What about the X1000 video bus performance?

[takemehomegrandma]

OK, I was reading in this thread, and the first post contains some "RageMem" benchmarks between various OS4 machines. In the same thread, there is this post from Vox that adds X1000 numbers to the picture.

I have merged those two posts below and done some cleaning and rearranged the order to be fastest first, then descending. I also marked the X1000 in red to separate it from the G4's.

Code: [Select]

--- CPU ---

MAX MIPS: 4194 // a1-xe @ 1.4 Ghz
MAX MIPS: 3797 // peg2 @ 1266 Mhz
[COLOR=Red]MAX MIPS: 3084  // X1000 @ 1.8 Ghz[/COLOR]

It's not news (but still surprising to many who expected more from reading marketing materials about the PA6T) that the X1000 performs worse than aggressively clocked G4's. Many benchmarks has confirmed this. The published specs about the PA6T (8800 MIPS, dualcore, wasn't it?) simply doesn't seem correct. Ah, well. It was a dead-end CPU before it was seriously commercialized anyway, so I guess it doesn't matter anymore.

Code: [Select]

--- L1 cache ---

[COLOR=red]READ64: 13677 MB/Sec  // X1000 @ 1.8 Ghz[/COLOR]
READ64: 10660 MB/Sec  // a1-xe @ 1.4 Ghz
READ64:  9650 MB/Sec  // peg2 @ 1266 Mhz

[COLOR=red]WRITE32: 6850 MB/Sec  // X1000 @ 1.8 Ghz[/COLOR]
WRITE32: 4569 MB/Sec  // a1-xe @ 1.4 Ghz
WRITE32: 4136 MB/Sec  // peg2 @ 1266 Mhz

Not commenting the values, but the results pretty much goes hand in hand with the clock frequency of the CPU's. As expected; L1 caches runs at the same speed as the CPU, so the more "GHz", the higher L1 transfer values.

Code: [Select]

--- RAM ---

[COLOR=red]READ32: 2860 MB/Sec  // X1000 @ 1.8 Ghz[/COLOR]
READ32:  233 MB/Sec  // a1-xe @ 1.4 Ghz
READ32:  146 MB/Sec  // peg2 @ 1266 Mhz

[COLOR=red]WRITE64: 3388 MB/Sec  // X1000 @ 1.8 Ghz[/COLOR]
WRITE64:  645 MB/Sec  // a1-xe @ 1.4 Ghz
WRITE64:  387 MB/Sec  // peg2 @ 1266 Mhz

WRITE: 733 MB/Sec (Tricky)  // peg2 @ 1266 Mhz
WRITE: 663 MB/Sec (Tricky)  // a1-xe @ 1.4 Ghz
[COLOR=red]WRITE: 352 MB/Sec (Tricky)  // X1000 @ 1.8 Ghz[/COLOR]

I think it would be interesting to know how how many memory modules Amigakit puts in the X1000 (is both memory controllers being used)? And how fast does the memory run?

BTW, what is the "Tricky" test? The X1000 obviously doesn't like doing tricky stuff...

Anyway, here comes the point of the post:

Code: [Select]

[b]--- VIDEO BUS ---

WRITE: 221 MB/Sec  // peg2 @ 1266 Mhz with unknown video card
WRITE: 169 MB/Sec  // a1-xe @ 1.4 Ghz with Radeon 9000 Pro
[COLOR=red]WRITE: 161 MB/Sec  // X1000 @ 1.8 Ghz with Radeon HD 6870 1GB[/COLOR][/b]

First, there is nothing unexpected in the G4 machine tests.

The maximum theoretical bandwidth for AGP 1x is 266 MB/s, but that's theoretical maximum and not what you get in practice. The Pegasos 2 has "AGP 1x", and the 221MB/s in this test is similar to other benchmarks and is about what you realistically can expect from a Pegasos 2.

The A1-XE is based on the notoriously flawed Articia-S Northbridge from MAI. It was marketed and sold as an AGP 2x chip (meaning a theoretical max transfer speed of 533MB/s). However, some benchmarks (here is one) did establish a long time ago that you in real life would only reach about "AGP 0,5x" (for a G3) to "AGP 0.7x" (for a G4) on a MAI Teron board (sold as "AmigaOne" by Eyetech). This benchmark confirm this. Terrible performance from a AGP 2x computer, but again, very expected when it comes to Articia-S. We already knew this.

But the X1000?! Whoa!! What's the matter with that?!

Either the benchmark is terribly flawed someway (it worked for the G4 systems though), or something is really borked in either the X1000 or PA6T HW, or in OS4 or the driver implementations. That number is so terrible that it simply can't possibly be true!! It should be close to twenty-five times faster!

So where is the flaw?

The test? The HW? The OS/Drivers?

:confused:

[takemehomegrandma]

Normal CPU transfer over PCIe is slow. Nowhere near PCIe bandwidth. Can be slower than AGP machines. Saw it on x86 PC with AROS. Also on Linux/X11 (x11perf) with unaccelerated drivers.

Wait a minute, what are you saying here?

That there is no DMA in the drivers? Or that the test is explicitly using CPU for video bus transfers?

:confused:

[takemehomegrandma]

@Hans and itix

Hans: "The test is 'flawed.' Ragemem uses its own custom CPU copy routine to copy to VRAM, and it's not very good."

Itix: "Those tests are using CPU to manipulate video memory."

OK, thank you both for your clarifications! I learned something here!