Um, no, it's not an idiom; "could care less" is just a typo. "Couldn't care less" is the idiomatic phrase, a poetic rendering of "don't care". Actually, show me an idiom that is the simple reverse of its literal counterpart... they tend to be more subtle, complex, or archaic than that.
(Besides, one could argue that there is no "fixed" system of meaning; it's all more or less reliant on accepted cultural understanding. But let's digress.)
Yes, the G400 was really fast as such (in 32bit especially), but it was plaqued by driver problems, particularly in OpenGL. The one thing where it was the absolute king was image quality -- the RAMDACs and the analog outputs were just lightyears better than what others had on their cards. That's why G400 continued to be a favourite video card for business workstations, until LCDs finally took over the whole scene.
Hey, I didn't mean to badmouth the TNT -- it was a solid, reliable performer. And considering it was an elegant single-chip solution, while a Voodoo2 needed three chips but *still* didn't have 2D VGA or 32bit colour... 3dfx was already beginning to lag behind. (And shot themselves in the head by grabbing the card making to themselves -- all their former vendors had to buy Nvidia. Ouch.)
Completely agree about ATI's drivers. They have a long tradition of being a nightmare to the user. Nowadays thye are in pretty good shape.
On an Aeroflot flight to Moscow, a passenger suddenly pulls an automatic rifle, storms into the cockpit, and demands: "I'm hijacking this plane to Moscow! Take us to Moscow immediately or everybody dies!" The captain attempts to explain: "Calm down, this *is* the flight to Moscow." The man screams: "Yes I know, but it's been hijacked every time I've tried to get there!"
No, they have one core with 320 scalar ALUs (ATI Radeon HD 2900) and very good thread management (also capable of bundling up 4 ALUs for a vector operation). Maybe you were thinking about the number of pixel pipelines, the high end parts have currently 24 of them.
I've corrected you on the "NV30 made by 3dfx engineers" error in an earlier post. Here I'll just point out that TNT wasn't the king of the hill -- Voodoo 2 SLI was -- nor was TNT2 because Matrox's last solid card G400 beat it (and other cards) in 32-bit colour. But I also agree, the rest since Geforce DDR have been the top dogs, except the FX5800 fiasco of course. Also agree about the drivers, Nvidia understood early on how crucially important they are to preformance and reliability; although I seem to remember Matrox was the first to offer open-sourced Linux drivers.
No, it was well reported back then that the 3dfx guys were just integrated into the existing teams. NV30 was a logical continuation to the 4 Ti and kept big chunks of it almost intact. Where Nvidia went wrong was they overestimated Z-testing (the "super pipelines" brouhaha), and critically, they went all-out for FP32 while ATI got FP24 accepted for the Full Precision of DX9; And completely unrelated to design prowess, there were TSMC's surprise problems with the 130 nm fabbing node, hence the ghastly dustbuster cooling.
Your info that NV30 was a 3dfx design is simply false, and probably has been propagated by diehard 3dfx fans. What 3dfx was working on -- SAGE geometry chip and Rampage rasterizer -- was quite different, although not very much is known about it, and their planned FEAR/Fusion follow-up based on Gigapixel 's tiler tech was still more different.
By the way, where have you actually seen the phrase "Do No Evil"?
From www.google.com/corporate/tenthings.html:
6. You can make money without doing evil.
The chapter isn't about grand ideals for a corporation, it's about better advertising practices.
"Do No Evil" has never been Google's corporate motto. Nor have they ever claimed anything like that. (Sorry if I burst somebody's bubble, "Do No Evil" has been a popular Slashdot meme for quite a while...)
You can also add advertising that is carefully selected to insult your intelligence and make you wish nobody passing by saw your screen. Then you can add a clutter of configuration and settings tool buttons with only cryptic images as clues; make sure to scatter these randomly. Also, make the app always run at startup, and bury the option to disable this as deep as possible; make the app also pop up another window for completely irrelevant news and an unwanted mail service. As a final touch, call the chat program a "messenger". Name it all, oh I don't know, Windows Live or something.
Intel became a vocal proponent of raytracing virtually overnight when they saw it as a way to create buzz for Larrabee... but most recent papers on the future of realtime rendering don't quite see raytracing as the holy grail walking all over rasterizing. Both have their strengths and weaknesses. (Yes, raytracing too.) The currently dominant method of OpenGL style textured polygons + Renderman style shaders has lots of potential in how it can develop further. And some interesting hybrids have been proposed, combining the excellent global lighting of raytracing to the inherent simplicity/speed of rasterization. The Carmack has some funky new ideas on voxels as base geometry primitives, I'm not even sure where that falls regarding the rasterization/raytracing divide... But anyway, look beyond Intel for the full picture of the future of PC gaming:-)
"So if your definition of Linux-like uses Linux as a kernel, then yes, OSX is Linux-like because it is basically BSD, which most people admit is Linux-like, as they are both Open Source kernels and Unix-like."
OS X does not use a BSD derivative kernel though, it uses a proprietary (and closed source) Mach derivative kernel. Sure, OS X has lots of BSD in the system services on top of its kernel. But if you base your likeness definitions on the kernel, then... but I kinda keep losing myself in your definition above:-P
(Privately I think Apple should ditch their proprietary kernel as an unnecessary effort, adopt the FreeBSD kernel, and keep producing an Apple Unix with all the API and GUI and app and maintenance tool goodness they have in OS X.)
(Aside: I calculated 12,000 times in a very simple matter. 3000 megahertz times dual-edge clocking times dual processor == 3000 x2 x2 == approximately 12,000 times. Is there a better way to measure how much faster a Core Duo 3000 is compared to the venerable 1 megahertz C=64 CPU?)
No "dual edge" though, that's for DDR, how memory chips get their clockspeed timing; nothing to do with CPUs. And dual core only doubles the horsepower on occasions when you really have work to keep both cores pegged at 100%. What is more important is the core's architecture, and obviously a Core 2 core has way, way more parallel execution resources (integer, FP64, SIMD) than a 6510 core, and a way better memory and cache subsystem to keep those execution resources maximally busy. (Your 12000 times faster might well be a pretty good guess in many processing situations, but exact figure will vary depending on workload du jour. Raw execution resources per each core could be summed up and compared, though, but I'm too lazy for that...) Hope this clarified!
In case it is still discussed somewhere... aside from the overall joke theme, it has some glaring factual errors about rendering methods. For example, the part on raytracing claims that raytracing is scalable while rasterization is not. This is just silly. It is simple and easy to parallelize rasterization (like how Nvidia's G80 introduced 128 shader processors and 24 pixel pipelines, and now they are talking quad-card SLI which seems to be efficiency limited only by interconnect bandwidth) while it actually takes a lot of work to write an efficiently scaling raytracer -- there are lots of gotchas about how to bundle up rays with regard to data dependencies, how to handle internode communication. I'm no expert myself but most papers on scalable raytracers seem to discuss these challenges a lot; you don't just fire off a a thousandfold amount of rays and get a thousandfold framerate increase, you've got to design it very thoughtfully.
I know there's this "30 fps is enuf, that's all your eye can see the differences" meme floating around, but that's not counting the flickers at the edge of awareness, which, for me anyway, doesn't disappear until 60-70 fps.
There is a simpler explanation. A film camera has a shutter time: every frame it captures is slightly blended over time. If you pause a video, the still image you see is often surprisingly blurry. Whereas computer graphics images are always prefectly sharp, like zero-time shutter shots. (Unless you use temporal anti-aliasing -- blending successive frames together before display -- but current video cards don't offer this; Silicon Graphics had a model with a hardware accumulation buffer for TAA, and Voodoo 5's "T-buffer" could be used for it too; by the way, it might have been good when flat panels were still severely framerate limited.) This sharpness is why you need about 60 fps for the illusion of smooth continuous movement.
I posted about this above but it bears repeating... the "do no evil" veneer is painted on by slashdotters, not by Google Inc. Their corporate website doesn't even have that phrase anywhere. What they do have is a section labeled "6. You can make money without doing evil" -- which is about better designed advertising.
The "do no evil" meme has been something of a/. invention all along. Google is just a company making a buck, they haven't claimed to be anything beyond that. Let's not put burdens on them *they* didn't want in the first place!
but rather that the industry won't abandon rasterization because each has strengths and weaknesses.
Aside from intrinsic strengths and weaknesses, he also observes that the industry has reason to stick to rasterization simply because very powerful rasterization hardware is ubiquitous, and that's hard to argue with. Nvidia and ATI aren't going to make it a walk in the park for Larrabee when it appears.
On hybrid approaches, it's interesting that Intel now seems to be talking about serious texture sampling capabilities for the initial 16 to 24 core Larrabee, as if backing away from the "pure ray-tracer" idea.
As to large data structures, the "Megatexturing" of id Tech 5 is apparently built closely around a streaming subsystem (reminding one how he talked about "virtual texturing" years ago in a.plan post), so I'd expect a continuation of that theme here too. In a comment some way above he indeed mentioned streaming to the extent of realtime online distribution becoming common in id Tech 6's timeframe. I'll expect some serious emphasis on smooth data traffic.
Nvidia didn't start with a polygons & textures renderer, though. NV1 (and NV2) was based on quadratic patches, thus natively capable of perfect curves.
It tanked because it proved to be a PITA to build a working game engine (especially collision detection and viewport clipping) around that principle, and Microsoft chose the OpenGL model instead for Direct3D. Nvidia nearly went under because of that fiasco. (But of course the elegant if not highest-performance NV3/Riva ZX single-chip graphics solution returned them in the good books of OEMs; and NV4/Riva TNT and NV5/Riva TNT2 were already worthy contenders to 3Dfx, Matrox, S3, and ATI offerings for the performance crown; and the rest is familiar history.)
In other words, it's not only about the pretty rendering results, it's also about how easy it is to build a complete game engine around the rendering method. The polygon meshes of a game's world model are also used by the physics and AI subsystems, level editors, and so on.
Which configuration is faster depends on the properties of the total workload under execution at any given time. But with the majority of apps out there still unoptimized for more than one core, the triple 2.6 GHz is likely to beat the quad 2.0 GHz in quite many instances (of varied "home computer" use) simply because it completes a single thread in less wallclock time.
All in all you just can't (sanely) boil it down to a simple case of "add up all the GHz", unless you are processing nothing but NOPs. In any real use there is a myriad of factors at play; so you have to look at it case by case and look at the details and characteristics of your combined worload.
(Caches and core interconnects and other relevant stuff omitted from discussion for simplicity.)
You made an excellent point and I'm glad to see your post up there at 5 where is belongs. However, this begs a comment:
Microsoft will actually be doing a disservice to teens if they don't explain the hard realities of copyright law.
I'm uncomfortable with the idea that a megacorporation takes upon itself -- or that we delegate on one -- the duty of educating teens (regardless of the topic). They are pervasive enough in our societies as it is, so many important areas of people's lives (health care, social activities, finances) revolving around and arranged by their employer corporation rather than something with a more benevolent or "neutral" basis. Sure an individual always has a choice there -- or does he or she in everyday reality?
But I'm not sure this is what you meant at all. And maybe I'm just being an alarmist here. Just felt like saying it out loud. Please comment, folks, would help me clarify my somewhat muddy thoughts here...
15% of the google codebase consitutes a "small user base"?
Why, yes. Google is a big user but still just one user. So if 15% of Google uses it, it's only 0.15 users. That's a very small user base. Even Mono has a larger user base (Miguel).
This doesn't affect W's rep in the academic world because it has none and should have none. It's helluva useful (for just about everybody there) but it cannot be depended on like a work with established procedures and responsible editors -- it cannot have and need not have that sort of authority (to be a small miracle of a tool). Who cares about a co-founder, it could be Dubya and still not affect the almighty community process one bit.
Slashdot Mirror
Um, no, it's not an idiom; "could care less" is just a typo. "Couldn't care less" is the idiomatic phrase, a poetic rendering of "don't care". Actually, show me an idiom that is the simple reverse of its literal counterpart... they tend to be more subtle, complex, or archaic than that.
(Besides, one could argue that there is no "fixed" system of meaning; it's all more or less reliant on accepted cultural understanding. But let's digress.)
Yes, the G400 was really fast as such (in 32bit especially), but it was plaqued by driver problems, particularly in OpenGL. The one thing where it was the absolute king was image quality -- the RAMDACs and the analog outputs were just lightyears better than what others had on their cards. That's why G400 continued to be a favourite video card for business workstations, until LCDs finally took over the whole scene.
Hey, I didn't mean to badmouth the TNT -- it was a solid, reliable performer. And considering it was an elegant single-chip solution, while a Voodoo2 needed three chips but *still* didn't have 2D VGA or 32bit colour... 3dfx was already beginning to lag behind. (And shot themselves in the head by grabbing the card making to themselves -- all their former vendors had to buy Nvidia. Ouch.)
Completely agree about ATI's drivers. They have a long tradition of being a nightmare to the user. Nowadays thye are in pretty good shape.
On an Aeroflot flight to Moscow, a passenger suddenly pulls an automatic rifle, storms into the cockpit, and demands: "I'm hijacking this plane to Moscow! Take us to Moscow immediately or everybody dies!" The captain attempts to explain: "Calm down, this *is* the flight to Moscow." The man screams: "Yes I know, but it's been hijacked every time I've tried to get there!"
No, they have one core with 320 scalar ALUs (ATI Radeon HD 2900) and very good thread management (also capable of bundling up 4 ALUs for a vector operation). Maybe you were thinking about the number of pixel pipelines, the high end parts have currently 24 of them.
I've corrected you on the "NV30 made by 3dfx engineers" error in an earlier post. Here I'll just point out that TNT wasn't the king of the hill -- Voodoo 2 SLI was -- nor was TNT2 because Matrox's last solid card G400 beat it (and other cards) in 32-bit colour. But I also agree, the rest since Geforce DDR have been the top dogs, except the FX5800 fiasco of course. Also agree about the drivers, Nvidia understood early on how crucially important they are to preformance and reliability; although I seem to remember Matrox was the first to offer open-sourced Linux drivers.
No, it was well reported back then that the 3dfx guys were just integrated into the existing teams. NV30 was a logical continuation to the 4 Ti and kept big chunks of it almost intact. Where Nvidia went wrong was they overestimated Z-testing (the "super pipelines" brouhaha), and critically, they went all-out for FP32 while ATI got FP24 accepted for the Full Precision of DX9; And completely unrelated to design prowess, there were TSMC's surprise problems with the 130 nm fabbing node, hence the ghastly dustbuster cooling.
Your info that NV30 was a 3dfx design is simply false, and probably has been propagated by diehard 3dfx fans. What 3dfx was working on -- SAGE geometry chip and Rampage rasterizer -- was quite different, although not very much is known about it, and their planned FEAR/Fusion follow-up based on Gigapixel 's tiler tech was still more different.
By the way, where have you actually seen the phrase "Do No Evil"?
From www.google.com/corporate/tenthings.html:
6. You can make money without doing evil.
The chapter isn't about grand ideals for a corporation, it's about better advertising practices.
"Do No Evil" has never been Google's corporate motto. Nor have they ever claimed anything like that. (Sorry if I burst somebody's bubble, "Do No Evil" has been a popular Slashdot meme for quite a while...)
You can also add advertising that is carefully selected to insult your intelligence and make you wish nobody passing by saw your screen. Then you can add a clutter of configuration and settings tool buttons with only cryptic images as clues; make sure to scatter these randomly. Also, make the app always run at startup, and bury the option to disable this as deep as possible; make the app also pop up another window for completely irrelevant news and an unwanted mail service. As a final touch, call the chat program a "messenger". Name it all, oh I don't know, Windows Live or something.
YPL also requires the preservation of all copyright and attribution notices within modified versions of the ZCS Open Source Edition.
You should have bolded this part. I skimmed over it twice before it hit me how much it means.
Intel became a vocal proponent of raytracing virtually overnight when they saw it as a way to create buzz for Larrabee... but most recent papers on the future of realtime rendering don't quite see raytracing as the holy grail walking all over rasterizing. Both have their strengths and weaknesses. (Yes, raytracing too.) The currently dominant method of OpenGL style textured polygons + Renderman style shaders has lots of potential in how it can develop further. And some interesting hybrids have been proposed, combining the excellent global lighting of raytracing to the inherent simplicity/speed of rasterization. The Carmack has some funky new ideas on voxels as base geometry primitives, I'm not even sure where that falls regarding the rasterization/raytracing divide... But anyway, look beyond Intel for the full picture of the future of PC gaming :-)
"So if your definition of Linux-like uses Linux as a kernel, then yes, OSX is Linux-like because it is basically BSD, which most people admit is Linux-like, as they are both Open Source kernels and Unix-like."
:-P
OS X does not use a BSD derivative kernel though, it uses a proprietary (and closed source) Mach derivative kernel. Sure, OS X has lots of BSD in the system services on top of its kernel. But if you base your likeness definitions on the kernel, then... but I kinda keep losing myself in your definition above
(Privately I think Apple should ditch their proprietary kernel as an unnecessary effort, adopt the FreeBSD kernel, and keep producing an Apple Unix with all the API and GUI and app and maintenance tool goodness they have in OS X.)
(Aside: I calculated 12,000 times in a very simple matter. 3000 megahertz times dual-edge clocking times dual processor == 3000 x2 x2 == approximately 12,000 times. Is there a better way to measure how much faster a Core Duo 3000 is compared to the venerable 1 megahertz C=64 CPU?)
No "dual edge" though, that's for DDR, how memory chips get their clockspeed timing; nothing to do with CPUs. And dual core only doubles the horsepower on occasions when you really have work to keep both cores pegged at 100%. What is more important is the core's architecture, and obviously a Core 2 core has way, way more parallel execution resources (integer, FP64, SIMD) than a 6510 core, and a way better memory and cache subsystem to keep those execution resources maximally busy. (Your 12000 times faster might well be a pretty good guess in many processing situations, but exact figure will vary depending on workload du jour. Raw execution resources per each core could be summed up and compared, though, but I'm too lazy for that...) Hope this clarified!
Didn't William Gibson actually introduce the term "Matrix" as well? (Or was it Bruce Sterling?)
In case it is still discussed somewhere... aside from the overall joke theme, it has some glaring factual errors about rendering methods. For example, the part on raytracing claims that raytracing is scalable while rasterization is not. This is just silly. It is simple and easy to parallelize rasterization (like how Nvidia's G80 introduced 128 shader processors and 24 pixel pipelines, and now they are talking quad-card SLI which seems to be efficiency limited only by interconnect bandwidth) while it actually takes a lot of work to write an efficiently scaling raytracer -- there are lots of gotchas about how to bundle up rays with regard to data dependencies, how to handle internode communication. I'm no expert myself but most papers on scalable raytracers seem to discuss these challenges a lot; you don't just fire off a a thousandfold amount of rays and get a thousandfold framerate increase, you've got to design it very thoughtfully.
I know there's this "30 fps is enuf, that's all your eye can see the differences" meme floating around, but that's not counting the flickers at the edge of awareness, which, for me anyway, doesn't disappear until 60-70 fps.
There is a simpler explanation. A film camera has a shutter time: every frame it captures is slightly blended over time. If you pause a video, the still image you see is often surprisingly blurry. Whereas computer graphics images are always prefectly sharp, like zero-time shutter shots. (Unless you use temporal anti-aliasing -- blending successive frames together before display -- but current video cards don't offer this; Silicon Graphics had a model with a hardware accumulation buffer for TAA, and Voodoo 5's "T-buffer" could be used for it too; by the way, it might have been good when flat panels were still severely framerate limited.) This sharpness is why you need about 60 fps for the illusion of smooth continuous movement.
I posted about this above but it bears repeating... the "do no evil" veneer is painted on by slashdotters, not by Google Inc. Their corporate website doesn't even have that phrase anywhere. What they do have is a section labeled "6. You can make money without doing evil" -- which is about better designed advertising.
/. invention all along. Google is just a company making a buck, they haven't claimed to be anything beyond that. Let's not put burdens on them *they* didn't want in the first place!
The "do no evil" meme has been something of a
Google's corporate website doesn't seem to have this famous phrase verbatim anywhere. There is this one item in their corporate philosophy section:
6. You can make money without doing evil.
It's mostly about less intrusive advertising.
I think Google's "no-evilness" is blown out of proportion (here and elsewhere). They've never claimed to be on some crusade for a better world.
but rather that the industry won't abandon rasterization because each has strengths and weaknesses.
.plan post), so I'd expect a continuation of that theme here too. In a comment some way above he indeed mentioned streaming to the extent of realtime online distribution becoming common in id Tech 6's timeframe. I'll expect some serious emphasis on smooth data traffic.
Aside from intrinsic strengths and weaknesses, he also observes that the industry has reason to stick to rasterization simply because very powerful rasterization hardware is ubiquitous, and that's hard to argue with. Nvidia and ATI aren't going to make it a walk in the park for Larrabee when it appears.
On hybrid approaches, it's interesting that Intel now seems to be talking about serious texture sampling capabilities for the initial 16 to 24 core Larrabee, as if backing away from the "pure ray-tracer" idea.
As to large data structures, the "Megatexturing" of id Tech 5 is apparently built closely around a streaming subsystem (reminding one how he talked about "virtual texturing" years ago in a
Nvidia didn't start with a polygons & textures renderer, though. NV1 (and NV2) was based on quadratic patches, thus natively capable of perfect curves. It tanked because it proved to be a PITA to build a working game engine (especially collision detection and viewport clipping) around that principle, and Microsoft chose the OpenGL model instead for Direct3D. Nvidia nearly went under because of that fiasco. (But of course the elegant if not highest-performance NV3/Riva ZX single-chip graphics solution returned them in the good books of OEMs; and NV4/Riva TNT and NV5/Riva TNT2 were already worthy contenders to 3Dfx, Matrox, S3, and ATI offerings for the performance crown; and the rest is familiar history.) In other words, it's not only about the pretty rendering results, it's also about how easy it is to build a complete game engine around the rendering method. The polygon meshes of a game's world model are also used by the physics and AI subsystems, level editors, and so on.
Forgive me, I am little confused here.
Yup.
Is Quad 2.0GHZ a total of 8GHZ?
No, it's quad 2.0 GHz.
Is Triple 2.6 GHZ a total of 7.8GHZ?
No, it's triple 2.6 GHz.
Which configuration is faster depends on the properties of the total workload under execution at any given time. But with the majority of apps out there still unoptimized for more than one core, the triple 2.6 GHz is likely to beat the quad 2.0 GHz in quite many instances (of varied "home computer" use) simply because it completes a single thread in less wallclock time.
All in all you just can't (sanely) boil it down to a simple case of "add up all the GHz", unless you are processing nothing but NOPs. In any real use there is a myriad of factors at play; so you have to look at it case by case and look at the details and characteristics of your combined worload.
(Caches and core interconnects and other relevant stuff omitted from discussion for simplicity.)
And like I said, nobody programmed their programs to split the processing into three parts, only 1, 2, or 4.
Yes, this is why the triple-core Xbox 360 has practically no games for it.
Oh wait.
You made an excellent point and I'm glad to see your post up there at 5 where is belongs. However, this begs a comment: Microsoft will actually be doing a disservice to teens if they don't explain the hard realities of copyright law. I'm uncomfortable with the idea that a megacorporation takes upon itself -- or that we delegate on one -- the duty of educating teens (regardless of the topic). They are pervasive enough in our societies as it is, so many important areas of people's lives (health care, social activities, finances) revolving around and arranged by their employer corporation rather than something with a more benevolent or "neutral" basis. Sure an individual always has a choice there -- or does he or she in everyday reality? But I'm not sure this is what you meant at all. And maybe I'm just being an alarmist here. Just felt like saying it out loud. Please comment, folks, would help me clarify my somewhat muddy thoughts here...
15% of the google codebase consitutes a "small user base"? Why, yes. Google is a big user but still just one user. So if 15% of Google uses it, it's only 0.15 users. That's a very small user base. Even Mono has a larger user base (Miguel).
This doesn't affect W's rep in the academic world because it has none and should have none. It's helluva useful (for just about everybody there) but it cannot be depended on like a work with established procedures and responsible editors -- it cannot have and need not have that sort of authority (to be a small miracle of a tool). Who cares about a co-founder, it could be Dubya and still not affect the almighty community process one bit.
Congratulations, you managed to pick the only Metric unit we don't use ;) (decimeter)