http://keithp.com/blogs/Sharpening_the_Intel_Driver_Focus/
If you wondered why Intel sucks on Fedora read this article, it expains all complexities of Intel drivers and for me it shows hope that Intel drivers are becoming better.
Valent.
Valent Turkovic wrote:
If you wondered why Intel sucks on Fedora read this article, it expains all complexities of Intel drivers and for me it shows hope that Intel drivers are becoming better.
intel sucks on anything but ms, because intel joined the ms whore house years ago along with many other oem suppliers because of their fear of not being included in ms specs.
in off quote of b.g., 'exclusively ms or be left out'.
many 'old heads' are aware of this, and it is what has made building linux drivers for a lot of good software very difficult when oem refuses to release specs on systems.
given enough time, driver writers will be able to over come the ms monopoly so that there will be more hardware working in other than ms.
page was a good read. thanks for posting.
On Saturday, May 16th 2009 at 15:00 -0000, quoth g:
=>Valent Turkovic wrote: => =>> If you wondered why Intel sucks on Fedora read this article, it =>> expains all complexities of Intel drivers and for me it shows hope =>> that Intel drivers are becoming better. => =>intel sucks on anything but ms, because intel joined the ms whore house =>years ago along with many other oem suppliers because of their fear of =>not being included in ms specs. => =>in off quote of b.g., 'exclusively ms or be left out'. => =>many 'old heads' are aware of this, and it is what has made building =>linux drivers for a lot of good software very difficult when oem refuses =>to release specs on systems. => =>given enough time, driver writers will be able to over come the ms =>monopoly so that there will be more hardware working in other than ms.
I need to go back and see the beginning of this thread, just in case anyone thinks I know what I'm talking about...
But for years now, I have had a marked preference for AMD processors. This is based on my past experience in compilers, especially in optimizers.
The Intel line is so massively pipelined that it's almost impossible to write an optimizer that doesn't have to flush its entire cache every few instructions.
Not that it matters, but the Alpha chip was one impressive processor. That thing could clock off as many as 6 instructions per clock tick because of intelligent pipelining.
The AMD line doesn't run as fast as the Intel chips on a clock-cycles per dollar basis, but it makes up for it in how it caches instructions. That's both why they're comparable and why it's even possible to write an optimizer for it.
Gcc is a truly amazing piece of work, and it certainly even qualifies as art. It supports a very wide range of architectures, but I would have to say that it does a better job of optimizing AMD code than Intel code. Then facton in that the processor itself does a better job of caching instructions, and you can see where I get my preferences.
Valent Turkovic wrote:
If you wondered why Intel sucks on Fedora
I cannot say that "Intel sucks on Fedora", but some problems are apparent (like, kms won't work on i915, like the display freezing when trying to play videos, etc, but this sounds like a new 'paradigm' for the display driver, so we need to be patient (like we were when the desktop's new paradigm took shape in 2008 - and how it has matured!). Like the article says, there are so many interleaved projects, like intel driver, xorg, mesa, compositing, dri2, and whatnot. I am optimistic that it will be wonderful soon.
Steven W. Orr wrote:
On Saturday, May 16th 2009 at 15:00 -0000, quoth g:
=>Valent Turkovic wrote: => =>> If you wondered why Intel sucks on Fedora read this article, it =>> expains all complexities of Intel drivers and for me it shows hope =>> that Intel drivers are becoming better. => =>intel sucks on anything but ms, because intel joined the ms whore house =>years ago along with many other oem suppliers because of their fear of =>not being included in ms specs. => =>in off quote of b.g., 'exclusively ms or be left out'. => =>many 'old heads' are aware of this, and it is what has made building =>linux drivers for a lot of good software very difficult when oem refuses =>to release specs on systems. => =>given enough time, driver writers will be able to over come the ms =>monopoly so that there will be more hardware working in other than ms.
I need to go back and see the beginning of this thread, just in case anyone thinks I know what I'm talking about...
I don't know what I'm talking about either.
But for years now, I have had a marked preference for AMD processors. This is based on my past experience in compilers, especially in optimizers.
AMD seems to be having some manufacturing problems, as well as financial problems, but they do know CPU design. Some of us always cheer for the underdog. And it looks like AMD just won big in the EU court decision.
The Intel line is so massively pipelined that it's almost impossible to write an optimizer that doesn't have to flush its entire cache every few instructions.
Intel finally realized that pipeline flushing was the main thing the processor was doing. The "new" (I7) architecture has fixed this problem, with very impressive results.
Not that it matters, but the Alpha chip was one impressive processor. That thing could clock off as many as 6 instructions per clock tick because of intelligent pipelining.
Most of the Alpha crew went to work for Intel. I think it paid off for Intel.
The AMD line doesn't run as fast as the Intel chips on a clock-cycles per dollar basis, but it makes up for it in how it caches instructions. That's both why they're comparable and why it's even possible to write an optimizer for it.
You'll notice that both AMD and Intel have turned down the clocks speeds and are trying to make up lost performance with more cores per chip. For some HPC workloads, it isn't working.
Gcc is a truly amazing piece of work, and it certainly even qualifies as art. It supports a very wide range of architectures, but I would have to say that it does a better job of optimizing AMD code than Intel code. Then facton in that the processor itself does a better job of caching instructions, and you can see where I get my preferences.
Agreed. You'll see better performance on Intel chips using the Intel (ICC) compiler.
Regards,
John
| From: Steven W. Orr steveo@syslang.net
| The Intel line is so massively pipelined that it's almost impossible to | write an optimizer that doesn't have to flush its entire cache every few | instructions.
Don't confuse companies with technologies.
What you don't like is the Pentium 4 design (and things with the same design -- some Celerons, some Xeons, Pentium D, ...). It is the processor with such long pipelines.
The Pentium 3 was fine. AMD's K7 was better.
Intel thought it could regain the lead with P4. It worked for a while but the experiment seems to have failed in the end. The P4 is history. (Funnily enough, I bought my first P4 CPU today.)
The Intel Core 2 is technically ahead of the latest AMD products in most ways. The i7 finally gets an on-chip memory controller (as AMD did way back when, copying the Alpha, if I remember correctly).
| From: g geleem@bellsouth.net
| intel sucks on anything but ms, because intel joined the ms whore house | years ago along with many other oem suppliers because of their fear of | not being included in ms specs.
I actually subscribe to your conspiracy theory.
Having said that, there are levels of evil.
Intel has been the most supportive of the major video controller makers historically. For example, Keith Packard is a long-time core X developer and Intel pays him for things that benefit X in general as well as the Intel X drivers.
ATI (now AMD) was open with specs, to a very useful degree, until roughly 10 years ago. Then they closed up tight. A couple of years ago, they started to make a large effort opening up. That's still not complete. I don't know of anyone from ATI contributing to X in general the way Keith is doing. They are in a tough spot financially.
nVidia is the laggard. On the other hand, many in the Linux community appreciate their well-maintained closed source Linux driver. I'm not one of them.
On Sun, May 17, 2009 at 4:52 AM, Petrus de Calguarium kwhiskerz@gmail.com wrote:
Valent Turkovic wrote:
If you wondered why Intel sucks on Fedora
I cannot say that "Intel sucks on Fedora", but some problems are apparent (like, kms won't work on i915, like the display freezing when trying to play videos, etc, but this sounds like a new 'paradigm' for the display driver, so we need to be patient (like we were when the desktop's new paradigm took shape in 2008 - and how it has matured!). Like the article says, there are so many interleaved projects, like intel driver, xorg, mesa, compositing, dri2, and whatnot. I am optimistic that it will be wonderful soon.
For me the performance is the mail issue. I tried a simple game on Fedora 10 and Ubuntu - "World of Goo" and it is not playable because it is too slow on Fedora, but on Vista with same hardware (i945gm) it works like a charm
.
Petrus de Calguarium wrote:
Valent Turkovic wrote:
If you wondered why Intel sucks on Fedora
I cannot say that "Intel sucks on Fedora", but some problems are apparent (like, kms won't work on i915, like the display freezing when trying to play videos, etc, but this sounds like a new 'paradigm' for the display driver, so we need to be patient (like we were when the desktop's new paradigm took shape in 2008 - and how it has matured!). Like the article says, there are so many interleaved projects, like intel driver, xorg, mesa, compositing, dri2, and whatnot. I am optimistic that it will be wonderful soon.
Any Relevance? http://feedproxy.google.com/~r/Phoronix/~3/0t49mIt2Oy0/vr.php
Frank
Frank Murphy (Frankly3d) wrote:
Any Relevance? http://feedproxy.google.com/~r/Phoronix/~3/0t49mIt2Oy0/vr.php
I really couldn't say, as I have no understanding of the workings of these things, but, now that you point it out, it seems quite plausible that these symptoms are caused by the stripping of old supports.
g wrote:
intel sucks on anything but ms, because intel joined the ms whore house years ago along with many other oem suppliers because of their fear of not being included in ms specs.
in off quote of b.g., 'exclusively ms or be left out'.
Balderdash! Intel appears to be working quite hard to make graphics on Linux work well. They're certainly putting money into developing the drivers. The problem currently seems to be that a lot of different areas of the driver have been re-architected at once, and rapid development has never been known to produce good quality. Sad, but true. We'd probably be better off if there were a "stable" branch of the Intel driver.
On Sat, 2009-05-16 at 21:07 -0700, john wendel wrote:
Intel finally realized that pipeline flushing was the main thing the processor was doing. The "new" (I7) architecture has fixed this problem, with very impressive results.
I think you're confusing this with the original Core architecture which more than halved the number of pipeline stages relative to the later models in the Netburst family (up to 31 stages for a late-model Pentium-D, down to just 14 in Core/Core2/Nehalem).
The major change with Nehalem is the on-die memory controller and switch from FSB to NUMA multiprocessor organisation using the QuickPath Interconnect (QPI).
Regards, Bryn.
g wrote:
Valent Turkovic wrote:
If you wondered why Intel sucks on Fedora read this article, it expains all complexities of Intel drivers and for me it shows hope that Intel drivers are becoming better.
intel sucks on anything but ms, because intel joined the ms whore house years ago along with many other oem suppliers because of their fear of not being included in ms specs.
in off quote of b.g., 'exclusively ms or be left out'.
many 'old heads' are aware of this, and it is what has made building linux drivers for a lot of good software very difficult when oem refuses to release specs on systems.
given enough time, driver writers will be able to over come the ms monopoly so that there will be more hardware working in other than ms.
page was a good read. thanks for posting.
Contrary to your unsubstantiated statements Intel is in fact very pro linux, and runs it inhouse.
My PC with the linux video drivers works with EVERY linux version that I have tried (COMPIZ and more)
OK, No 3D, but for the desktop or server platforms, it is A1.
On 05/16/2009 04:14 PM, Steven W. Orr wrote:
Not that it matters, but the Alpha chip was one impressive processor. That thing could clock off as many as 6 instructions per clock tick because of intelligent pipelining.
Note that many of the Intel compiler people are former Digital compiler people.