YOU Get a Supercomputer!

[jakkuh_t]

 

[Kazooduck]

Please wait for everyone to get a regular computer  

[Outdoors-Filter8]

Any reason you'd need a Workstation like this in your home instead of having them in a server room and remoting into it?

[Ripred]

literally just saw the YouTube notification jakkuh is on top of things

[Ripred]

think I'm going to get one for facebook and checking emails

[Hutch]

What's the model of those pumps? I'd like to take a closer look at those.

[Radium_Angel]

26 minutes ago, Outdoors-Filter8 said:

Any reason you'd need a Workstation like this in your home instead of having them in a server room and remoting into it?

Bragging rights?

[AyesC]

1 hour ago, Outdoors-Filter8 said:

Any reason you'd need a Workstation like this in your home instead of having them in a server room and remoting into it?

Being able to access the machine over LAN is far better than remoting into it over the internet for high bandwidth or low latency applications

[Crabnutz]

1 hour ago, Hutch said:

What's the model of those pumps? I'd like to take a closer look at those.

I'd love to know too. From a screen grab I think the model number is something like NTWB201PA1 or NTW8201PA1 but neither show up anywhere.

[JoshB2084]

Finally found to be able to play Minesweeper at insane FPS!

[ImorallySourcedElectrons]

2 hours ago, Outdoors-Filter8 said:

Any reason you'd need a Workstation like this in your home instead of having them in a server room and remoting into it?

Working on large simulation date sets comes to mind. Running paraview on a remote cluster is not fun!

[Outdoors-Filter8]

1 hour ago, AyesC said:

Being able to access the machine over LAN is far better than remoting into it over the internet for high bandwidth or low latency applications

I feel like a lot of ML/AI workloads aren't all that latency dependent or interactive, right? Like you'd mostly run training on something like this, right?

[Crunchy Dragon]

I helped build an AI workstation that cost about 25% as much as this one, and while it wasn't too much more than your typical PC/gaming enthusiast with expendable income, it was pretty wild.

 

I can definitely see the demand for a system that costs 4 times as much for people who do crazier things.

[James Weeb]

Does anyone have information on the brand/model of those QD fittings for the water lines?

[ImorallySourcedElectrons]

15 minutes ago, Eric Kolotyluk said:

The AMD Epyc version would be better, but I would like to see a version with nVidia Grace. Sorry, I just don't see any future in x86... There may be better ARM processors than Grace, but the point is, you could get more CPU power on a single 208 Volt 20 Amp circuit like I have in my man cave. AMD + Intel = Wast Heat

 

I live in Vancouver... does Linus offer facility tours?

I'd much rather see them implement FPGA co-processors, ARM vs x86 is a bit of a silly argument given how wasteful most software developers and compilers are.

[ImorallySourcedElectrons]

4 minutes ago, Eric Kolotyluk said:

Hey, as a software developer, I resemble that comment...

I am quite serious though, needless virtualisation and abstraction cost far more power than any CPU architecture decision. Take modern day desktop applications for something as simple as minesweeper, the current day implementation would bring an early 2000s workstation to its knees, while nothing particularly changed other than some minor static eye candy. 

 

6 minutes ago, Eric Kolotyluk said:

Actually, compilers for ARM, and RISC in general, are pretty good. CISC compilers have to do a lot more work second guessing the more complex micro-code of the instruction set. Given how ARM is taking over Data Centres and Super Computers, there is likely a lot more research going into RISC compilers. Now that Apple have changed the personal computing landscape with M1, M2, etc., we will see even more emphasis on better compilers.

Yes, but for workstation code you're putting them up against latest generation x86 Fortran compilers, those things are in a league of their own when it comes to optimization. They're incredibly fun to play with in combination with a cluster!

 

Quote

However, ARM vs x86 is most relevant when comparing CPU Power per Watt. I want the most CPU power possible on the end of my BC Hydro circuit, and x86 is just not the way to go.

I think a more nuanced view is applicable here, let me play the devil's advocate once more: RISC processors often need far more instructions to achieve the same result. For example, it gets especially painful for most ARM CPUs once your code is heavy on divisions that aren't easily broken up into divisions by powers of 2. CISC also gives the CPU significantly more freedom when doing OoE, since parts of instructions can potentially be planned ahead of time, which isn't necessary the case with RISC implementations since you're not necessarily sure if something else is going to touch the registers in question in between. etc.

 

I think Intel is going down the right path with their efficiency cores, it'd be sweet if they'd add FPGA cores like they did in certain Xeons.

[sub68]

5 hours ago, Outdoors-Filter8 said:

Any reason you'd need a Workstation like this in your home instead of having them in a server room and remoting into it?

I have a uncle working in research, it used to be based in New york but sense the pandamic they have spread to all over the world.

I don't know what they run, but if the main devs need to do some computing I can't imagin trying remote and having lag, it just seems like a mess

[Fendrick]

27 minutes ago, Eric Kolotyluk said:

Sorry, unless you state your actual education and professional experience, I cannot trust that you are not trolling me... People who "play the devil's advocate" are often trolls. Myself, as a scientist, I try to advocate for reality.

I have software and hardware experience since 1970, a BSc in Computer Science since 1981, an MSc in Computing Science since 1995, and a professional career since 1979, working and teaching in universities for 13 years, and the rest of the time in the real world.

+1 to this, too many motor mouths without real experience, wait until they call you a boomer. You know you have hit the jackpot then.

[igormp]

2 hours ago, Eric Kolotyluk said:

Sorry, unless you state your actual education and professional experience, I cannot trust that you are not trolling me... People who "play the devil's advocate" are often trolls. Myself, as a scientist, I try to advocate for reality.

I have software and hardware experience since 1970, a BSc in Computer Science since 1981, an MSc in Computing Science since 1995, and a professional career since 1979, working and teaching in universities for 13 years, and the rest of the time in the real world.

As uni teacher you should be well aware of the differences between RISC and CISC, Patterson's book was already available by the end of your MSc and is taught in any basic Computer Architecture class. CISC code is indeed smaller, and since you're from the old times you should be well aware of how little code space you'd need for a 8051 compared to any other RISC mcu (good look making your program as small on any modern ARM mcu ).

 

Also, making use of the credentials fallacy makes me doubt your integrity, specially with a newly made account and when the user that replied to you was just adding more information about facts. If you require every user here to state credentials before engaging with in a discussion with you, you are honestly better in some other forum.

 

Anyhow, comparing ISAs for performance per watt is pretty much stupid when the front-end only makes up for a small part of a CPU design and the execution units are the ones responsible for most of the power usage.

[ImorallySourcedElectrons]

6 hours ago, Fendrick said:

+1 to this, too many motor mouths without real experience, wait until they call you a boomer. You know you have hit the jackpot then.

If we are going to call degrees, I raise you by EE PhD. I must admit my experience with CPU design is more related to DSPs than general purpose computing. But honestly, a discussion is pointless if we're going to call degrees and years of experience as defence.

 

3 hours ago, igormp said:

As uni teacher you should be well aware of the differences between RISC and CISC, Patterson's book was already available by the end of your MSc and is taught in any basic Computer Architecture class. CISC code is indeed smaller, and since you're from the old times you should be well aware of how little code space you'd need for a 8051 compared to any other RISC mcu (good look making your program as small on any modern ARM mcu ).

 

Also, making use of the credentials fallacy makes me doubt your integrity, specially with a newly made account and when the user that replied to you was just adding more information about facts. If you require every user here to state credentials before engaging with in a discussion with you, you are honestly better in some other forum.

 

Anyhow, comparing ISAs for performance per watt is pretty much stupid when the front-end only makes up for a small part of a CPU design and the execution units are the ones responsible for most of the power usage.

8051, too bad Si Labs stopped making the fun 100 MHz one instruction per clock cycle ones.

 

But yeah, the difference between RISC and CISC has gotten really ill-defined at the hardware side.

 

And folks rarely look into what's causing that power consumption. I think many would be surprised how much small inconsequential parts contribute, just something as "simple" as PCIe interfacing can cost you dearly in your power budget.

[Fendrick]

Well I raise you my PhD in nothing, I just simply agreed because I agreed.

[bcc]

So.. at which companies do I have to apply to get this under my desk :)?

[Bramimond]

Speaking of performance gains by comparing hardware improvements with developer effort and given how this community is leaning towards gaming, I was fascinated to learn that a single developer rewrote Super Mario 64 to make it run twice as fast and implemented a split-screen multiplayer mode where Luigi is playable. Something that Nintendo famously planned but had to cut because of "hardware restraints".

 

And I feel this mentality all over the world. Websites are bloated. Operating systems are bloated. Software is bloated. "The hardware will catch up eventually, anyway".

 

No one is really interested in spending the time to optimize their code anymore, it seems.

 

 

With a few exceptions. I once found a forum where clicking on a topic brought the page up faster with everything fully loaded than my local e-reader app would turn a page.

 

It would be nice if we could have a cultural shift towards more efficient software.

[Radium_Angel]

7 hours ago, bcc said:

So.. at which companies do I have to apply to get this under my desk :)?

Gov't agencies...for whom budgets are an abstract issue. (I work for a gov't agency...cost is not considered a factor when looking into new toys....err...hardware)

[igormp]

On 1/11/2023 at 6:33 PM, Eric Kolotyluk said:

but not relevant in my previously stated context.

You could have simply mentioned that instead of bringing up credentials, and it does have a minor effect since you need to trade-off between a beefier encoder and micro-op cache vs larger instruction cache, and caches always demand some power. But I digress, those are still not as important when it comes to the bigger picture anyway.

< removed quoted part >

On 1/11/2023 at 6:33 PM, Eric Kolotyluk said:

Well, comparing Apple Silicon to AMD/Intel on performance per watt is not stupid at all, unless you are claiming Apple is stupid. I guess I should have been more specific. Largely, Apple has really good reasons switching from x86 to ARM, and I tend to respect their credentials more than Intel...

 

AMD routinely has better CPU performance per watt than Intel, but their x86 products still fall short of Apple Silicon in this metric. I really wish AMD would release their ARM projects as products...

AMD's Zen 4 perf/watt is really close to the M1/M2 since they are now on a similar node. Turns out most of Apple's perf/watt strengths actually came from a good node process. 

The reason why apple isn't using x86 is likely due to licensing, I bet they could've made a really similar x86 CPU as the M1 if they had the freedom to do so.

 

As a side note, AMD's skybridge project was basically using the same architecture for both thei ARM and x86 designs, with changes mostly on the front-end.

 

And I never mentioned comparing Apple Silicon to AMD/Intel, but rather ISAs. You can find awful perf/watt in many other ARM designs, just take a look at most CPUs based on ARM's proprietary µarches. The ones that fare best are the custom ones that could be easily done with any other ISA, be it RISC-V, Power or x86.

 

On 1/11/2023 at 6:45 PM, Bramimond said:

It would be nice if we could have a cultural shift towards more efficient software

That's a double-edged sword, since it'd take way more time to validate and deliver stuff, which means constrained cash flow.

 

The super mario example was a good one, since they couldn't manage to optimize the game further due to the lack of time. Most of the optimizations shown in performance were really simple but would take a lot of time to validate and whatnot, and it's not like you could just patch a game after the fact in the old days.

 

As donald knuth said, premature optimization is the root of all evil.

Edited January 13, 2023 by LogicalDrm