLOPS.

So look! Here I am, calling 3 TERAFLOPS (!!!) a low number. :crazy_face:
That means that since my first post five years ago the raw computing power (which I’d like to remind us all, does not equal performance) of a decently priced phone has a bit more than quadrupled.

So my current phone is now roughly four million times faster than my first PC.

Next, let’s look at the CPUs and GPUs of proper PCs to blow our minds:

As mentioned above I am considering buying a new PC either toward the end of the year or in Q1 2025.
I am not made out of money so I cannot consider the fastest options for CPU or GPU. I also plan to run Linux on the PC so I will probably go for AMD CPU and GPU (the first time since the Radeon 9800 Pro in 2003) because AMDs drivers are a bit better.

So…

As far as I can see the AMD Ryzen 9 7950X seems to be one of the fastest affordable (below 1000€) CPUs right now.
It was surprisingly hard to find a raw FLOPS claim for it, that’s because nowadays performance in real world applications is just more important. But I found one, based on some benchmark method called HPCC. It puts the CPU at around 8000 GFLOPS. Impressive but not too outlandish.
(Man, I am jaded, those are 8 TERAFLOPS!)

But the GPU is where the fun really starts. We probably all know that GPUs are vastly more powerful in pure computing power, but looking at the numbers is still… wow.

The GPU chipset I am looking at is the AMD RX 7900 XT, a mid-level one. You can get graphics cards with it for around 750€.
That GPU is not nearly at the top. It isn’t bad by any means, but there are ones on the market (Nvidia RTX 4090, for around 2000€) that have 40% or so better performance.

The FLOPs claim is 51.48 TFLOPS (Tera!) for 32bit precision floating point operations, and 103 TFLOPS for 16 bit precision.

I’ll use the lower 32bit one for the comparisons, because 32bit floating point operations are what I believe the ones that we have been doing since the 80386 in the late 1980s.

In fact I found an old article about a 80386 variant with a 1167 maths coprocessor proposed by Weitek Corp, that had a FLOPS claim of 0.32 MFLOPS in 1987 (funnily enough the article calls 32bit “double precision”. That’s because 16bit processors were the norm).

Anyway, here we are now with the raw numbers. So what can we do with them now?
First thing we can do is comparing them with our numbers from 2019.

Back then the 2070 super (which is still in my PC right now) had - depending on who you ask - something between 9 and 14 TFLOPS. I’ll take 10 here.
So that makes the RX7900XT (from 2023) around 5 times as fast.
The new GPU is also around 16 times faster than the new phone, while the old GPU was around 10 times as fast as the old phone.

So it looks like if we roughly quadrupled raw performance in five years.

The RX7900XT easily makes the top500 supercomputer list of 2010, at rank 150 in fact.

It also just makes the top500 list of November 2011, the last rank on that list has 50.94 TFLOPS.

So that’s also interesting. My old computer from 2019 would just make it onto the list of 2008 (12 years before it was built), a ~2400€ gaming PC from 2024 would still have counted as a top500 supercomputer in 2011 (13 years earlier) it seems.

(Yeah, I know, things are more complicated than that, this is just a fun thread).

So at least on this (superficial) level it seems we can see that things are slowing down a bit, as @sobek remarked correctly.

BTW, the Nvidia RTX 4090 makes the top500 list of November 2012 at rank 434 with its 82.58 TFLOPS.
Congrats to y’all who can call that card their own, that’s amazing performance.
In fact that GPU would have been the fastest computer in the world in 2004, and even the RX7900XT would have been ranked on the third place that year.

Ok, that’s it for now. Now I have to see if I really want to build that PC, but that’s a question for another thread.

1 Like
18

Do it.

Nvidia might be fast, but their asking price and power requirements moved past us all into stupidland. AMDs strategy is much more down to earth. I wish they had more success.

8-core X3D chips still rule all gaming benchmarks with plenty of slack to the next contender. Watch out for 9800X3D news next week. :star_struck:

19

Despite predictions above, Moore’s Law still works! When will we have the first quantum computing phone? Maybe my lifetime if I stop drinking. Nah, not worth it.

20

Somewhat, if you see it as raw number of transistors per the entire CPU, perhaps. Transistor density, nope.