A1222 vs Sam460LE benchmarks?

Are there any benchmarks to compare CPUs on those and different GPUs running under both of those system? Just curious.

@PixelHi
for quick view I recommend The QuickStart Guide for New A1222+ Users from Eliyahu.

@PixelHi

In my opinion and based on what I analysed so far - the only area where SAM460 is faster/better is FPU speed for standard FPU code.

Of course its hard to tell when will this bother you since all other parts of A1222 are better/faster than SAM460. This question is still not answered since improvements on A1222 are still comming. For example L2 cache speed increased dramatically in last update.

Below is a good benchmark (newest stream-5.10-AOS from Aminet/OS4depot) that can ilustrate FPU emulation vs native SPE performance on A1222:

Emulated FPU (standard FPU code)
Function Best Rate MB/s Avg time Min time Max time
Copy: 842.5 0.194572 0.189902 0.201057
Scale: 150.8 1.063566 1.061034 1.066340
Add: 160.0 1.510033 1.499692 1.524443
Triad: 153.6 1.569161 1.563005 1.577607

A1222 SPE FPU (SPE code)
Function Best Rate MB/s Avg time Min time Max time
Copy: 831.8 0.197500 0.192355 0.211105
Scale: 542.5 0.301120 0.294940 0.309383
Add: 617.1 0.400034 0.388917 0.413630
Triad: 592.5 0.413958 0.405089 0.420792

I dont have SAM460 benchmark results for the newest stream-5.10-AOS to compare, but maybe owners of SAM460 can provide us with this info?

But it is clearly visible that emulated code vs SPE code for FPU operation on A1222 is somehow only "4x slower" in speed than native code. This really shows how SPE FPU emulator is very effective and fast implementation. Usually when you emulate FPU on CPU that does not have real FPU - its not usable at all. Speed number are then unusable. On A1222 FPU emulation si usable because A1222 has SPE FPU with its own math function so emulator translates those to standard FPU. There are equivalent functions but with speed penalty.

But in general you dont really want to use very intensive FPU code on A1222 via FPU emulation. It not usable. Only SPE port can solve that!

However A1222 SPE FPU is also in the same time a 64-bit vectorial unit (like "small" Altivec). I cant wait moment when developers start to explore that advantage. There are vectorial instructions in A1222 SPE unit. Right now ports to A1222 are just recompile to use SPE math functions instead of standard ones.

Also important thing to compare A1222 vs SAM460 is that P1022 in A1222 is dual core. If multicore arrives, even the most primitive one - then there will be no base to compare it to SAM460. It will be much faster.

The only problem I see now with SAM460 is effective support for modern graphic cards (RX cards, GART). If I understand correctly GART enables fast memory transfers between main memory and GFX card memory. But that may be solved in the future.

stream v5.10 - Sam460LE 1.14GHz:

Function Best Rate MB/s Avg time Min time Max time
Copy: 779.3 0.206296 0.205306 0.210359
Scale: 372.6 0.431840 0.429410 0.436890
Add: 447.7 0.540051 0.536067 0.542271
Triad: 443.4 0.544179 0.541314 0.547519

for explanation, stream tests double-precision floating-point values this way:
Copy: a(i) = b(i)
Scale: a(i) = q * b(i)
Add: a(i) = b(i) + c(i)
Triad: a(i) = b(i) + q*c(i)

SAM460ex 1.15GHz

Thank you! I would say it is comparable to A1222 (as a single core obviously) and considering price difference between both I will be happy with Sam460le for a while.

@PixelHi

I'm really happy with my Sam460LE (alongside with my X1000) : I can take it with me when going in events.

Please bear in mind that the infamous GART problem is still present : 3D is nowhere as fast as it should be.

Hans tried to circumvulate this problem but never managed to find a solution :(

Too bad since many games are quite slow because of this.