72-bit cpu 72-bit Processor

[Schyken]

Right, so AMD created the whole amd64 thing. A 64-bit x86 instruction set architecture.That's great. We'll be using that technology for quite a few years into the future. Now, I've been studying first, the i386 processors and then the amd64 processors, how they work and process data. (Honestly, nothing has changed at all over the years other than size and transistor count.) So, after quite a bit of think-tanking with a few people, I've done some work on my own and have come up with a 72-bit x86 based instruction set processor. Note: It hasn't been built yet, at least not in the sense of a finished CPU that can be strapped into a motherboard, however everything has been run through extensive simulations and virtual testing, and it appears to be very effective. Granted, 72-bit is an INSANE thing to consider... Especially the memory part... Don't worry, I've been working on something pretty neat for that too. Now, I have no real application for something like this; at least not a practical one. I will go on to say, that for now, I am mainly working on architectural improvements for efficiency and performance. The model I've been tinkering with is a single core processor, yet seems to surpass performance in testing than it's amd64 first generation counterpart. That being said, this is all simulation and there's no way for me to manufacture this, especially at such a small scale. Let me know if you'd like to know more about this. I can probably go into a bit of detail as to how I'm handling development. I can't exactly put screenshots of anything in here, as I do have future plans to sell if things go as I hope. A processor like this has much more application in server and supercomputing than in general computing. That being said, will probably be more effective for certain projects like folding. I'm really working on creating a dual-core version of this at the moment, however architectural improvements will help me the most. Also, I think I should state that it is incredibly hard to run simulations on this... It is so complex, my system gets really toasty just loading it in. I'm pretty sure I'm going to have to invest in some more suited hardware if I plan to improve upon this.

[Nena Trinity]

Why not 128bit? Also I do not think such a thing is needed yet is the reason for it not to be made...

[Arcanekitten]

Sounds neat, Keep going.

[Virtual01]

How did you become a wizard who can design and simulate new types of processors?

[MoYo]

What would be the benefit to a 64-bit processor? To my knowledge the main reason why we went to 64-bit is because of the memory limit of 32-bit (around 4gb). With 64-bit this is 16 exabytes (10^9 gb) and we are no where near that, so would we really need something like that? Even in servers.

 

Also wouldn't 128-bit be more logical?

 

Still sounds awesome to tinker with though!

[deWaardt]

4 minutes ago, Virtual01 said:

How did you become a wizard who can design and simulate new types of processors?

I think architecture simulation is possible, but how the hell do you simulate 72 bit on a 64 bit CPU?

I mean... There's no single way to get 64 bit code to run on a 32 bit CPU, not even using emulation (if I'm right that is)

[SSL]

Sure you have, which is why you're announcing this on a tech forum for teens in a garbled wall of text.

 

Give a presentation at a major conference or publish a paper in a respected journal and I'll be impressed.

[Schyken]

23 minutes ago, Nena360 said:

Why not 128bit? Also I do not think such a thing is needed yet is the reason for it not to be made...

The reason for 72 instead of 128 is because unfortunately, I don't really have the knowledge nor attention span to handle something quite like that yet.

 

12 minutes ago, MoYo said:

What would be the benefit to a 64-bit processor? To my knowledge the main reason why we went to 64-bit is because of the memory limit of 32-bit (around 4gb). With 64-bit this is 16 exabytes (10^9 gb) and we are no where near that, so would we really need something like that? Even in servers.

 

Also wouldn't 128-bit be more logical?

 

Still sounds awesome to tinker with though!

Yeah, honestly, 128-bit does make more sense, I'm just not quite there yet. The reason for the jump, by the way, is it will allow me to use a new form of RAM that is a lot more similar to the CPU cache without the latency of travel from another part of the motherboard. The issue with this, is because I will have a lot more 'bridge' points between individual modules, every set of data (1/0) ends up being multiplied exponentially because of the 3D configuration of both actual storage points and logic processors, essentially being about 3-4x more effective than HBM while at the same time having such a low latency (about CL- 8 @4800MHz (NOTE: VERY THEORETICAL... BASED ON SIMULATIONS))

 

20 minutes ago, deWaardt said:

I think architecture simulation is possible, but how the hell do you simulate 72 bit on a 64 bit CPU?

I mean... There's no single way to get 64 bit code to run on a 32 bit CPU, not even using emulation (if I'm right that is)

This is actually a funny story... I started off doing it inside of a program I made that ran heavily off of the GPU, using floating point as my main calculation, however this became incredibly difficult after it got more complicated and I ended up killing a pair of 980s... I then resorted to using a dual CPU motherboard and I ordered a couple of 18 core CPUs and had to write a completely new (command line) Operating System off of the Linux kernel, this essentially taking mw about three months just to get it to boot... I then rewrote the program I had originally used to function based on CPU calculations rather than GPU, eventually killing both processors and having to put I two new ones. I eventually got it working to where, upon boot, passthrough of only certain parts of one processor and the entirety of the other created a direct passthrough, allowing me to, at the very least, run a simulation. Now, as it gets more complex, the issue becomes that I have already surpassed what both of these CPUs together are capable of, and ended up resorting back to using floating point. I'm currently running 3 Titan X's, putting me at a total loss on this project of about $13k. That being said, I did actually manage to create an operating system that ONLY runs off of the GPU. The downside of this, is that there is nothing else I will ever be able to use the hardware for.

[Djole123]

That thing would be AWESOME!!!!

 

I can only imagine the use case scenario where you need a ton of memory (like in a datacenter), and for consumers would mean a bit of performance boost!

I guess you can contact VIA about it, so if they make a 72bit CPU, that would make them (and you) some money!

[Schyken]

17 minutes ago, SSL said:

Sure you have, which is why you're announcing this on a tech forum for teens in a garbled wall of text.

 

Give a presentation at a major conference or publish a paper in a respected journal and I'll be impressed.

I don't need you to be impressed. I posted this to see what people thought of the idea, as well as get some opinions. Feel however you'd like, I remain determined to complete this project successfully, should I not go broke beforehand.

[BuckGup]

3 minutes ago, Schyken said:

I don't need you to be impressed. I posted this to see what people thought of the idea, as well as get some opinions. Feel however you'd like, I remain determined to complete this project successfully, should I not go broke beforehand.

How old are you and where did you study?

[SSL]

2 minutes ago, Schyken said:

I don't need you to be impressed. I posted this to see what people thought of the idea, as well as get some opinions. Feel however you'd like, I remain determined to complete this project successfully, should I not go broke beforehand.

 

You sound fantastically delusional, but if Intel announces the move to 72-bit server architecture within the next 10 years I'll concede that I was wrong.

[Schyken]

8 minutes ago, SamStrecker said:

How old are you and where did you study?

I'm actually 16 years old. My father is an engineer at Raytheon and his father was an engineer at IBM. (Funny how that works, huh?). I took classes at University High (AZ) in computer engineering, so that helped a lot too. A lot of it really, is just taking things apart, studying them, realizing they're way too small to study, and then reading almost every documentation I could find and testing it.

 

7 minutes ago, SSL said:

 

You sound fantastically delusional, but if Intel announces the move to 72-bit server architecture within the next 10 years I'll concede that I was wrong.

Well, if I'm going to be delusional, I guess fantastically is the best way to be

[BuckGup]

4 minutes ago, Schyken said:

I'm actually 16 years old. My father is an engineer at Raytheon and his father was an engineer at IBM. (Funny how that works, huh?). I took classes at University High (AZ) in computer engineering, so that helped a lot too. A lot of it really, is just taking things apart, studying them, realizing they're way too small to study, and then reading almost every documentation I could find and testing it.

Could you post a video on this

[juretrn]

OK...

so someone without any real education in engineering can just throw together an architecture these days... whaddya know.

 

Tell me what you based your design upon. What did you change? How? 

I would love to see some technical documentation, or at least some kind of schematic.

 

Also, how did you test your design? You have access to the 10k $ (per year) Cadence licence?

I'm sorry if I have my doubts. 

[Schyken]

2 minutes ago, SamStrecker said:

Could you post a video on this

Umm, probably. I don't really know what I'd say though

[BuckGup]

2 minutes ago, Schyken said:

Umm, probably. I don't really know what I'd say though

Just a video of it in action.

[Schyken]

2 minutes ago, juretrn said:

OK...

so someone without any real education in engineering can just throw together an architecture these days... whaddya know.

 

Tell me what you based your design upon. What did you change? How? 

 

Also, how did you test your design? You have access to the 10k $ (per year) Cadence licence?

I'm sorry if I have my doubts. 

No, I can understand why you'd have doubts. My design was originally based off of amd64, granted, almost all of modern day processors are these days. I eventually got to the point where it wasn't possible to get 72-bit working based on amd64 once I surpassed 37k transistors. It just didn't handle very well. So, I started looking at ARM processors that understand x86 and there was so much more room for improvement, I stuck with it. That being said, I think that qualifies it as an ARM -based x86 processor. On the note of what I had to change... Almost everything. It's not necessarily that I had to completely replace everything, its that I had to add so many more paths, that I had to rework the actual architecture itself. And because I'm not an engineering genius, doing this effectively is really hard.

[Schyken]

6 minutes ago, SamStrecker said:

Just a video of it in action.

The system running it? Or the processor itself? There's no physical product at this point, everything except for a very primitive model I made is all virtual.

[BuckGup]

12 minutes ago, Schyken said:

The system running it? Or the processor itself? There's no physical product at this point, everything except for a very primitive model I made is all virtual.

Yeah just the virtual stuff and your setup

[Schyken]

2 minutes ago, SamStrecker said:

Yeah just the virtual stuff and your setup

Yeah, I could do that. Probably show off the fried stuff as well. Honestly, I got really scared when I fried the GPUs because I had never smelled burning PCBs before then and I had actually run around the house looking for fire, not realizing it was underneath my desk.

[BuckGup]

Just now, Schyken said:

Yeah, I could do that. Probably show off the fried stuff as well. Honestly, I got really scared when I fried the GPUs because I had never smelled burning PCBs before then and I had actually run around the house looking for fire, not realizing it was underneath my desk.

How exactly did you kill them?

[Schyken]

1 minute ago, SamStrecker said:

How exactly did you kill them?

Sadly, it was really lame. I had actually attempted to run my simulation as if the GPUs were functioning as the actual processor itself, and I'm not actually sure why it completely fried, my guess is it was pretty much like plugging a fork into an electrical outlet. They're just not compatible. I kinda wish I was able to log what happened, because that could actually help me designing a real physical processor. To clarify, when I say fried, I mean even the PCIE slot melted in. It was probably the saddest way I could have killed some 980s. The CPUs getting killed, that was me putting waaay to many volts into something that should barely have a couple.

[BuckGup]

3 minutes ago, Schyken said:

Sadly, it was really lame. I had actually attempted to run my simulation as if the GPUs were functioning as the actual processor itself, and I'm not actually sure why it completely fried, my guess is it was pretty much like plugging a fork into an electrical outlet. They're just not compatible. I kinda wish I was able to log what happened, because that could actually help me designing a real physical processor. To clarify, when I say fried, I mean even the PCIE slot melted in. It was probably the saddest way I could have killed some 980s. The CPUs getting killed, that was me putting waaay to many volts into something that should barely have a couple.

Huh

[deWaardt]

I'm not sure, not sure at all, but this is starting to sound a little bit like bull.