First thing: I think the whole "USia" and "USians" is stupid and you were trolling by trying to piss of people from the U.S., known as Americans
I'd never thought about the term 'american' much until some Columbians pointed out to me at SigGraph once that they were Americans too (just like Germans are Europeans)and they resented the 'USians' (for lack of a better term) taking over their self identity.
After a while it dawned on me - we here in the US are the only people in the world who don't have name that uniquely represents ourselves.
Personally I knind of like the on-line equivalent 'merkins it only because of the double-entndre:-)
I find I do my most creative work in the times when my mind is not directly focused on the problem at hand - my best ideas (and the solutions to problems that have me blocked) seem to come when I'm in a quiet, somewhat disassociated, state - like taking a long hike in the hills, (or in town), or on the train, taking a long cross country drive, or one of the best - taking a long shower.
I suspect it has something to do with a combination of physical relaxation, quietness and lack of distractions (no radios/web/irc/etc)
Dave has also suffered the wrath of the Scientologists by exposing their space aliens and e-meter really a tricked out multimeter.
While doing so he's done it in the guise of academic scholarly study - pointing out why the 1st amendment is important and why 'fair use' is needed for a free and open society
AMD's doing a lot right, but there are a couple of additional things they need to do. One is to get the damn SMP chipsets out. They're not going to break into the lucrative server markets without a strong SMP offering.
I agree - I buy a lot of dual's, dozens at a time, and would much prefer to by AMD.... but they have to be as cheap or cheaper at the low end. I buy the cheapest cycles out there - I'm continually trying to guess the peak of the (daily) all up system price/performance curve.
The problem I see with future AMD SMP chipsets is that they are NOT going to be main-stream - the AMD FSB is point-to-point - that means the chipset needs about 100 odd pins per CPU - mainstream (1 cpu) chipsets aren't going to have those extra pins the SMP chipsets are going to be a low volume (and as a result probably more expensive) item. I worry that when they're finally here there will be no point in buying them because there will be no advantage
Reading the article it sounds like they want an interface that's still a lot more like a 'mac' interface than an old Unix one (ASR33->VT100) - mostly they seem to be quibbling about details rather than the whole - and hopefully with our current UI competition (KDE vs. Gnome) we're in a much position for real evolution than say the Mac or Windows
I disagree that the industry is near slowing down, and I'll tell you why: we are about to see a big jump in display qualities. HDTV's run at twice the resolution most of us are currently set to. 2048x1280 if I recall, which is quite a boost if you ask me.
I think the point I was making was that there's a big problem with consumer (ie something that costs a couple of hundred $$ and fits on a PCI card) 3D getting faster - it's memory bandwidth into off-the-shelf DRAMs - in fact HDTV makes it worse - 2048x1024x4bytes/pixelx85Hz is 700Mb/sec just to read the pixels and put them on the screen - peak is really closer to 900Mb/sec because there's no need to fetch pixels during blanking - remember a 200MHz DDR DRAM bank only produces 1.6GB/sec best case (realiisticly you might get 1.2-1.4 - this doesn't leave an awfull lot of bandwidth for rendering..... 3D on HDTV is going to be slow (in fact the bandwidth sucked up by display refresh is roughly proportional to the square of the screen resolution) - plus more pixels also means you need more bandwidth and GPU oomph to do a good job....
This latest crop of cards only barely support those resolutions, and certainly not at 85+fps with however many millions of triangles/sec. Lots of room for development there.
Right - but they're going to have to use a different technology that simple consumer DRAMS if they are going to keep producing order of magnitude performance increases every generation. However as I pointed out elsewhere - screens aren't going to get much bigger for a while so the memory bandwidth they suck wont keep increasing generation to generation - and I bet the 'sweet spot' the chip designers aim at wont be 3D on HDTV any time soon - in fact it's more likely to be something like 1kx512 scaled and filtered on-the-fly to HDTV resolutions which uses a lot less bandwidth (1/4)
Concerned that HDTV is vaporware?
Nope - as a logic designer I've been designing HDTV capable displays for almost a decade now - I still remember showing my first card driving what was at the time one of less than 10 HDTV monitors in the US at a show 7 years or so back - that thing weighed a ton and needed a forklift to get it into the booth.
As HDTV picks up some steam (lord knows it's been long enough) Those people who adopt will realize that their TV sets are much higher resolution than their monitors.
Yup - and in fact Radeon claims to support HDTV out YPbPr (I don't know if this is standard on the new cards or a future feature - I'd guess standard through a fudged VGA cable) - but then ATI makes set-top graphics chips too...
I'd agree that bandwidth constraints are important, but I don't think they're the end of the world. At the moment (on-card) bandwidth primarily limits framebuffer read/write and texture sampling;
Yes - and with more bits/pixel (now 32 is becoming common) and bigger displays you need a lot more bandwidth for just driving the screen (4x1280x1024x75 = ~400Mb/sec alone to get the data out of the memory - and you figure you need at least the same amount to get it if you're doing 1 frame/frame so close to 1Gb/sec just for the frame buffer. The good news here is that neither the number of pixels/screen or the refresh rate is going up at a horrble rate - in fact the pixel count and bits/pixel will probably plateau at HDTV sizes (modulo 4x or 16x for antialiasing) - which in a couple of generations of silicon will fit on-die (remember my comment was about the use of off-the-shell DRAMs - a technology that tends to drive density rather than speed) - once you move that (and the Z buffer) on-board you can have radically different memory architectures (ie really wide and/or radically banked for parallelism) that wil probably allow you another order of magnitude DRAM performance
(then again maybe we should bring back VRAM to double our bandwidth in the mean time - the problem being that now days there's 100s of 1000s of gates in the display backend past the DRAM outputs where in the days of VRAMs we used to put a simple mux and DAC)
the hardware focus now is shifting more towards single-pass multitexture (which can actually reduce bandwidth requirements in some cases), per-pixel shading (which is more computation than memory access) and on-card vertex manipulation (which is a really wide-open field, and is already causing historical API models to creak a bit at the seams).
Yup and all these require different programming - I wonder who's really driving the APIs - the chip architects or the game developers - both have pretty long development cycles - I bet it's pretty hit and miss...
The comparison with 2D graphics is possibly misleading - 2D hit a plateau because it became "fast enough" for just about anything. I mean, once you can comfortably redraw the entire screen every frame there's nowhere much to go
well to be fair the same could be said for 3D cards - benchmarking things at 100+ frames/sec is really quite meaningless if the screen is doing 75Hz... which is one of the reasons we're now seeing all this stuff that provide better image quality (for example Radeon has T&L to get poly counts up, bump mapping, skins etc etc)
There are ways around the bandwidth wall if you want to badly enough; move blending ops into RAM as some companies have done to avoid the read-modify-write round trip, or use a Talisman-style tiled architecture and parallel pipelines.
Yup Talisman sorts of stuff was what I was refering to when I was talking about 'radically different architectures' in my previous post - in general we're going to have to play with the memory hierarchy to get faster memory access - bringing some of in on-chip as in Talisman-like image-space stuff or even hardware ray tracing (the number of rays you have to trace is 1st order proportional to the number of screen pixels which as I pointed out above isn't going up particularly fast - there has to be a cross-over point where this becomes usefull)
I think it will slow down - but probably not for the reason you think - there's a memory bandwidth wall fast approaching in the consumer space - by which I mean that if you are using off-the-shelf DRAMs there's only so much bandwidth you can get out of them and they're already pushing them to the max - no more factor of 10 performance boosts in the pipes without gross rearchitecting of people's graphics pipes (and software/drivers/etc). We saw this same thing happen with 2D - for a long time companies were fighting to outdo each in 2D graphics performance - 64-bit cards, 128, 256 etc - but you can only push bytes into dram so fast and wider buses mean more pins but only factor of 2 performance increases - eventually there came a point where everyone's performance became pretty much the same and no-one really cares any more provided it's 'fast enough'.
Instead in the 3D space I think you'll see a broadening of the features available - and more optimization to avoid drawing pixels at all - Radeon seems to already have some of this with it's 'Hyper-Z' stuff
Remember when you had to be a non-professional to compete? Back when the amount of money floawing around the IOC was soooo much smaller - well I think they should go back to the old-style olympics as the Greeks envisioned them.... keep them in one place, don't move them around and make them non-commercial...
Oh, and did I mention?.... everyone will have to compete naked.... that whould keep the NBCs of the world away.....
Durons do have a set of pins to request the multiplier.
But, if they're external pins, this doesn't get around the 'fake chip' problem - better they are on-die in some form to make it hard for people to be selling overclocked chips.
All of this may be moot in the future - with multiplier-agile mobile parts appearing that can change their multipliers on the fly these pins are becoming simply a suggestion (what frequency you should come out of reset at - but this is probably going to be set low to save power at reset)
Reason Number 1. It would mean having different CPU dies for each chip they sell. They aren't going to do that. They prefer to simply build a batch of chips and depending on how clean they come out you put a label on to claim a specific clock speed. Yes. Specific clock speeds are determined after the fact before labeling is done, not before.
Not quite - if they are locking frequencies then what they're probably doing now is bonding out the frequency programming pads when the package the die - what this means is that the robot that solders the tiny wires between the chip carrier and the die wires up different wires depending on whether wafer sort decided that the dies were fast enough ie they decided how fast the die ought to be before they packaged it, not after - most chips get 2 sorts of tests - before and after in order to weed out bad die early - packaging is an expensive step - also all the die on a wafer tend to run at the same speed because they've all received the same processing - sometimes extra circuitry is added to a wafer to allow it to be easily characterised.
What they could do instead is to add an extra set of pins saying what speed they think the chip should run at and make those available in an internal register for the OS to print at boot time. This way you could have overclocking and a CPU that announced how fast its manufacturer thinks it should run.
that's OK you can still write your congresspeople.... and aactually having control of your own mind you can tell them what you want them to - rather than prrotting the words of Chairman Bill....
MacHack has gone a long way to promote open-source on the mac - for way longer than 'Open Source' was a popular refrain - namely the proceedings disks containing all the hacks - including most of the sources.....
sadly I haven't been able to go to 'The Hack' for some years but IMHO it still rates as one of the most fun conferences there is - the numbers are small, everyone stays in the same hotel - sessions start at lunch and run 'till midnight (except for the business track which occurs 9-12 while everyone's asleep) - there's rows of machines to play on - and a contest to produce the best 'hack' at the show - and completely worthless (and highly prized) prizes for the winners....
Sadly the Linux conferences seem much more like the big PC shows rather than the more intimate feel of the Hack - I think it's more like Usenix was back in the mid 80s
Well if I'm sitting in a commercial plane flying over his launch site when he launches I care - he has to get the FAA waivers and NOTAMs filed to keep air traffic away from where he's flying.
But in my posting I was refering to the next step which is where FAA jurisdiction cuts out (~100k ft I think) and the DoT OCS (I think that's what they're called) agency cuts in - their job is to make sure he doesn't drop his whopping big tank, or himself, on say Vancouver CA (or Seattle for that matter) - so when you're space shot goes awry and you're charged with 'criminal manslaughter' or whatever - we'll write you in jail..... in whatever country you end up dropping in on.... say Turkey?
yeah that's about it - he'll have to get more than the FAA permission he's asking for there's a portion of the DOT who's job it is to make sure that stuff doesn't fall on other countries and cause diplomatic incidents etc.... they have to get involved for flights over 100k ft (~20 miles)
I fly (not quite that big) rockets for fun out in the NV desert - friends of mine made an attempt at over 100kft a while back (officially a sub-orbital flight) - the paper work is amazing - you have to do a lot of faiilure analysis, all this population density downrange statistics etc etc finally resulting in a final number estimate of the fraction of a person you will kill (statistically) during your flight....
Personally I think he's crazy - I've seen a LOT of rockets go wrong (Murphy LOVES rockets they're his favorite thing:-) - I'd want at least a half dozen successfull unmanned flights under my belt brfore I strapped my skin into something like that
I live in Berkeley and every year it has a parade in which people poke fun at Berkeley's (and other people's) stereotypes of itself. My favorite float (and always a crowd pleaser) is the PETA one (People Eating Tasty Animals - I assume it's the same crowd). It's covered with people dressed in meat on top there's a BBQ or two throwing sausages to the crowd.
Of course it's always followed by the goose-stepping 'greenshirts' of the vegan-reich....
Slashdot Mirror
I'd never thought about the term 'american' much until some Columbians pointed out to me at SigGraph once that they were Americans too (just like Germans are Europeans)and they resented the 'USians' (for lack of a better term) taking over their self identity.
After a while it dawned on me - we here in the US are the only people in the world who don't have name that uniquely represents ourselves.
Personally I knind of like the on-line equivalent 'merkins it only because of the double-entndre :-)
this is of course not a new idea ... it's been done in the crypto world a few times already ....
hmmmm .... I wonder where I'll put it ..... is the judge a prude?
I suspect it has something to do with a combination of physical relaxation, quietness and lack of distractions (no radios/web/irc/etc)
1234-3456-5678-8888-1201@visa.gov ... that way the spammers can cut out the middle man (you)
maybe it wil be more like:
householder@303.north-seminary-avenue.park-ridge.i l-60068-3048.usps.gov
with 'householder' replaced with your name for personal stuff ....
While doing so he's done it in the guise of academic scholarly study - pointing out why the 1st amendment is important and why 'fair use' is needed for a free and open society
I agree - I buy a lot of dual's, dozens at a time, and would much prefer to by AMD .... but they have to be as cheap or cheaper at the low end. I buy the cheapest cycles out there - I'm continually trying to guess the peak of the (daily) all up system price/performance curve.
The problem I see with future AMD SMP chipsets is that they are NOT going to be main-stream - the AMD FSB is point-to-point - that means the chipset needs about 100 odd pins per CPU - mainstream (1 cpu) chipsets aren't going to have those extra pins the SMP chipsets are going to be a low volume (and as a result probably more expensive) item. I worry that when they're finally here there will be no point in buying them because there will be no advantage
Reading the article it sounds like they want an interface that's still a lot more like a 'mac' interface than an old Unix one (ASR33->VT100) - mostly they seem to be quibbling about details rather than the whole - and hopefully with our current UI competition (KDE vs. Gnome) we're in a much position for real evolution than say the Mac or Windows
I think the point I was making was that there's a big problem with consumer (ie something that costs a couple of hundred $$ and fits on a PCI card) 3D getting faster - it's memory bandwidth into off-the-shelf DRAMs - in fact HDTV makes it worse - 2048x1024x4bytes/pixelx85Hz is 700Mb/sec just to read the pixels and put them on the screen - peak is really closer to 900Mb/sec because there's no need to fetch pixels during blanking - remember a 200MHz DDR DRAM bank only produces 1.6GB/sec best case (realiisticly you might get 1.2-1.4 - this doesn't leave an awfull lot of bandwidth for rendering ..... 3D on HDTV is going to be slow (in fact the bandwidth sucked up by display refresh is roughly proportional to the square of the screen resolution) - plus more pixels also means you need more bandwidth and GPU oomph to do a good job ....
This latest crop of cards only barely support those resolutions, and certainly not at 85+fps with however many millions of triangles/sec. Lots of room for development there.
Right - but they're going to have to use a different technology that simple consumer DRAMS if they are going to keep producing order of magnitude performance increases every generation. However as I pointed out elsewhere - screens aren't going to get much bigger for a while so the memory bandwidth they suck wont keep increasing generation to generation - and I bet the 'sweet spot' the chip designers aim at wont be 3D on HDTV any time soon - in fact it's more likely to be something like 1kx512 scaled and filtered on-the-fly to HDTV resolutions which uses a lot less bandwidth (1/4)
Concerned that HDTV is vaporware?
Nope - as a logic designer I've been designing HDTV capable displays for almost a decade now - I still remember showing my first card driving what was at the time one of less than 10 HDTV monitors in the US at a show 7 years or so back - that thing weighed a ton and needed a forklift to get it into the booth.
As HDTV picks up some steam (lord knows it's been long enough) Those people who adopt will realize that their TV sets are much higher resolution than their monitors.
Yup - and in fact Radeon claims to support HDTV out YPbPr (I don't know if this is standard on the new cards or a future feature - I'd guess standard through a fudged VGA cable) - but then ATI makes set-top graphics chips too ...
Yes - and with more bits/pixel (now 32 is becoming common) and bigger displays you need a lot more bandwidth for just driving the screen (4x1280x1024x75 = ~400Mb/sec alone to get the data out of the memory - and you figure you need at least the same amount to get it if you're doing 1 frame/frame so close to 1Gb/sec just for the frame buffer. The good news here is that neither the number of pixels/screen or the refresh rate is going up at a horrble rate - in fact the pixel count and bits/pixel will probably plateau at HDTV sizes (modulo 4x or 16x for antialiasing) - which in a couple of generations of silicon will fit on-die (remember my comment was about the use of off-the-shell DRAMs - a technology that tends to drive density rather than speed) - once you move that (and the Z buffer) on-board you can have radically different memory architectures (ie really wide and/or radically banked for parallelism) that wil probably allow you another order of magnitude DRAM performance
(then again maybe we should bring back VRAM to double our bandwidth in the mean time - the problem being that now days there's 100s of 1000s of gates in the display backend past the DRAM outputs where in the days of VRAMs we used to put a simple mux and DAC)
the hardware focus now is shifting more towards single-pass multitexture (which can actually reduce bandwidth requirements in some cases), per-pixel shading (which is more computation than memory access) and on-card vertex manipulation (which is a really wide-open field, and is already causing historical API models to creak a bit at the seams).
Yup and all these require different programming - I wonder who's really driving the APIs - the chip architects or the game developers - both have pretty long development cycles - I bet it's pretty hit and miss ...
The comparison with 2D graphics is possibly misleading - 2D hit a plateau because it became "fast enough" for just about anything. I mean, once you can comfortably redraw the entire screen every frame there's nowhere much to go
well to be fair the same could be said for 3D cards - benchmarking things at 100+ frames/sec is really quite meaningless if the screen is doing 75Hz ... which is one of the reasons we're now seeing all this stuff that provide better image quality (for example Radeon has T&L to get poly counts up, bump mapping, skins etc etc)
There are ways around the bandwidth wall if you want to badly enough; move blending ops into RAM as some companies have done to avoid the read-modify-write round trip, or use a Talisman-style tiled architecture and parallel pipelines.
Yup Talisman sorts of stuff was what I was refering to when I was talking about 'radically different architectures' in my previous post - in general we're going to have to play with the memory hierarchy to get faster memory access - bringing some of in on-chip as in Talisman-like image-space stuff or even hardware ray tracing (the number of rays you have to trace is 1st order proportional to the number of screen pixels which as I pointed out above isn't going up particularly fast - there has to be a cross-over point where this becomes usefull)
Instead in the 3D space I think you'll see a broadening of the features available - and more optimization to avoid drawing pixels at all - Radeon seems to already have some of this with it's 'Hyper-Z' stuff
Ummmm - not quite - they are shipping today - that's what they're announcing - many online sellers have had them in their catalogs since last week ....
Oh, and did I mention? .... everyone will have to compete naked .... that whould keep the NBCs of the world away .....
But, if they're external pins, this doesn't get around the 'fake chip' problem - better they are on-die in some form to make it hard for people to be selling overclocked chips.
All of this may be moot in the future - with multiplier-agile mobile parts appearing that can change their multipliers on the fly these pins are becoming simply a suggestion (what frequency you should come out of reset at - but this is probably going to be set low to save power at reset)
Not quite - if they are locking frequencies then what they're probably doing now is bonding out the frequency programming pads when the package the die - what this means is that the robot that solders the tiny wires between the chip carrier and the die wires up different wires depending on whether wafer sort decided that the dies were fast enough ie they decided how fast the die ought to be before they packaged it, not after - most chips get 2 sorts of tests - before and after in order to weed out bad die early - packaging is an expensive step - also all the die on a wafer tend to run at the same speed because they've all received the same processing - sometimes extra circuitry is added to a wafer to allow it to be easily characterised.
What they could do instead is to add an extra set of pins saying what speed they think the chip should run at and make those available in an internal register for the OS to print at boot time. This way you could have overclocking and a CPU that announced how fast its manufacturer thinks it should run.
why after I turned my monitor upside down the web page look just like any other Linux web page ....
that's OK you can still write your congresspeople .... and aactually having control of your own mind you can tell them what you want them to - rather than prrotting the words of Chairman Bill ....
until they attack Tokyo, knock down a few buildings and have a face off with Godzilla ...
MacHack has gone a long way to promote open-source on the mac - for way longer than 'Open Source' was a popular refrain - namely the proceedings disks containing all the hacks - including most of the sources .....
there have also been a couple of ports of GCC to run under MPW
Sadly the Linux conferences seem much more like the big PC shows rather than the more intimate feel of the Hack - I think it's more like Usenix was back in the mid 80s
But in my posting I was refering to the next step which is where FAA jurisdiction cuts out (~100k ft I think) and the DoT OCS (I think that's what they're called) agency cuts in - their job is to make sure he doesn't drop his whopping big tank, or himself, on say Vancouver CA (or Seattle for that matter) - so when you're space shot goes awry and you're charged with 'criminal manslaughter' or whatever - we'll write you in jail ..... in whatever country you end up dropping in on .... say Turkey?
I fly (not quite that big) rockets for fun out in the NV desert - friends of mine made an attempt at over 100kft a while back (officially a sub-orbital flight) - the paper work is amazing - you have to do a lot of faiilure analysis, all this population density downrange statistics etc etc finally resulting in a final number estimate of the fraction of a person you will kill (statistically) during your flight ....
Personally I think he's crazy - I've seen a LOT of rockets go wrong (Murphy LOVES rockets they're his favorite thing :-) - I'd want at least a half dozen successfull unmanned flights under my belt brfore I strapped my skin into something like that
Of course it's always followed by the goose-stepping 'greenshirts' of the vegan-reich ....