X86 Equivalent For Arm Architecture?

[cbp]

I googled around but I couldn't find exact answers on this.

What are the equivalent x86 processors* (CPU, GPU) for:
- ARM® Cortex™-A8 600Mhz+ CPU
- 430-MHz TMS320C64x+™ DSP Core

*equivalent processors for mobile devices not desktops. Also, concentrate only on processing power and memory/cache.

Thanks,

 

[cbox]

It's like comparing apples and oranges.

 

[sindbad]

In very broad and erroneous terms (Dhryston MIPS), the cortex is more powerfull than an Intel Atom and even a VIA Nano.

The DSP (Digital Signal Processor) is made for tasks such as video/audio decoding/encoding.

 

[cbp]

In very broad and erroneous terms (Dhryston MIPS), the cortex is more powerfull than an Intel Atom and even a VIA Nano.

The DSP (Digital Signal Processor) is made for tasks such as video/audio decoding/encoding.

Aren't there any benchmarks that can be done? i.e. 2D/3D rendering times, calculating PI to 1M decimals, etc...


I know that they are different, but in terms of raw processing power, what would the equivalent be?

 

[Squidge]

I know that they are different, but in terms of raw processing power, what would the equivalent be?

Benchmarks are useless. Actual performance depends on the application. x86 is better at some things, Cortex is better at others. You can't say a xMhz Cortex is equivalent to a yMhz x86, as they will both perform differently given different tasks.

It's not really fair to run the same code on both and compare results that way either. For example, for floating point work, you would use the FPU on the x86, but the Neon on the Cortex. Using the FPU on the Cortex would make it much slower than the x86, but the x86 doesn't have Neon...

 

[Kyo]

perhaps we should create a crossarchitectural unit for processor power

let's call it megagooditude

 

[mechanizeddeath]

It's even become difficult to compare x86 processors to each other, let alone to another architecture. Try looking up x86 benchmarks, and even in a situation where model B is arguably better than model C, there will always be that one test, maybe even two, where model C performed better.

Like squidge said, actual performance depends on the application. You just can't accurately compare the Cortex to the AMDtel Pentathlon Octocore Extreme VIII.

 

[cbp]

Benchmarks are useless. Actual performance depends on the application. x86 is better at some things, Cortex is better at others. You can't say a xMhz Cortex is equivalent to a yMhz x86, as they will both perform differently given different tasks.

It's not really fair to run the same code on both and compare results that way either. For example, for floating point work, you would use the FPU on the x86, but the Neon on the Cortex. Using the FPU on the Cortex would make it much slower than the x86, but the x86 doesn't have Neon...

Thanks for the clarification.

If we look at ARM11 MPCore and Cortex A8 which one has the edge?

Would applications/games be compatible between these two without the need to be ported?

I am saying this because if NVidia releases its Tegra APX there will be loads of apps/games flowing towards the ARM architecture and thus pandora.

 

[Squidge]

"The ARM11TM MPCoreTM synthesizable processor implements the ARM11 microarchitecture and can be configured to contain between one and four processors delivering up to an aggregate 2600 Dhrystone MIPS of performance."

Well, if it's configured for four processors, each with 64KB of cache, it's going to be a little nippier than the Cortex (assuming all cores are used properly)

As for compatability, Linux binaries that just use basic ARM features will typically run on all ARM processors (within reason - for example, it's possible to create code that will work on Gp2x, Wiz and Pandora without detecting device type.). However, use something specific thats not available in the other ARM, and it'll obviously crash on the later chip when it reaches the unimplemented stuff, such as trying to use Neon instructions on a gp2x or Wiz.

 

[Exophase]

However, Tegra is a single core implementation of ARM11 MP, and is thus inferior to Cortex-A8 clock for clock.

 

[icurafu]

Once the Craigix's Ministry of "information" gets formally setup, we should compare a Menlow 833MHz (lowest) MID playing Q3A timedemo against a Pandora clocked to 833MHz to show the Pandora after the optimisations.

 

[Squidge]

Fruit discussion moved here: http://www.gp32x.de/board/index.php?showtopic=44052