Page 1 of 1
x5000/020 vs x5000/040 tests and benchmarks
Posted: Fri Dec 10, 2021 7:52 pm
by kas1e
@All
Let's collect all information about speed of x5000/020 and x5000/040 as it looks like x5000/040 in some areas is very slow (slower than x5000/020 much). I noticed the same under AmigaOS4 with 2 different testers, so, let's see if there is the same issues under Linux, so we can rule out AmigaOS4 or not, and find out if it issue on OS side, or Uboot/Hardware/etc.
@Roland
But also I have tried to find such test program for Linux and have not found any. Instead, there is for MorphOS 'MemtTest' program, which gives with the 'vram' parameter the write and read performance for VRAM. And also there the read performance is VERY low for X5040, even lower than with OS 4.1 (e.g. 12 Mbit/sec MOS vs. 20 Mbit/sec OS4.1 with a HD 7450). So I think there is indeed something wrong at lower level than the operating systems themselves.
Do you mean you compile mos's "MemTest" for both Linux and OS4 on X5000/040 and tried it? Probably we need to put a binarys there, so we all can test it on different 020/040 machines.
Re: x5000/020 vs x5000/040 tests and benchmarks
Posted: Fri Dec 10, 2021 11:27 pm
by Roland
kas1e wrote: Fri Dec 10, 2021 7:52 pm
Do you mean you compile mos's "MemTest" for both Linux and OS4 on X5000/040 and tried it? Probably we need to put a binarys there, so we all can test it on different 020/040 machines.
No, unfortunately not... As 'MemTest' is a component of MorpOS, only the MOS developer team has access to the source code.
But as MOS can also be run under Linux using Qemu, it would be interesting to see what kind of result MemTest would give if run that way... Unfortunately I do not have all the files needed for emulation in this machine.
Christian, if you still have Qemu installed and ready to run MorphOS, could you please run "MemTest vram" in Cli window?
Re: x5000/020 vs x5000/040 tests and benchmarks
Posted: Sat Dec 11, 2021 11:05 am
by xeno74
Roland wrote: Fri Dec 10, 2021 11:27 pm
But as MOS can also be run under Linux using Qemu, it would be interesting to see what kind of result MemTest would give if run that way... Unfortunately I do not have all the files needed for emulation in this machine.
Christian, if you still have Qemu installed and ready to run MorphOS, could you please run "MemTest vram" in Cli window?
MorphOS doesn't run with KVM so it can't use the virtualization support of the CPU (KVM = Type 1 hypervisor with mapping the physical CPU directly to Virtual CPU) so it is too slow for benchmarking. Additionally MorphOS hasn't support for VirtIO devices (Paravirtualized drivers) for example the VirtIO GPU, VirtIO block devices like an HD, VirtIO input devices like keyboard and mouse. With the paravirtualized VirtIO drivers, guest I/O latency decreases and throughput increases to near bare-metal levels.
The X5000 Linux kernel supports the KVM-HV type 1 hypervisor and paravirtualized VirtIO drivers so it can works in virtual e5500 QEMU machines near to the bare-metal performance level.
Screenshot:
- Christian
Re: x5000/020 vs x5000/040 tests and benchmarks
Posted: Sat Dec 11, 2021 11:20 am
by Roland
xeno74 wrote: Sat Dec 11, 2021 11:05 am
MorphOS doesn't run with KVM so it can't use the virtualization support of the CPU (KVM = Type 1 hypervisor with mapping the physical CPU directly to Virtual CPU) so it is too slow for benchmarking. Additionally MorphOS hasn't support for VirtIO devices (Paravirtualized drivers) for example the VirtIO GPU, VirtIO block devices like an HD, VirtIO input devices like keyboard and mouse. Wit the paravirtualized VirtIO drivers, guest I/O latency decreases and throughput increases to near bare-metal levels.
The X5000 Linux kernel supports the KVM-HV type 1 hypervisor and paravirtualized VirtIO drivers so it can works in virtual e5500 QEMU machines near to the bare-metal performance level.
Ok... While waiting your answer, I got Qemu reconfigured so that I could again run MorphOS. But the 'Memtest vmem' did not work at all, it just gave this error message: "Could not lock pubscreen or allocate the vmem bitmap". The basic RAM test worked, and gave 326/495MB/sec writing speed and 130MB/sec reading speed

. Indeed far from the Ragemem results run under OS4.1...