• Announcements

    • Robin

      Welcome!   03/05/2016

      Welcome, everyone, to the new 910CMX Community Forums. I'm still working on getting them running, so things may change.  If you're a 910 Comic creator and need your forum recreated, let me know and I'll get on it right away.  I'll do my best to make this new place as fun as the last one!
WR...S

Story, Wednesday September 21, 2016

Recommended Posts

19 hours ago, Vorlonagent said:

Maybe their software only leveraged the Xeons?

Nope, I monitored their performance logs and those cards were actually doing a lot of the work.

 

I've been trying to think of things that might explain the excessive demands. Things I can think of is that they were probably applying effects to simulate how the printing of the graphics on the material affected colors, and how the surface texture scattered the light and other such things. But still the combined CPU and GPU loads seems excessive for such a simple task.

Those Tesla cards were quite expensive, and yet, it's my experience, they were way more prone to die on you than the high end gaming cards that used the same GPU core generation. And the price difference was drastic. A high end gaming card would cost about $700 to $800, a high end workstation card about $4000 to $5000, and a high end GPU processing card in the range of $5000 to $7000.

I just had a look at the latest and greatest, and the current top of the line appears to be the Tesla P100, and the price for them ranges from $5900 for the 12GB version to $9400 for the 16GB SXM2 version. At those prices I sure hope those are reliable...
 

Share this post


Link to post
Share on other sites
39 minutes ago, Cpt. Obvious said:

I just had a look at the latest and greatest, and the current top of the line appears to be the Tesla P100, and the price for them ranges from $5900 for the 12GB version to $9400 for the 16GB SXM2 version. At those prices I sure hope those are reliable...

If they are like politicians I wouldn't bet my shoes on it. Just one Danish politician I can think of cost his constituency almost half a billion dollars, and he wasn't particularly reliable. And given that the municipality he came from consisted of just 20.000 people, that means he cost his citizens nearly 25.000 dollars each. Not really worth it at the price.

Admittedly he ran that up over the course of some twelve years, so they only got cheated of some two thousand dollars per head per year. Even so, they still didn't want to re-elect him once they found out. People are so ungrateful.

Share this post


Link to post
Share on other sites
8 minutes ago, The Old Hack said:

If they are like politicians I wouldn't bet my shoes on it. Just one Danish politician I can think of cost his constituency almost half a billion dollars, and he wasn't particularly reliable. And given that the municipality he came from consisted of just 20.000 people, that means he cost his citizens nearly 25.000 dollars each. Not really worth it at the price.

Admittedly he ran that up over the course of some twelve years, so they only got cheated of some two thousand dollars per head per year. Even so, they still didn't want to re-elect him once they found out. People are so ungrateful.

Sounds like he could be a distant ancestor of Kornada...
 

Share this post


Link to post
Share on other sites
On 2016-10-06 at 2:40 AM, Scotty said:

I'm thinking we have to be getting to the point where graphics quality can't be noticeably improved any further and so hardware would be able to get ahead, as far as I can tell, we've plateaued on audio quality years ago (I base this mainly on the fact that I haven't heard anyone saying I need to use X audio card instead of the one integrated with the motherboard) though audio mixing is a thing that requires different hardware altogether but can be done on the fly easily enough with little impact on CPU, most people when rendering video, can't do much else until it's done.

Audio has only plateaued in the areas of encoding, decoding, and playback of simple audio, one big thing we can't do yet without noticeable performance degradation—or heck, even doing it in the first place—is the generation of proper, real time audio effects, like echos, muffled sound, true 3d positional audio, or just how audio spreads past obstacles of different materials.

As a comparison, realistic echos require complex calculations that are at least as complicated global illumination, which has been something of a pipe dream in graphics programming for ages.

On 2016-09-23 at 2:43 AM, CritterKeeper said:

We haven't discussed this lately, have we?  Hmm....a form becomes permanent if you get pregnant while transformed....and Tedd now thinks he can figure out how to change someone else's "default" sex....so maybe you could transform into an 18-year-old girl, then get pregnant, have the baby, and then if you're male gender, you get Tedd to turn you to a now-19-year-old male version of the girl you became permanently by getting pregnant.  If they all plan ahead and the guys get gender-swapped-but-otherwise-unchanged versions of themselves saved, then they can become their own normal self but younger.

No need for the pregnancy, Tedd thinks that (s)he can make transformation wands that changes someone base form, and presumably it would work with any transformation. Just load up that "me when I was 18" form into a perma-wand and BAM! I'm definitely asking about this next Q&A.
 

Edited by lonjil
Better wording.

Share this post


Link to post
Share on other sites
On 10/8/2016 at 7:35 AM, Cpt. Obvious said:

Nope, I monitored their performance logs and those cards were actually doing a lot of the work.

 

I've been trying to think of things that might explain the excessive demands. Things I can think of is that they were probably applying effects to simulate how the printing of the graphics on the material affected colors, and how the surface texture scattered the light and other such things. But still the combined CPU and GPU loads seems excessive for such a simple task.

Those Tesla cards were quite expensive, and yet, it's my experience, they were way more prone to die on you than the high end gaming cards that used the same GPU core generation. And the price difference was drastic. A high end gaming card would cost about $700 to $800, a high end workstation card about $4000 to $5000, and a high end GPU processing card in the range of $5000 to $7000.

I just had a look at the latest and greatest, and the current top of the line appears to be the Tesla P100, and the price for them ranges from $5900 for the 12GB version to $9400 for the 16GB SXM2 version. At those prices I sure hope those are reliable...
 

Me too.  I know there are hardware differences between Workstation boards and Desktop (gaming) ones besides the one costing way more than the other.  I don't know what they actually are because I don't do much with them besides plug them into systems.  I've found regular Quadro boards to be as or more reliable than Nvidia's gaming boards using the same base chip.  I am aware Tesla is not "regular quadro". 

The Tesla boards you're talking about are the ones I'm thinking of, they should have either 2 or 4 physical GPU chips on the one card as well.  That's a lot of heat generation in a small space.  They may not be too because Nvidia Marketing has over-used the Tesla name a bit...

Share this post


Link to post
Share on other sites
24 minutes ago, Vorlonagent said:

Me too.  I know there are hardware differences between Workstation boards and Desktop (gaming) ones besides the one costing way more than the other.  I don't know what they actually are because I don't do much with them besides plug them into systems.  I've found regular Quadro boards to be as or more reliable than Nvidia's gaming boards using the same base chip.  I am aware Tesla is not "regular quadro".

The Tesla boards you're talking about are the ones I'm thinking of, they should have either 2 or 4 physical GPU chips on the one card as well.  That's a lot of heat generation in a small space.  They may not be too because Nvidia Marketing has over-used the Tesla name a bit...

Yes, current generations there are quite big differences between the gaming cards, workstations cards, and computation cards. But a few years back that wasn't always the case. In fact there were some generations where the gaming and workstation cards were more or less identical, and by changing a single resistor you could promote the gaming card to accept both bios and drivers used for the workstation model. Oh and then there were the price premium of two to four times the price for the workstation card...

AMD (ATI) and Nvidia also differ a bit in just what they add to the workstation series, but I'm not up to date on the details for current generations. Differences between workstation cards and computation cards are mainly that the later has no, or extremely gimped support for a monitor, the amount of memory and ECC support for the computation card, and as you wrote the number of GPU's on board.

There is also quite strict guide lines for the cooling of these, and the one thing not found in these is a upper limit for sound... Pushing that much air though a chassis isn't something I think can be done quietly. I remember one computer manufacturer who marketed a quiet workstation with incredible specifications. I got hold of one of those and loaded it up with my usual stress test cocktail and stood back as the machine apparently was revving up for takeoff. At full tilt it could compete with some of the nosiest 19" rack servers I've built. Turns out there's not an easy way to quietly cool a computer that pulls more than a KW at the wall, but I pity the users who bought into the "Quiet" hype. Yes it was quiet, if you didn't push it. Load it down with some well optimized software that can max out the CPU's and GPU's concurrently and you will want to invest in earplugs.
 

Share this post


Link to post
Share on other sites
3 hours ago, Cpt. Obvious said:

Turns out there's not an easy way to quietly cool a computer that pulls more than a KW at the wall

That shouldn't be surprising. Remember that computer is very similar to electric heater: almost ALL energy you put into it will change to heat. 500W immersion heaters are small but still practical.

There ARE steps which can make computer quieter: using bigger heatsinks and bigger fans, basically. But neither will make computer quiet - just quieter. Only removal of all fans will make the computer quiet ... but usually, also dead.

Share this post


Link to post
Share on other sites
5 hours ago, hkmaly said:

There ARE steps which can make computer quieter: using bigger heatsinks and bigger fans, basically. But neither will make computer quiet - just quieter. Only removal of all fans will make the computer quiet ... but usually, also dead.

There are systems for water-cooling computers. This gets the heat away from the critical spots to places where there is more room for bigger radiators - get big enough and you don't need fans on them.

I have never had a computer hot enough to need such a thing.

Share this post


Link to post
Share on other sites
1 hour ago, Don Edwards said:

There are systems for water-cooling computers. This gets the heat away from the critical spots to places where there is more room for bigger radiators - get big enough and you don't need fans on them.

I have never had a computer hot enough to need such a thing.

Some of the late model G5 towers were "water" cooled.  I strongly suspect that fact that they never could get the heat problems solved on G5  chips is why Apple went Intel.  For example even the "cool running" G5 chips were too hot to make a laptop with.  The pre Intel laptops were over clocked G4

Share this post


Link to post
Share on other sites
3 hours ago, mlooney said:

Some of the late model G5 towers were "water" cooled.  I strongly suspect that fact that they never could get the heat problems solved on G5  chips is why Apple went Intel.  For example even the "cool running" G5 chips were too hot to make a laptop with.  The pre Intel laptops were over clocked G4

Pffft. Liquid helium is where it's at. :demonicduck:

Share this post


Link to post
Share on other sites
18 hours ago, Don Edwards said:
23 hours ago, hkmaly said:

There ARE steps which can make computer quieter: using bigger heatsinks and bigger fans, basically. But neither will make computer quiet - just quieter. Only removal of all fans will make the computer quiet ... but usually, also dead.

There are systems for water-cooling computers. This gets the heat away from the critical spots to places where there is more room for bigger radiators - get big enough and you don't need fans on them.

I have never had a computer hot enough to need such a thing.

The water is too dense to flow freely fast enough. You need to have some turbine pushing it, and this turbine is noisy. Sure, not as noisy as air fan, but still ...

(You may get under the HDD noise, but if you mean it with being quiet, you are using SSD's anyway.)

13 hours ago, The Old Hack said:

Pffft. Liquid helium is where it's at. :demonicduck:

Normal computers wouldn't be able to work at such temperatures, they NEED the metals to have resistance. You need special construction for superconducting computers.

For normal computers, liquid nitrogen is more than enough.

Share this post


Link to post
Share on other sites
On 10/7/2016 at 0:02 PM, Cpt. Obvious said:

Hmm... Yes, I should have caught that myself. But what we can take away from this is that a top of the line PC CPU today is way faster than a one off speed monster back in the early eighties.

Sometimes I see things that makes me wonder just what people are doing wrong when they work with 3D. I built a computer for a customer a few years back. It had dual Xeon processors, the fastest available at the time, and eight (8!) Tesla cards, again the fastest available, might have been K80 or M60. This was used to provide real time rendering of 3D models for the products that they sold. Their customers would log in on a website and use a web app that allowed them to slap graphics on the products and see what they would look like, supposedly in real time. I got to see the web app working and I was not impressed by the performance, and most importantly I can't see how what they were doing could require the immense amount of computational power that they had in that machine. But they wanted eight Tessla cards so they got eight Tessla cards. At full load that machine could draw almost 3KW of power... I did some burn in at full tilt and boy was it loud and hot.
 

Is it weird that I heard someone mention Elon Musk trying to create an AI using a computer that can do 170 teraflops, do a search for what can do 170 teraflops, and find basically what you described?

Share this post


Link to post
Share on other sites

AI isn't just about sheer processing speed--it also needs sufficient RAM (or soft-wired circuits) to simulate all of the thought processes. Then of course there's the software side of things--we don't know enough about how our own intelligence functions to be able to devise algorithms for it, so simply throwing extra processing power and memory at the problem is one of those "throw everything at the wall and see what sticks" approaches, more or less.

Share this post


Link to post
Share on other sites
7 hours ago, ijuin said:

AI isn't just about sheer processing speed--it also needs sufficient RAM (or soft-wired circuits) to simulate all of the thought processes. Then of course there's the software side of things--we don't know enough about how our own intelligence functions to be able to devise algorithms for it, so simply throwing extra processing power and memory at the problem is one of those "throw everything at the wall and see what sticks" approaches, more or less.

It's not necessarily a totally bad approach. I wouldn't like to try to program AI on a VIC-20.

Share this post


Link to post
Share on other sites
32 minutes ago, ijuin said:

Yes, processing power is necessary, but not sufficient.

To be fair... ahem... there are those who spend so much time bragging about the processing power of their rigs that they make me wonder if they are compensating for something. I have more than once seen people throw massive amounts of money at a monster hugely more powerful than what they really needed.

Share this post


Link to post
Share on other sites
13 hours ago, ijuin said:

AI isn't just about sheer processing speed--it also needs sufficient RAM (or soft-wired circuits) to simulate all of the thought processes. Then of course there's the software side of things--we don't know enough about how our own intelligence functions to be able to devise algorithms for it, so simply throwing extra processing power and memory at the problem is one of those "throw everything at the wall and see what sticks" approaches, more or less.

Well ... our intelligence works using 100 billion neurons all running in parallel, without common clocks and in analog mode. Computers are not really good in running so many things in parallel, wouldn't work without single clock for whole chip and are strictly digital. Even ignoring the problem of how those neurons are connected (which is where the intelligence actually is), it's obvious that direct 1:1 simulation would be extremely ineffective. But ... can intelligence appear if we try something different than 1:1 simulation?

Share this post


Link to post
Share on other sites

Our intelligence also evolved using genetic algorithms, which require sufficient hardware to run large numbers of independent experiments in parallel.

Considering that each intelligent entity is likely to involve a massively-parallel processor...

Developing AIs the same way would take quite a large amount of hardware.

Share this post


Link to post
Share on other sites
18 hours ago, Don Edwards said:

Our intelligence also evolved using genetic algorithms, which require sufficient hardware to run large numbers of independent experiments in parallel.

Considering that each intelligent entity is likely to involve a massively-parallel processor...

Developing AIs the same way would take quite a large amount of hardware.

Some of the science fiction I've seen involving AIs often has it appear spontaneously from very large computer systems devised to operate in parallel. Good examples would be Skynet from the Terminator movies and Mycroft from Heinlein's The Moon is a Harsh Mistress. (Which last I highly recommend -- it is some of Heinlein's best work, I feel.)

Share this post


Link to post
Share on other sites
1 hour ago, The Old Hack said:

Some of the science fiction I've seen involving AIs often has it appear spontaneously from very large computer systems devised to operate in parallel. Good examples would be Skynet from the Terminator movies and Mycroft from Heinlein's The Moon is a Harsh Mistress. (Which last I highly recommend -- it is some of Heinlein's best work, I feel.)

Yeah ... and the most likely computer system AI can spontaneously evolve from this way is probably Google.

On the other hand, other science fictions have AI developing differently, with greater influence of chance ... like, sentience evolving in computer virus which was hiding in unreliable video RAM ...

Share this post


Link to post
Share on other sites
6 hours ago, The Old Hack said:

Some of the science fiction I've seen involving AIs often has it appear spontaneously from very large computer systems devised to operate in parallel. Good examples would be Skynet from the Terminator movies and Mycroft from Heinlein's The Moon is a Harsh Mistress.

 

5 hours ago, hkmaly said:

Yeah ... and the most likely computer system AI can spontaneously evolve from this way is probably Google.

There's a good murder mystery series by Donna Andrews which she started back at the turn of the century after a challenge to do a story with a protagonist that had never been done before.  Turing Hopper is an artificial intelligence program that started out as a library research computer.  (Google existed at this point, but wasn't nearly as ubiquitous as it is now.)  The service used several different versions of the assistance program, with different "personalities" to appeal to different tastes, and Turing Hopper was the most popular and thus got more and more resources.  She could adapt herself to make herself more useful to customers, and decided that learning how to present a more appealing and helpful personality, as well as getting better at her job, would fulfill that purpose and keep herself 'alive' to boot.  Sentience developed from there.

Share this post


Link to post
Share on other sites
16 hours ago, CritterKeeper said:

There's a good murder mystery series by Donna Andrews which she started back at the turn of the century after a challenge to do a story with a protagonist that had never been done before.

This kind of challenge gets harder and harder every day ...

 

Share this post


Link to post
Share on other sites
3 hours ago, hkmaly said:
19 hours ago, CritterKeeper said:

There's a good murder mystery series by Donna Andrews which she started back at the turn of the century after a challenge to do a story with a protagonist that had never been done before.

This kind of challenge gets harder and harder every day ...

While not quite the same thing, are you aware that Robert A Heinlein wrote, as R. A. Heinlein, three stories for teen age girls that were publish in "Calling All Girls" of all places.  He like the central character so much she lost some weight, got a name change to Podkayne and moved to Mars.

Share this post


Link to post
Share on other sites
8 minutes ago, mlooney said:
3 hours ago, hkmaly said:

This kind of challenge gets harder and harder every day ...

While not quite the same thing, are you aware that Robert A Heinlein wrote, as R. A. Heinlein, three stories for teen age girls that were publish in "Calling All Girls" of all places.  He like the central character so much she lost some weight, got a name change to Podkayne and moved to Mars.

The "should have stayed on Mars" Podkayne? No, I wasn't aware, but it won't tell me much as I don't know what "Calling All Girls" is.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now