How fast is the Alchemy CPU?

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
There sure seems to be a gain of momentum in the mobile market with faster ARM-based CPUs, and if I understand correctly, the Cortex A8 yields up to 800 MIPS @600Mhz and the Dragonfly up to 2000MIPS @ 1GHz ....

Then there are those Alchemy processors which can be found in some of Cowon's devices. They sure brag how powerful they are and how many megahertz they can do but nothing about it's performance.

So, does anyone know how well they perform on benchmarks, how many MIPS, FLOPS or at least CPI does the Au1250 and the Au1300 do? I suspect they are not that good for otherwise the manufacturer wouldn't put such an effort to hide this from the sheets.

 

wuzy

Distinguished
Jun 1, 2009
900
0
19,010
If they're based on ARMv11 then they should be fairly close to Cortex-A8 (an extension of ARMv11) in terms of MIPS/Hz.

Cowon are multimedia devices, so the the focus on them should be about software being optimised to use built-in codec decoding extensions for lower CPU usage rather than pure MIPS number. If used in a gaming device then the focus shifts to support and performance of OpenGL ES 2.0.

Pure crunching power on ARM based CE devices aren't as significant today.
 

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
The Alchemy processors feature a stripped down MIPS32-based core. MIPS32 is a 32-bit RISC architecture meant to compete with the ARM architecture, so it is not ARM-based. When looking at the ARM CPUs alone we see that the Instructions-Per-Cycle (DMIPS/MHz) vary widely from 1 to 2 IPC from which one can conclude that a 600MHz CPU yields anywhere from maybe 400 up to 1200 MIPS.

If I understand computer architecture correctly; the key trait of a CPU that enables it to yield up to 1 IPC is called pipelining, and the key trait that enables one core to yield more than 1 IPC is called superscalarity (see superscalar pipelines). When reading through the MIPS32/64 manuals superscalarity is mentioned but it is not mentioned anywhere in the Alchemy documentation so I assume that the Alchemy CPUs are not superscalar (if it were, it would be like a sportscar manufacturer/seller who markets a car without mentioning what engine there is inside, which is not likely).

Now, the SOC implementation is an advantage which can mean that the Alchemy circuit may pack a punch in terms of performance, but I seriously doubt is measures up to Cortex and Dragonfly based units.

So I don't think it is as hot as the ARM CPUs in terms of performance but it may have an advantage when looking at the power consumption alone.
 

TRENDING THREADS