Slashdot Mirror
ARM VP To Keynote AMD Developer Conference
MojoKid writes "AMD is hosting its first AMD Fusion Developer Summit (AFDS) this summer, from June 13-16. The conference will focus on OpenCL and upcoming AMD Llano performance capabilities under various related usage models. One interesting twist is that the keynote address will be given by Jem Davies, currently ARM's VP of technology. To date, AMD's efforts to push OpenCL as a programming environment have been limited, particularly compared to the work NV has sunk into CUDA. With its profit margins and sales figures improving, AMD is apparently turning back to address the situation — and ARM's a natural ally. The attraction of OpenCL is that it can potentially be used to improve handheld device performance. AMD's explicit mention of ARM hints that there might be more than meets the eye to this conference as well."
How about spending a few engineering dollars and releasing GOOD well documented drivers? I'm a regular reader of the XBMC forums and anyone that wants to use Linux more or less needs to buy Nvidia hardware.
I'm not in the 'anti-closed binary' camp, I just want the best tool for the job. Nvidia provides great CUDA and VDPAU support and it more or less 'just works'. ATI & Intel decided to jump on the Linux bandwagon by opening up everything and so far it seems like the community really hasn't jumped on it. I paid money for your hardware, why not pay an engineer to write software I can actually use?
When I go car shopping and the sales associate shows me 2 cars. One is completely built, works well enough and has good factory support BUT I'm not allowed to modify it. Or the second one which is actually just in a crate. It comes partially assembled... but don't worry. There is complete documentation for every single loose part and instructions on how to put it together. And the 2 cars cost nearly the same.
I'm going to choose the first car. My time IS worth something and I'd rather have something I can't modify but works great as is (NVidia's drivers) to something that really is useless unless I, or someone else, uses the documentation to do something (ATI). Especially when the hardware costs are nearly the same.
I've been hearing this drivel for years.
It's beaten out other techs for a reason. ARM is not replacing x86 on the desktop any time soon, thank goodness. Maybe joining it, but not replacing it.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
People have tried and failed before. Repeatedly. It's not impossible but it is improbable. ARM sure has a whole lot of work to do if it is to go from being an anorexic featureless simpleton CPU into a flexible *and* powerful desktop and server CPU.
Speaking to EE Times during a discussion of ARM's first quarter financial results CEO Warren East said: "AMD is a successful company selling microprocessors. ARM is in the business of licensing microprocessor designs. It is perfectly natural that we should have been trying to sell microprocessor designs to AMD for about the last ten years. Hitherto we haven't been successful." East also said: "AMD has signaled they are going through a rethink of their strategy, and that must provide a heightened opportunity for ARM. They might use ARM microprocessors in the future and you've got to expect that we would be trying to persuade them of that." http://www.eetimes.com/electronics-news/4215518/ARM-working-on-AMD-to-drop-x86
No ARM Will not replace x86 on the desktop. They point is to keep x86 away from other platforms. The desktop is what it is, a swiss army knife where a powerful CPU with a big feature set is called for, because the system will do a little of everything. Its also true that different feature sets are needed by different users but packing it into a single one size fits all chip is the way to go because the incremental cost of adding features so won't use are less that making a wider array of products. Electrical efficiency is desirable but not at the cost of features or raw performance. Nothing but perhaps POWER will ever compete with x86 there.
x86 is ill suited to mobile devices for all the same reasons its great on desktops and laptops that can afford large heavy batteries, and are generally used where AC is available.
What AMD/Marvell/VIA/NVIDIA *need* to do is keep the old WinTel mind set and lock in out of these new devices. Which they can because they are more limited in scope, and most of the software for these things is already cross platform. Its a matter of keeping it that way so the market remains open and competitive.
Repeal the 17th Amendment TODAY! Also Please Read http://www.gnu.org/philosophy/right-to-read.html
Comment removed based on user account deletion
And the reason is Intel had the resources to make more of what they were selling than anybody else and other low end players could get better sales by copying them instead doing their own thing. Itanic was competing with Intel's core product and did not have a chance within Intel. Other places got bought out by other companies that didn't see anything past Intel making a lot of money. I've got no idea what IBM are doing with Cell - try to buy something with it and they tell you to get something else.
2012 will be the year of ARM on the desktop.
It's beaten out other techs for a reason. ARM is not replacing x86 on the desktop any time soon, thank goodness. Maybe joining it, but not replacing it.
And what do you think this reason is?
Hint: it's not technical superiority.
Phillip.
Property for sale in Nice, France
Which features is the ARM architecture missing compared to x86?
Donate free food here
Features of x86 that are currently missing in ARM? How about out-of-order execution, 64-bit operation, speed boost (some cores shut down to let other cores run faster), and a top-end speed around 3GHz just to name a few.
Of course the lack of those features lets it run cooler which makes the ARM processor ideal for low-power applications like cell phones.
It's missing various addressing modes, some tiny 8-bit instructions, a separate I/O address space, and probably more. Depending on what you do, however, missing these features may itself be a feature. ;)
Except for 64-bit operations, those features have nothing to do with x86 vs ARM. They're part of the microarchitecture. And by the way, the Cortex-A9 does support out-of-order execution.
Features of x86 that are currently missing in ARM? How about out-of-order execution,
The Cortex-A9 is out-of-order.
64-bit operation,
They indeed don't have 64 bit ALU or memory addressing support yet.
speed boost (some cores shut down to let other cores run faster),
That's unrelated to the architecture. And at least NVidia's Tegra dual-core cpu's shut down one of the two cores if it's not in use. I don't think they automatically overclock the other one to run faster when doing so though.
and a top-end speed around 3GHz just to name a few.
Yes, in absolute performance per core they are still trailing x86. I was mainly reacting to the "anorexic featureless simpleton CPU" remark with my question though.
Of course the lack of those features lets it run cooler which makes the ARM processor ideal for low-power applications like cell phones.
And server farms.
Donate free food here
Server farms? Kind of a bitch without ECC but I guess it works in some cases. Granted, what do I know, maybe they will be slapping ECC support into them.
Server farms? Kind of a bitch without ECC but I guess it works in some cases. Granted, what do I know, maybe they will be slapping ECC support into them.
Why do you believe ARM doesn't support ECC right now?
Donate free food here
I agree that has been extremely important until now. As the success of Android and iOS devices demonstrates, it's getting much less so though, even in the consumer space. Even migrations from one architecture to another have been pulled off by Apple already twice in a quite successful way (although in that case increasing performance of the new architecture is quite important). And in the server space, the underlying architecture is almost irrelevant.
Donate free food here
I think the only one of those that would stand a chance is Alpha... *MAYBE* MIPS.
Sparc? Doubt it. It's nice for highly parallel tasks, but not so great for high-demand single-thread, which matches a lot of desktop apps.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
Adaptability and technological superiority for the tasks on a desktop.
So, yeah, it is in part, technological superiority.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
Actually, ARM wouldn't take much work to make a good server CPU.
Most servers are better suited by a lot of mediocre or slightly below mediocre cores, rather than one or a few heavier-duty cores. ARM's low power and high performance/power ratio makes it a very likely contender for the server market in the next couple years, if it is developed properly.
For the end-user segment (desktops and notebooks), where low-thread brute force tends to be a more relevant factor, ARM isn't as good of a choice.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
With the plethora of JIT compiled languages doing high-end tasks today, and the increasing number of cross platform/arch libraries, I'm not sure that x86 compatibility is such a killer.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
RISC architecture is going to change everything.
(I couldn't resist)
x86 is ill suited to mobile devices for all the same reasons its great on desktops and laptops that can afford large heavy batteries, and are generally used where AC is available.
This is complete BS. As ARM starts to match the performance of x86 the power draw is beginning to match too. If intel sticks it out in another generation or two i'm betting Atom will absolutely dominate the power/performance curve because there isn't anything fundamentally in x86 that makes it draw more power. If anything building an OOO ARM is harder due to the fact that every single instruction can be predicated.
Intel has learned the integration lesson too, which should bring the platform power aspects under control within a generation or two.
The "legacy" portions of x86 are such a tiny portion of the overall die as to be nearly meaningless. ARM isn't exactly legacy free either, plus they continue to have a fairly fragmented market due to a half dozen ABI's, and probably ~100 different vendors producing different versions of the chips with different memory controllers/DMA engines/etc. A SOC x86 with graphics/IO/slow memory/etc will be comparable when one gets built. ARM will continue to sell in huge numbers, but in order to absolutely control the market place they MUST have generational compatibility between devices. Otherwise, they run the risk that in any single generation they loose marketshare. This actually happens all the time, its just that their market has so far been diverse enough that it just doesn't matter, loose one vendor, gain another.
Finally, half the reason the tablets/phones seem fast isn't because they have fast hardware, but rather its a focus on building a system that works within their constrains, unlike windows or general purpose Linux. This is why no one particularly cares about x86 in a tablet. You have to customize the environment sufficiently that the ABI compatibility that x86 brings, along doesn't provide any value. No one wants a smart phone running windows 7 (hell a lot of people don't even want a desktop running windows7).
BTW: I have a fair number of ARM devices, including an OpenRD (BTW: 7 watts wall power at full load, compare with the CPU draw) which acts as the primary network server at my house. A few months ago I had this discussion with my coworker and measured the performance/watt numbers of a number of devices. The end result is that the latest Intel processors quite literally are about 10x faster than the best ARM devices you can buy, and the power/watt is roughly comparable even though the server is eating 150-300 watts (consider disks/etc) and the arms are eating 10.
s/is going to change/has changed/
Given that, to my knowlege, all x86 CPUs are internally RISC, that change may be necessary.
However, there's nothing wrong with exposing the x86 arch while internally using their own, separate, instruction set - this allows a model which can perform standard optimizations, keeping the programmers from having to worry about implementing them, themselves. Also, it allows for changing the executed instruction set, to improve performance, without causing problems to existing applications.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
> ARM is not replacing x86 on the desktop any time soon, thank goodness. Maybe joining it, but not replacing it.
That's a beginning: when someone use an iPad instead of a PC, he "replace" an x86 by an ARM.
Currently iPads and the like are mostly PC 'companions' but it wouldn't surprise me that they could become PC replacements for many users..
Nice try, Intel.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
I feel like x86 compatibility itself doesn't matter anymore either. The majority of users seem to depend on only a very small number of applications. You pretty much get all the average folk with a web browser, Flash, Microsoft Office, and maybe iTunes. Adobe, Microsoft, and Apple have demonstrated a willingness to work with whatever platforms are popular.
There are certainly large niches that matter too, like video games, but would companies/developers for those applications hold things up? I don't know.