Loongson Hardware Is Still A Tough Find

[phoronix]

Phoronix: Loongson Hardware Is Still To Tough Find

While the Loongson MIPS64 CPUs have been available for a while now as a Linux-friendly chip, they are still tough to find in the western countries. New benchmarks reveal that the ARM SoCs are becoming a much more compelling offer for those caring about performance...

http://www.phoronix.com/vr.php?view=MTI0NzE

[necro-lover]

Phoronix: Loongson Hardware Is Still To Tough Find

While the Loongson MIPS64 CPUs have been available for a while now as a Linux-friendly chip, they are still tough to find in the western countries. New benchmarks reveal that the ARM SoCs are becoming a much more compelling offer for those caring about performance...

http://www.phoronix.com/vr.php?view=MTI0NzE

Its a well known fact that the first performance version of the Loongson will be the:
Loongson 3B-32nm with 8mb L3 cache@1,3ghz coming for sale in the first quarter of 2013
All other versions before that version was just "development" versions for developers.
And yes this one will be 3-4 times faster than the Loongson 3A.
Yes without there own GPU its just worst because the lowend-amd-chipset-gpu's used with the open-source drivers are just a nightmare.
But they work on a GPU with only OpenGL+OpenCL support without directX support for the year 2014.

Right now its pointless to buy Loongson hardware but in 2013 there 8 core and 16 core(28nm) will hit the market then maybe its a more interesting comparison.

[LinuxID10T]

Uselessly slow and power hungry nice...

[chris8976]

Phoronix: Loongson Hardware Is Still To Tough Find

While the Loongson MIPS64 CPUs have been available for a while now as a Linux-friendly chip, they are still tough to find in the western countries. New benchmarks reveal that the ARM SoCs are becoming a much more compelling offer for those caring about performance...

http://www.phoronix.com/vr.php?view=MTI0NzE

I own a Lemote Yeelong Netbook, and I think that software support is a much bigger problem than the not-quite-competitive performance. The days where I compiled the kernel myself are long gone (and I don't want to go back there), to say nothing of compiling X, Gnome, ...

On my Lemote, I have a partial version of Fedora 13 and more complete Debian. Even Debian still has some holes - e.g. there is a OpenJDK, but without JIT compilation, so Java is more-or-less useless. And it uses the o32 abi which doesn't make full use of the processors capabilities. Things may be different if you can use the chinese OS versions, but unless some of the bigger distros decides to support mips as a tier-one target, I consider this a lost cause.

BTW, I also did some simple benchmarks, which you can find here:
http://bokesan.blogspot.de/2011/05/l...enchmarks.html

Even though my Stream TRIAD result is quite a bit better than the one on openbenchmarking, I think that bad memory performance is one of the major reasons for the bad overall performance.

[willmore]

Phoronix: Loongson Hardware Is Still To Tough Find

Should be "Too Tough" unless you're going to tough.

[crazycheese]

Should be "Too Tough" unless you're going to tough.

Or
Loongson Hardware Is Still Too Tough To Find
Or
Loongson Hardware Is Still A Tough Find

But they work on a GPU with only OpenGL+OpenCL support without directX support for the year 2014.

Completely open hardware, implementing only open standards comes only from China.

Bite me please.

[artivision]

Uselessly slow and power hungry nice...

The L3C part (2013) will have 1024bit-Fmac for floating point and another similar length for integer, that is 12-15 dmips/mhz wile Ivybridge is 9.5. All this with a consumption less than an Atom, 43m transistors per core vs 47mt of an Atom. Also Mips can have a less complex 512bit-Fmac (2.5drystone) interface, with 1mt logic and 1,5mt L1 memory, make that 4mt with integrated 3D instructions and shared cache. So i prefer an L3C@2.5ghz-28nm instead of an 1.5-2ghz Atom-32nm, its 5-6 times faster and 3.5-4 times on x86 emulation mode. Also Intel is giving 20mt per 256bit-Fmac graphics and in the future probably 10mt, Mips and probably others can do that with 1/5 the transistors and with just a software rasterizer. So if you have 16 small cores with 32 instructions and 2.5ghz you have 1.3tflops.

[necro-lover]

The L3C part (2013) will have 1024bit-Fmac for floating point and another similar length for integer, that is 12-15 dmips/mhz wile Ivybridge is 9.5. All this with a consumption less than an Atom, 43m transistors per core vs 47mt of an Atom. Also Mips can have a less complex 512bit-Fmac (2.5drystone) interface, with 1mt logic and 1,5mt L1 memory, make that 4mt with integrated 3D instructions and shared cache. So i prefer an L3C@2.5ghz-28nm instead of an 1.5-2ghz Atom-32nm, its 5-6 times faster and 3.5-4 times on x86 emulation mode. Also Intel is giving 20mt per 256bit-Fmac graphics and in the future probably 10mt, Mips and probably others can do that with 1/5 the transistors and with just a software rasterizer. So if you have 16 small cores with 32 instructions and 2.5ghz you have 1.3tflops.

i think there is another reason why they don't will win any benchmark: because they do not have the man-power to fix all the source code of all apps to use all these SIMD features.
because without hand optimised code its only a 8/16 core 64bit integer cpu@1,3ghz with the need of 2 clocks per instruction.

and compared to this intel do have the market power to push there SIMD hand optimised code into every app closed and opensource.

[smitty3268]

i think there is another reason why they don't will win any benchmark: because they do not have the man-power to fix all the source code of all apps to use all these SIMD features.
because without hand optimised code its only a 8/16 core 64bit integer cpu@1,3ghz with the need of 2 clocks per instruction.

and compared to this intel do have the market power to push there SIMD hand optimised code into every app closed and opensource.

99% of apps just rely on whatever is in the compiler to automatically generate code.

[necro-lover]

99% of apps just rely on whatever is in the compiler to automatically generate code.

Sure and the people always only benchmark the 1% of the apps just because they need the speed in the 1% of the apps.

In the end you buy your hardware because of these benchmarks in the 1% of the apps.