(Sorry for the off topic post -- I hope some of you will find it informative -- and some of those people might even have mod points to burn:-)
Clearcase sucked big time -- at least the last time I looked (admittedly years ago) Performance just went in the toilet as soon as you started throwing apps built from 10's of thousands of cpp files. Never mind having hundresds of developers manipulating 10s of versions of everything.
We use Perforce, and it rocks -- handles branching and mergeing great, scales to ridiculously large projects, is rock solid, and works on Irix, Solaris, HPUX, AIX, Linux, Windows and Mac OSX. (And yes, we need it to run across that many platforms.)
A careful study of the asm docs will lead you to the conclusion that it's aimed at four seperate target markets;
1: Gamers / graphics;
-- 4 deep vectors of single precision float -- various interesting but ulimately useless-for-games parts of the IEEE FP spec not implemented.
2: Video codecs. Parent post has the details
3: Crypto -- Support for arbitrary precision math. (details in parent post). popcount, count leading zeros. I expect various 3 letter agencies to buy boatloads of these things.
4: Supercomputing math geeks. 2 deep Double precision vectors with full IEEE support and compliance. (Currently not fully pipelined, but I expect that that situation will not last for long.)
What's missing for me;
- no saturated fixed point math.
- round, ceil, floor, trunc are all very slow (compared to everything else).
Those estimate instructions did not place useful information in the low order bits of the estimate. Those processors did not have an instruction whose only porpose was to process those special low order bits to produce more accurate results. This aspect of Cell is really new.
Not just buzzword compliant, but confusing as hell for those of us who have been in the know for a while.
To me, SPU always made me think "Scalar Processing unit", while PPE made me think "Parallel Processing Element".
Of course that's exactly backwards.
That, and I choke on words like "synergistic" because they peg my bullshiat-o-meter way off in the red.
In my opinion one of the coolest features of this architecture are the way the reciprocal estimate and reciprocal square root estimate instructions work.
In a single cycle you get 13 good bits of precision -- with the low order bits filled with information to be used by the floating point interpolate instruction.
You can get a full precision (32 bit ieee float) reciprocal in about 6 cycles, and a 1/sqrt in 7 or so. Oh, and that's 4 results in that time. Averaging 1.5 cycles per FP divide, and slightly more for sqrt. times 7, times 3.2 billion per second, and the bandwidth to feed it.
That's several orders of magnitide faster that you could do with any x86 part out there.
This guy's ramblings on that web page are quite thoroughly disconnected from reality.
Let me give him a hint; wishing for something to be true does not make it so. The real world is quite unconcerned with your desires and keeps chugging along in its own way regardless of them.
I have a fair amount of experience throwing off my vestibular canals and ignoring them (pilot, flew aerobatics on the competition circuit for a few years). I tried to walk in a straight line while the device was trying to have me do otherwise. It was *extremely* difficult, but not impossible.
The feeling of lateral acelleration (where none was actually present) was very convincing.
I also thought this was one of the cooler things in the emerging technologies section at siggraph.
Not speaking for Alias, but it would be *very* hard to make Maya (or Studio Tools) use anything other than OpenGL. Both of these applications are high end apps that most of their users spend alot of time in. If they don't run well on a particular platform, but do run well on anotgher -- most users will just buy the good platform.
MS is cutting off their nose to spite their face on this one. It's a really stupid move on their part.
"The problem is not coming up with good ideas, but getting the political and financial resources to develop them."
Very true -- good ideas are a dime a dozen. Persuading a group of smart and skeptical people to fund 6 or 7 figures of development budget is harder. Harder still is seeing your idea through implementation to delivering a real revenue generating product. Always remember that the default outcome for software projects is failure. And you have to fight that tendency every day.
I agree with much of what you say about SGI -- they are a pale shadow of a once great company. And one that was really good to work for.
Alias is no longer part of SGI. Hasn't been for about a year now. Alias is now privately held by a venture capital firm Accel-KKR, and the Ontario Teachers Pension fund (a collegue of mine asked if this meant we were getting pesions now -- yes, Mike I'm talking about you:-)
Maya and Studio are targeted at very different markets. Studio is aimed at fairly high end industrial design and styling. It's a vertical market that is willing to pay well for tools that are really good at what they do. Studio does well, and, on average, doesn't make less money than Maya. But the two products have different customer sets, business models, and therefore different pricing structures.
Alias remains a very cool place with many innovative things happening. Yes, sometimes it's easy to think it was cooler when Rob Burgess was running it -- but the mind is always quick to paint nostalgic memories in the best light.
No, I don't always agree with the decisions that are made by those running the place. Not much different from any other company in that respect.
I was recently down in Doug's office (president) and told him I was conducting a palace coup. He told me the job was mine -- all I had to do was answer the next phone call:-)
And don't forget that when you figure energy density, you'll want to take into account the mass and volume that its container takes up too. This makes hydrogen and LNG suck even more as fuels for vehicles as their fuel tanks are much heavier than those you can use for gasoline or diesel.
I wasn't specifically talking about them -- but the problem I'm describing (imperfectly at that -- "would would"? I meant "world would") seems endemic to the graphics chip industry -- probably because their chips are so complex, and are never touched directly by most software anyways -- so it seems acceptable to band-aid over problems in the driver software.
But the problem is not confined to the gfx manufacturers -- I've heard horor stories about a certain maker of SCSI chips as well.
They don't release the specs because the hardware is generally full of bugs that they work around in their drivers. Different versions of the hardware have their own bugs and work arounds. Sometimes I think they're too embarassed to release the specs for fear the would would see what a dogs breakfast some of their work amounts to.
Further most of them have "specs" which barely qualify for the name -- often driver writers will go read the RTL, or talk directly to the hardware guys to figure out how something works -- often much of it is not written down, and usually not in one coherent document that could be called a spec.
I know this for a fact about several producers of hardware that serves various purposes -- who, for obvious reasons, I shall not name.
Others are anal about getting things right and writing great specs -- even if they never publish them -- But I think they're in the minority.
Not knowing what your situation is, I won't criticise you for putting food on the table. But I hope you're not involved in something as unethical as what you're describing -- That's outright vandalism -- IANAL, but I believe in Canada it would fall under the following section of the criminal code
Section 387(1): "Everyone commits mischief who wilfully (a) destroys or damages property, (b) renders property dangerous, useless, inoperative or ineffective, (c) obstructs, interrupts or interferes with the lawful use, enjoyment or operation of property, or (d) obstructs, interrupts or interferes with any person in the lawful use, enjoyment or operation of property."
Those who commit mischief in relation to private property may be found guilty of an indictable offence and are liable to imprisonment for five years.
I'd love to see whoever is responsible for that vandalous DRM scheme brought up on charges.
Here's an even scarier idea -- Macrovision has install shield secretly install a rootkit on your system which enables their evil interdiction and DRM schemes. (Virtuually every windows box has had install shield run on it several times.)
I better go file a patent double quick on this silly idea.
Hey Suresh! You reading this? If you steal my idea I'll hunt you down and sue your ass into next sunday! -- or at least invalidate any patent on it using this posting as evidence of prior art:-)
(I used to work with a guy named Suresh Balasubramanian -- cool guy back then -- a graphics pipeline microcode programmer at SGI - now director of product development at macrovision.)
I'm now expecting a call saying "Ian, you don't know the power of the dark side -- join us and together we will rule this planet!" complete with scuba gear breathing...
(I'm assuming it's the same Suresh at Macrovision)
Why then do you arguably get somewhere between 0 and 1 of those three when running Windows?
(I've switched from Windows to Mac as my primary development machine.)
Out with it soldier -- what is it you are trying to buy? I bet they'll ship it to me -- and then I can ship it to you. If not me, I'm sure there are dozens who would also like to help.
(But I wonder that it's a Canadian who is the first to offer to help here.)
I can be reached at ameline@a[ ]s.com ['lia' in gap]
Slashdot Mirror
These spores have clearly been designed by His Noodlyness. There is truly no limit to what He can do with merely the wave of His Noodly Appendage.
Ramen.
(Sorry for the off topic post -- I hope some of you will find it informative -- and some of those people might even have mod points to burn :-)
Clearcase sucked big time -- at least the last time I looked (admittedly years ago) Performance just went in the toilet as soon as you started throwing apps built from 10's of thousands of cpp files. Never mind having hundresds of developers manipulating 10s of versions of everything.
We use Perforce, and it rocks -- handles branching and mergeing great, scales to ridiculously large projects, is rock solid, and works on Irix, Solaris, HPUX, AIX, Linux, Windows and Mac OSX. (And yes, we need it to run across that many platforms.)
Hey, At least Maya runs on Linux, and Mac OSX. If you want to transfer your maya licence to linux or mac, contact Alias.
A careful study of the asm docs will lead you to the conclusion that it's aimed at four seperate target markets;
1: Gamers / graphics;
-- 4 deep vectors of single precision float -- various interesting but ulimately useless-for-games parts of the IEEE FP spec not implemented.
2: Video codecs. Parent post has the details
3: Crypto -- Support for arbitrary precision math. (details in parent post). popcount, count leading zeros. I expect various 3 letter agencies to buy boatloads of these things.
4: Supercomputing math geeks. 2 deep Double precision vectors with full IEEE support and compliance. (Currently not fully pipelined, but I expect that that situation will not last for long.)
What's missing for me;
- no saturated fixed point math.
- round, ceil, floor, trunc are all very slow (compared to everything else).
Those estimate instructions did not place useful information in the low order bits of the estimate. Those processors did not have an instruction whose only porpose was to process those special low order bits to produce more accurate results. This aspect of Cell is really new.
Not just buzzword compliant, but confusing as hell for those of us who have been in the know for a while.
To me, SPU always made me think "Scalar Processing unit", while PPE made me think "Parallel Processing Element".
Of course that's exactly backwards.
That, and I choke on words like "synergistic" because they peg my bullshiat-o-meter way off in the red.
In my opinion one of the coolest features of this architecture are the way the reciprocal estimate and reciprocal square root estimate instructions work.
In a single cycle you get 13 good bits of precision -- with the low order bits filled with information to be used by the floating point interpolate instruction.
You can get a full precision (32 bit ieee float) reciprocal in about 6 cycles, and a 1/sqrt in 7 or so. Oh, and that's 4 results in that time. Averaging 1.5 cycles per FP divide, and slightly more for sqrt. times 7, times 3.2 billion per second, and the bandwidth to feed it.
That's several orders of magnitide faster that you could do with any x86 part out there.
He's been smoking some seriously strong weed to come up with the crazy ass ideas in that article.
Of course I'm sure you don't find it at all ironic that you include spam in your very own signatture line, do you?
feh.
Yep --
This guy's ramblings on that web page are quite thoroughly disconnected from reality.
Let me give him a hint; wishing for something to be true does not make it so. The real world is quite unconcerned with your desires and keeps chugging along in its own way regardless of them.
20 stages? Where have you been -- it's over 30 these days.
I was there, and I tried it.
I have a fair amount of experience throwing off my vestibular canals and ignoring them (pilot, flew aerobatics on the competition circuit for a few years). I tried to walk in a straight line while the device was trying to have me do otherwise. It was *extremely* difficult, but not impossible.
The feeling of lateral acelleration (where none was actually present) was very convincing.
I also thought this was one of the cooler things in the emerging technologies section at siggraph.
Not speaking for Alias, but it would be *very* hard to make Maya (or Studio Tools) use anything other than OpenGL. Both of these applications are high end apps that most of their users spend alot of time in. If they don't run well on a particular platform, but do run well on anotgher -- most users will just buy the good platform.
MS is cutting off their nose to spite their face on this one. It's a really stupid move on their part.
Damn -- I have no mod points...
"The problem is not coming up with good ideas, but getting the political and financial resources to develop them."
Very true -- good ideas are a dime a dozen. Persuading a group of smart and skeptical people to fund 6 or 7 figures of development budget is harder. Harder still is seeing your idea through implementation to delivering a real revenue generating product. Always remember that the default outcome for software projects is failure. And you have to fight that tendency every day.
Silly question -- I believe that Itanium is getting all the respect it is due. It's just not due as much as you seem to think.
I agree with much of what you say about SGI -- they are a pale shadow of a once great company. And one that was really good to work for.
:-)
:-)
Alias is no longer part of SGI. Hasn't been for about a year now. Alias is now privately held by a venture capital firm Accel-KKR, and the Ontario Teachers Pension fund (a collegue of mine asked if this meant we were getting pesions now -- yes, Mike I'm talking about you
Maya and Studio are targeted at very different markets. Studio is aimed at fairly high end industrial design and styling. It's a vertical market that is willing to pay well for tools that are really good at what they do. Studio does well, and, on average, doesn't make less money than Maya. But the two products have different customer sets, business models, and therefore different pricing structures.
Alias remains a very cool place with many innovative things happening. Yes, sometimes it's easy to think it was cooler when Rob Burgess was running it -- but the mind is always quick to paint nostalgic memories in the best light.
No, I don't always agree with the decisions that are made by those running the place. Not much different from any other company in that respect.
I was recently down in Doug's office (president) and told him I was conducting a palace coup. He told me the job was mine -- all I had to do was answer the next phone call
Diesel has a higher energy density than gasoline.
And don't forget that when you figure energy density, you'll want to take into account the mass and volume that its container takes up too. This makes hydrogen and LNG suck even more as fuels for vehicles as their fuel tanks are much heavier than those you can use for gasoline or diesel.
I wasn't specifically talking about them -- but the problem I'm describing (imperfectly at that -- "would would"? I meant "world would") seems endemic to the graphics chip industry -- probably because their chips are so complex, and are never touched directly by most software anyways -- so it seems acceptable to band-aid over problems in the driver software. But the problem is not confined to the gfx manufacturers -- I've heard horor stories about a certain maker of SCSI chips as well.
They don't release the specs because the hardware is generally full of bugs that they work around in their drivers. Different versions of the hardware have their own bugs and work arounds. Sometimes I think they're too embarassed to release the specs for fear the would would see what a dogs breakfast some of their work amounts to.
Further most of them have "specs" which barely qualify for the name -- often driver writers will go read the RTL, or talk directly to the hardware guys to figure out how something works -- often much of it is not written down, and usually not in one coherent document that could be called a spec.
I know this for a fact about several producers of hardware that serves various purposes -- who, for obvious reasons, I shall not name.
Others are anal about getting things right and writing great specs -- even if they never publish them -- But I think they're in the minority.
Hey, anon DRM developer;
Not knowing what your situation is, I won't criticise you for putting food on the table. But I hope you're not involved in something as unethical as what you're describing -- That's outright vandalism -- IANAL, but I believe in Canada it would fall under the following section of the criminal code
Section 387(1): "Everyone commits mischief who wilfully (a) destroys or damages property, (b) renders property dangerous, useless, inoperative or ineffective, (c) obstructs, interrupts or interferes with the lawful use, enjoyment or operation of property, or (d) obstructs, interrupts or interferes with any person in the lawful use, enjoyment or operation of property."
Those who commit mischief in relation to private property may be found guilty of an indictable offence and are liable to imprisonment for five years.
I'd love to see whoever is responsible for that vandalous DRM scheme brought up on charges.
Here's an even scarier idea -- Macrovision has install shield secretly install a rootkit on your system which enables their evil interdiction and DRM schemes. (Virtuually every windows box has had install shield run on it several times.)
:-)
I better go file a patent double quick on this silly idea.
Hey Suresh! You reading this? If you steal my idea I'll hunt you down and sue your ass into next sunday! -- or at least invalidate any patent on it using this posting as evidence of prior art
(I used to work with a guy named Suresh Balasubramanian -- cool guy back then -- a graphics pipeline microcode programmer at SGI - now director of product development at macrovision.)
I'm now expecting a call saying "Ian, you don't know the power of the dark side -- join us and together we will rule this planet!" complete with scuba gear breathing...
(I'm assuming it's the same Suresh at Macrovision)
I know what would be good enough for the Slashdot crowd -- it's simple if you think for even a moment;
:-)
Full (and extended) frontal nudity for Natalie Portman.
That's all it would take.
(Ducking quickly
Why then do you arguably get somewhere between 0 and 1 of those three when running Windows? (I've switched from Windows to Mac as my primary development machine.)
I conceed that there may not be any I in "team", but remember, you can't spell "failure" without the letters U R A.
Out with it soldier -- what is it you are trying to buy? I bet they'll ship it to me -- and then I can ship it to you. If not me, I'm sure there are dozens who would also like to help. (But I wonder that it's a Canadian who is the first to offer to help here.) I can be reached at ameline@a[ ]s.com ['lia' in gap]
You watch -- it'll be used for porn and spam injected directly into your brain -- and I barely have room left for the voices.