There's an oppertunity for compilers to explore which - afaik - they don't do as of yet.
Modern chips have PRELOAD instructions, which allow you to specify an address for which the CPU will attempt to preload the cacheline in advance.
Graphics drivers typically exploit this when accessing vertex buffers and whatnot. The Preload instruction will go out and fetch the data in parallel to executing what it already has in cache.
For a compiler to do this, it'd need to recognize *p++ sequences in for loops and insert an additional PRELOAD instruction after executing the increment.
Do-able, maybe Intels compiler does it, or maybe VectorC - does anyone know?
What I couldn't find in the article is that it is possible to reach the maximum rate to SDRAM as an SDRAM chip has multiple banks - eg. one can issue a command to one bank, while still receiving data from another.
This avoids incurring latency since read commands can be issued in parallel to incurring the cas latency.
There's more details on this in the SDRAM specification (lost the URL but it's out there - I think its Intel who wrote it though).
Maybe I can contribute some thoughts on this since I'm currently 'on assignment' as they call it here (eg. internationally relocated).
Working for a multinational is a great idea if you want to go international and see something of the world, though since it is quite pricey (about 3x what you cost normally), you would need to give your company a very very very good reason as to why your presence elsewhere is absolutely critical.
The easiest way to get there is to join a big IT firm, start a career, make a name for yourself and work your way internationally. Working at a big firm can be frustrating at times, but hang in there, and keep focus on going international.
Then, if you're lucky, you should find yourself 'on assignment' after about 4-5 years..
Anyway, hope it helps, these things tend to be easier if you have "critical" skills, some sense of politics (friends in the right places), and then lots of luck.
I think the tough part is not in getting the specs (there are some out there for decent cards),
the tough part is, imesho, in writing the whole 3D infrastructure within your OS.
I for one have not seen a single hobby-os with decent 3D support, despite there being specs out there.
There's more to a 3D infrastructure than drivers alone, infact, if you re-use the x drivers like someone else suggested, your driver problems are pretty much solved.
Judging from the review : RH8 is still a geek-release of linux (not integrated GUI, and techy bits (X nVidia support drama) get in the way).
For RH to get a slice from the business mainstream, the application support and integration has to be improved.
This will inevitably involve masquerading more of the KDE and Gnome efforts into one coherent whole (no doubt to understandably great upset of contributors).
And while they're at it, I think eventually conclude that the typical joe-desktop will only need one office suite, not multiple.
Only one can win :
- Either a desktop suite (KDE/Gnome) grows dominant and it's applications provide the needed integration & support.
- Or RedHat does the integration for them, to great upset of the community.
We'll see.
Slashdot Mirror
If anyone writes an anti-spam tool, I need to distinguish between spam and non-spam, making non-spam equally valuable for spam-filter benchmarking.
Having a log with only spam makes it quite easy to achieve a 100% benchmark (simply reject it all!).
Couldn't find anything about this on the site, so unless I'm missing something, the value of such a log is limited at best.
If we could even remotely compare a brain to a CPU we'd have found true artificial intelligence a long time ago.
I'm probably just stating the obvious, but this marketing BS is seriously deteriorating the hope of developing sentient computers.
Computers are far superior to the human brain in the area for which they were designed : Computing <period>.
(It's not like we're living in the fifties sheesh)
(I'm serious - why?)
Modern chips have PRELOAD instructions, which allow you to specify an address for which the CPU will attempt to preload the cacheline in advance.
Graphics drivers typically exploit this when accessing vertex buffers and whatnot. The Preload instruction will go out and fetch the data in parallel to executing what it already has in cache.
For a compiler to do this, it'd need to recognize *p++ sequences in for loops and insert an additional PRELOAD instruction after executing the increment.
Do-able, maybe Intels compiler does it, or maybe VectorC - does anyone know?
This avoids incurring latency since read commands can be issued in parallel to incurring the cas latency.
There's more details on this in the SDRAM specification (lost the URL but it's out there - I think its Intel who wrote it though).
Working for a multinational is a great idea if you want to go international and see something of the world, though since it is quite pricey (about 3x what you cost normally), you would need to give your company a very very very good reason as to why your presence elsewhere is absolutely critical.
The easiest way to get there is to join a big IT firm, start a career, make a name for yourself and work your way internationally. Working at a big firm can be frustrating at times, but hang in there, and keep focus on going international.
Then, if you're lucky, you should find yourself 'on assignment' after about 4-5 years..
Anyway, hope it helps, these things tend to be easier if you have "critical" skills, some sense of politics (friends in the right places), and then lots of luck.
the tough part is, imesho, in writing the whole 3D infrastructure within your OS.
I for one have not seen a single hobby-os with decent 3D support, despite there being specs out there.
There's more to a 3D infrastructure than drivers alone, infact, if you re-use the x drivers like someone else suggested, your driver problems are pretty much solved.
Judging from the review : RH8 is still a geek-release of linux (not integrated GUI, and techy bits (X nVidia support drama) get in the way). For RH to get a slice from the business mainstream, the application support and integration has to be improved. This will inevitably involve masquerading more of the KDE and Gnome efforts into one coherent whole (no doubt to understandably great upset of contributors). And while they're at it, I think eventually conclude that the typical joe-desktop will only need one office suite, not multiple. Only one can win : - Either a desktop suite (KDE/Gnome) grows dominant and it's applications provide the needed integration & support. - Or RedHat does the integration for them, to great upset of the community. We'll see.