Truly Linux friendly Intel graphics
|
| Author | Content |
|---|---|
| CFWhitman Feb 11, 2014 10:18 AM EDT |
A few days ago BernardSwiss pointed out that a LinuxGizmos.com article described an Atom based board that had PowerVR graphics, and decent Linux support was limited to the kernels that PowerVR released a driver for. Here we actually have an Atom based board with Intel graphics, which have a well supported open source driver that Intel itself maintains. Since the release of the Bay Trail (both T and I) series of processors by Intel toward the end of last year, we should be seeing a lot more Atom based and truly Linux friendly devices appearing. As an added bonus, this series also has much improved power consumption and performance per clock cycle over previous iterations of the Atom processor. In fact, the Intel GPU on the new processors gives a much better reason to take Atom processors seriously as an alternative to ARM based processors for power efficient Linux based devices. (Though it still seems RISC based chips always have an advantage in price/efficiency ratio over CISC based ones.) |
| Bob_Robertson Feb 11, 2014 1:10 PM EDT |
> (Though it still seems RISC based chips always have an advantage in price/efficiency ratio over CISC based ones.) Ya know, I've heard this for a very long time. But if the price/performance ratio is actually better with RISC, why to supercomputer builders with relatively huge budgets and custom software use CISC chips? It can't be just because the Linux kernel was originally written for the i386. |
| CFWhitman Feb 11, 2014 3:17 PM EDT |
I didn't say the price/performance ratio. I said the price/efficiency ratio, as in power efficiency. That is, for example, it's cheaper to make a RISC chip that will play 1080p video for 10 hrs on a 31 Whr battery than it is to make a CISC one that will do the same thing. Intel can compensate for this somewhat by fabricating smaller than the competition is generally able to (smaller fabrication improves power efficiency) and by producing chips on a great scale. As for what's better for a server, where power efficiency is not as much of a concern as opposed to raw power, that argument is still going on. However, the November 2013 top ten list of fastest supercomputers has 5 CISC (4 Xeon, 1 Opteron) based systems and 5 RISC (4 Power, 1 SPARC) based systems, so I don't really think that CISC dominates in supercomputers.* All of the top ten have one thing in common, though; they all run Linux. Personally, I have my suspicions that the x86 race for the desktop computer market and the dominance of Windows has really helped keep CISC competitive on the server by pushing development for high powered computing (one purpose that AMD serves for certain is to push Intel to greater heights). RISC chips, on the other hand, have been much further developed in embedded applications because the price/power efficiency advantage was quickly recognized in that area, and there was no legacy operating system / chip architecture monopoly combination to get in the way. * Top Ten Supercomputers for November 2013: 1. Tianhe-2 (Xeon) 2. Titan (Opteron) 3. Sequoia (Power) 4. K Computer (SPARC64) 5. Mira (Power) 6. Piz Daint (Xeon) 7. Stampede (Xeon) 8. JUQUEEN (Power) 9. Vulcan (Power) 10. SuperMUC (Xeon) Source: http://www.top500.org |
| cr Feb 11, 2014 7:14 PM EDT |
> why to supercomputer builders with relatively huge budgets and custom software use CISC chips? I think one thing ARM has been facing is the fragmentation in system architecture. When you say "i386", the implication is that the machine's design outward from the CPU is ultimately derived from the PC-AT, so there's an awful lot that all those machines have in common. In ARM land, even the CPU is a variable (because ARM doesn't fab, instead licensees fab and often add their own secret sauce), let alone the architecture of the rest of the board-set, and, up until recently, they all didn't play nice in the kernel tree. Thus, anybody who wanted to put together a new ARM-based replacement for a server had to do a fair amount of kernel-level hacking and maintenance in addition to competitive hardware design, or have it done. |
| Bob_Robertson Feb 12, 2014 9:45 AM EDT |
> When you say "i386", the implication is that the machine's design outward from the CPU is ultimately derived from the PC-AT What I meant was, when Linus announced what became Linux, he noted that he was using the CISC i386 specific functionality, so it would obviously not be portable. As the list of top machines provided by CF above shows, they're based upon CISC cores. I'm quite pleased to see a SPARC machine at all, I've been a fan of the idea of OpenSPARC since its inception. Well, with electricity being more of a factor, and adding an architecture to Linux something that has been done many many times before, people will adapt when it's worth the RISC of doing so. |
| CFWhitman Feb 12, 2014 1:46 PM EDT |
One thing that I found interesting about the list was that five of the top ten machines were manufactured by IBM, all four of the Power based ones (of course) and one of the Xeon based ones. Cray produced two of the remaining five, the Opteron and one Xeon. Fujitsu manufactured the SPARC64 machine (Fujitsu being the largest backer of SPARC at this point). One of the remaining two Xeon machines was built by Dell and the other, the top machine, by NUDT (National University for Defense Technology) in China for the Chinese National Super Computer Center. IBM is obviously dedicating a lot of resources toward the manufacture of supercomputers. "Big iron" has always been one of IBM's strengths, and they want to keep it that way. |
| Bob_Robertson Feb 12, 2014 3:49 PM EDT |
> "Big iron" has always been one of IBM's strengths, and they want to keep it that way. The benefits of competition. The only way IBM can stay on top is to continuously innovate and improve their offerings. I think that is wonderful. |
| CFWhitman Feb 12, 2014 3:58 PM EDT |
@cr
I agree with your assessment of ARM not being completely consistent and making it harder to develop a general purpose operating system that runs on all the different versions. However, I'm not sure how much of this is attributable to ARM being licensed to different manufacturers. It seems to me that a lot of the differences in various versions of the processor were introduced by ARM themselves with their reference designs. Of course, various GPUs are paired with ARM CPUs by many different SoC manufacturers, and this contributes to the problem (especially because most of them have proprietary drivers). However, I think the root of the problem may lie in the fact that for a long time ARM was considered to be targeted at custom devices with custom operating systems, and it was up to device manufacturers to worry about the heavy lifting when they chose a reference design either to use or base their design on. It's really only with the advent of Android that a lot of different devices were expected to run essentially the same software stack with only a little modification by the device manufacturer. Up until Android most key people didn't seem to care if it was tough on the Linux kernel developers to cross between different versions of ARM because that type of thing was handled downstream. The problem affected hackers and hobbyists, but mobile device developers just made sure that their software stack worked on their device. Since they have started caring, newer designs have been streamlined and a lot of patches have flowed upstream and eased the burden on kernel developers (and made it so more ARM code actually got merged). Proprietary GPU drivers are still an issue though. |
| tuxchick Feb 12, 2014 6:48 PM EDT |
Speaking of IBM...everyone knows the company was built on goverment contracts, and they still rely heavily on government contracts, right? From their first tabulating machines for automating the census, to mainframes, to pretty much every technology they developed, all because of big government customers such as the US Navy and the census bureau. They have a 97% mainframe monopoly, and are a bit whiny about that missing 3%. Linux on ARM is huge, from little bitty phones and other embedded devices to whatever they call servers that are not mainframes, but still handle big loads. And soon desktops and laptops, and everything else all up and down the stack. ARM traditionally has been somewhat fragmented, as CF Whitman noted, but it's going to harmonize for Linux. All the big players are behind it, including IBM, and Samsung (who have become very interesting and busy) under their new CEO, TI, ASUS, and well, everyone. |
Posting in this forum is limited to members of the group: [ForumMods, SITEADMINS, MEMBERS.]
Becoming a member of LXer is easy and free. Join Us!