5 ms·

It was a common practice in the 90s for creative engineers to use extremely expensive "supercomputer" workstations to pay for productivity gains & live on the b

It was a common practice in the 90s for creative engineers to use extremely expensive "supercomputer" workstations to pay for productivity gains & live on the bleeding edge (e.g. Silicon Graphics workstations, NeXT workstations, and so forth). Question: What is the equivalent way to do this today? That is, is there a way to pay a lot of money to use a computer which is 5-10 years ahead of its time? Ok so I'm pretty biased here, but I think the answer lies in VR computing. There's no doubt VR computers are more expensive than their PC/laptop counterparts, but they allow you to adopt a bleeding edge technology which is essentially 5+ years ahead of its time in terms of where it is on the "commodity computing" frontier. A good quote from Alan Kay I find pretty inspirational on this front: https://youtu.be/id1WShzzMCQ?t=3345 https://youtu.be/id1WShzzMCQ?t=3345 Here he basically advocates for spending hundreds of thousands of dollars on a computing machine, in order to compute on what will be a commodity product 10-15 years into the future. VR computers aren't this extreme on the cost curve, but I think there is something to this point of view which I find really inspirational.[1] [1] Caveat: I'm one of the founders of SimulaVR (https://simulavr.com https://simulavr.com), so admittedly am very biased here. But I do think VR provides a way to convert money into "better compute" in a way that hasn't been available since the 70s-90s super workstation era.

arapacana 5y ago
I respect the smoothness and honesty of your plug.
Keyframe 5y ago
You could shell out for Nvidia's DGX A100. Not the same level of futureness compared to current off the shelf stuff as this was back then.
wongarsu 5y ago
I think todays equivalent is more running a desktop with a 64 core CPU and 128 GB RAM (or more), with as many screens as you want (I don't find more than two useful, but preferences vary). I could see VR computing as the mobile version of that. Laptops can be pretty powerful, but their screens are limiting. VR gives you the screen real estate without requiring you to set up monitors where you are.
anyfoo 5y ago
> Question: What is the equivalent way to do this today? I don't think there is an equivalent, because as I explained here, the very noticeable, very significant gains in almost every aspect that even simple hardware upgrades brought haven't been a thing for a long time: https://news.ycombinator.com/item?id=30951533 https://news.ycombinator.com/item?id=30951533 It's not just that back then an "extremely expensive supercomputer" workstation effectively enabled you to do a lot of things that we take for absolutely granted nowadays in the first place. When not having such a beast meant essentially not being able to do any serious CAD (whether mechanical or electrical engineering), large scale software engineering, number crunching, anything related to graphics/visuals... It's also that super mundane tasks like writing a letter, editing an image, or even just booting your computer received a significant, immediately noticeable boost for comparably plain seeming hardware upgrades like a 33MHz higher clock or a bit more L1 cache.
- zozbot234 5y ago
  True, but I think hyperscale gear that's $200k nowadays from specialty vendors (much like those workstations of old) might easily become "commodity" enough in the next 10 to 15 years, and that's what Alan Kay was really trying to get at. What might that be useful for, in the hands of a single person? That's a really interesting question, and I don't have a sure-fire answer. (Even AI inference on the huge models that seem to be popular as of late - an easy choice for something that will chew through a lot of data and compute - feels like way too much of a toy to be genuinely useful.)
aabajian 5y ago
I'm surprised nobody has stated the obvious. See the recent post on Dall-E 2. It's likely to change the face of creative design and was done using modern supercomputers (e.g. cloud-based deep learning). To make a direct analogy, I imagine some of the newest game AI-graphics and engines require training on such supercomputers.
asdff 5y ago
Today's equivalent is probably just a cluster and requesting more cpus or gpus vs having some ahead of time chip locally. I guess you could argue the filesystems on clusters tend to be pretty performant too so maybe that's ahead of time in terms of consumer ssd speeds.
blenderdt 5y ago
Not a lot of people put two RTX3090 cards in there desktop for over $2000 each, but I know some who do. When rendering times for production work is cut in half it pays back fast.

arapacana 5y ago

I respect the smoothness and honesty of your plug.

Keyframe 5y ago

You could shell out for Nvidia's DGX A100. Not the same level of futureness compared to current off the shelf stuff as this was back then.

wongarsu 5y ago

I think todays equivalent is more running a desktop with a 64 core CPU and 128 GB RAM (or more), with as many screens as you want (I don't find more than two useful, but preferences vary). I could see VR computing as the mobile version of that. Laptops can be pretty powerful, but their screens are limiting. VR gives you the screen real estate without requiring you to set up monitors where you are.

anyfoo 5y ago

> Question: What is the equivalent way to do this today? I don't think there is an equivalent, because as I explained here, the very noticeable, very significant gains in almost every aspect that even simple hardware upgrades brought haven't been a thing for a long time: https://news.ycombinator.com/item?id=30951533 https://news.ycombinator.com/item?id=30951533 It's not just that back then an "extremely expensive supercomputer" workstation effectively enabled you to do a lot of things that we take for absolutely granted nowadays in the first place. When not having such a beast meant essentially not being able to do any serious CAD (whether mechanical or electrical engineering), large scale software engineering, number crunching, anything related to graphics/visuals... It's also that super mundane tasks like writing a letter, editing an image, or even just booting your computer received a significant, immediately noticeable boost for comparably plain seeming hardware upgrades like a 33MHz higher clock or a bit more L1 cache.

zozbot234 5y ago
True, but I think hyperscale gear that's $200k nowadays from specialty vendors (much like those workstations of old) might easily become "commodity" enough in the next 10 to 15 years, and that's what Alan Kay was really trying to get at. What might that be useful for, in the hands of a single person? That's a really interesting question, and I don't have a sure-fire answer. (Even AI inference on the huge models that seem to be popular as of late - an easy choice for something that will chew through a lot of data and compute - feels like way too much of a toy to be genuinely useful.)

aabajian 5y ago

I'm surprised nobody has stated the obvious. See the recent post on Dall-E 2. It's likely to change the face of creative design and was done using modern supercomputers (e.g. cloud-based deep learning). To make a direct analogy, I imagine some of the newest game AI-graphics and engines require training on such supercomputers.

asdff 5y ago

Today's equivalent is probably just a cluster and requesting more cpus or gpus vs having some ahead of time chip locally. I guess you could argue the filesystems on clusters tend to be pretty performant too so maybe that's ahead of time in terms of consumer ssd speeds.

blenderdt 5y ago

Not a lot of people put two RTX3090 cards in there desktop for over $2000 each, but I know some who do. When rendering times for production work is cut in half it pays back fast.