QEMU Emulation vs Hardware CPU Benchmarks

I'm interested to see the raw CPU performance of QEMU on various machines vs real AmigaOS 4.x hardware. So, I've put together a smal script with a set of CPU benchmarks.

EDIT: You can download the benchmarks here.

Please post the results here (deleting the unneeded extra output in the Dhrystone & Quicksort tests), together with you're system's specs.

Hans

Here are my results:
System: QEMU 8.0.0 running on a Core i7 @ 2.60GHz

CPUBench v1.0


RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Motorola MPC 7447/7457 Apollo 1.1 @ 1533 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 32

---> CPU <---
MAX MIPS: 3119

---> L1 <---
READ32: 2034 MB/Sec
READ64: 2930 MB/Sec
WRITE32: 1881 MB/Sec
WRITE64: 3583 MB/Sec

---> L2 <---
READ32: 1950 MB/Sec
READ64: 2689 MB/Sec
WRITE32: 1602 MB/Sec
WRITE64: 3539 MB/Sec

---> RAM <---
READ32: 1433 MB/Sec
READ64: 2280 MB/Sec
WRITE32: 348 MB/Sec
WRITE64: 2523 MB/Sec
WRITE: 789 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 2105 MB/Sec
WRITE: 1773 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 1007.81 MB/sec
2 K Element : 998.39 MB/sec
4 K Element : 954.54 MB/sec
8 K Element : 982.14 MB/sec
16 K Element : 968.40 MB/sec
32 K Element : 981.72 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 240.00 BogoMips



Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends

Duration in seconds: 100.3
Microseconds for one run through Dhrystone: 2.0
Dhrystones per Second: 498517.2
Dhrystone MIPS (DMIPS) 283



Quicksort:
Total time taken by CPU: 7.84


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.002365 0.002365 701.0
10000 2261 19997 0.002887 0.003128 649.5
20000 4202 39989 0.005775 0.005875 700.7
40000 7836 79999 0.011550 0.011749 709.2
80000 14683 160001 0.023100 0.027161 620.7
160000 27607 319993 0.046199 0.054932 620.5
320000 52073 639997 0.092398 0.114746 600.3
640000 98609 1279997 0.184797 0.244141 569.9
1280000 187133 2559989 0.369594 0.786133 357.3
2560000 356243 5119997 0.739188 1.230469 460.6
5120000 679460 10239989 1.478376 2.656250 430.4
10240000 1299068 20479999 2.956751 8.632812 267.0
20480000 2488465 40960001 5.913503 44.218750 105.1

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.004554 (sec)
High MIPS = 709.2
Low MIPS = 105.1



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 43.7 seconds
C Converted Double Precision Whetstones: 114.4 MIPS


All tests done.

Machine: A1222

CPUBench v1.0


RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Freescale P10XX (E500 core) 5.1 @ 1199 Mhz
Caches Sizes: L1: 32 KB - L2: 256 KB - L3: none
Cache Line: 64

---> CPU <---
MAX MIPS: 2398

---> L1 <---
READ32: 4402 MB/Sec
READ64: 330 MB/Sec
WRITE32: 4411 MB/Sec
WRITE64: 322 MB/Sec

---> L2 <---
READ32: 630 MB/Sec
READ64: 182 MB/Sec
WRITE32: 969 MB/Sec
WRITE64: 321 MB/Sec

---> RAM <---
READ32: 606 MB/Sec
READ64: 181 MB/Sec
WRITE32: 924 MB/Sec
WRITE64: 317 MB/Sec
WRITE: 317 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 7 MB/Sec
WRITE: 158 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 2210.71 MB/sec
2 K Element : 2226.00 MB/sec
4 K Element : 2221.14 MB/sec
8 K Element : 2207.53 MB/sec
16 K Element : 565.15 MB/sec
32 K Element : 456.96 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 192.00 BogoMips



Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends

Duration in seconds: 41.6
Microseconds for one run through Dhrystone: 0.8
Dhrystones per Second: 1201938.1
Dhrystone MIPS (DMIPS) 684



Quicksort:
Elaborating quicksort of 1000 numbers repeated for 10 times
Total time taken by CPU: 4.62


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.001367 0.001367 1212.7
10000 2261 19997 0.001669 0.001758 1155.8
20000 4202 39989 0.003338 0.003515 1170.9
40000 7836 79999 0.006677 0.013672 609.5
80000 14683 160001 0.013353 0.045313 372.1
160000 27607 319993 0.026706 0.114064 298.8
320000 52073 639997 0.053412 0.246873 279.0
640000 98609 1279997 0.106824 0.524998 265.0
1280000 187133 2559989 0.213648 1.112499 252.5
2560000 356243 5119997 0.427296 2.900009 195.4
5120000 679460 10239989 0.854593 6.949997 164.5
10240000 1299068 20479999 1.709186 15.399933 149.7
20480000 2488465 40960001 3.418371 33.999939 136.7

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.006461 (sec)
High MIPS = 1212.7
Low MIPS = 136.7



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 414.5 seconds
C Converted Double Precision Whetstones: 12.1 MIPS


All tests done.

Machine: A1-X1000-40

CPUBench v1.0


RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Freescale P5040 (E5500 core) 1.2 @ 2200 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 64

---> CPU <---
MAX MIPS: 4397

---> L1 <---
READ32: 8183 MB/Sec
READ64: 16342 MB/Sec
WRITE32: 8184 MB/Sec
WRITE64: 16346 MB/Sec

---> L2 <---
READ32: 4660 MB/Sec
READ64: 8378 MB/Sec
WRITE32: 5494 MB/Sec
WRITE64: 9634 MB/Sec

---> RAM <---
READ32: 615 MB/Sec
READ64: 1055 MB/Sec
WRITE32: 1433 MB/Sec
WRITE64: 1438 MB/Sec
WRITE: 2455 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 21 MB/Sec
WRITE: 539 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 4129.14 MB/sec
2 K Element : 4145.09 MB/sec
4 K Element : 4150.98 MB/sec
8 K Element : 4155.51 MB/sec
16 K Element : 3730.50 MB/sec
32 K Element : 3625.49 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 352.00 BogoMips



Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends

Duration in seconds: 23.1
Microseconds for one run through Dhrystone: 0.5
Dhrystones per Second: 2165278.0
Dhrystone MIPS (DMIPS) 1232



Quicksort:
Elaborating quicksort of 1000 numbers repeated for 10 times
Total time taken by CPU: 2.47


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.000781 0.000781 2121.7
10000 2261 19997 0.000954 0.000976 2081.1
20000 4202 39989 0.001908 0.001757 2342.6
40000 7836 79999 0.003816 0.003517 2369.6
80000 14683 160001 0.007632 0.007811 2158.5
160000 27607 319993 0.015264 0.017185 1983.5
320000 52073 639997 0.030528 0.034370 2004.1
640000 98609 1279997 0.061056 0.100021 1391.0
1280000 187133 2559989 0.122111 0.312500 898.8
2560000 356243 5119997 0.244223 1.224976 462.7
5120000 679460 10239989 0.488445 3.999939 285.8
10240000 1299068 20479999 0.976890 9.799805 235.2
20480000 2488465 40960001 1.953781 21.799316 213.2

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.002749 (sec)
High MIPS = 2369.6
Low MIPS = 213.2



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 7.6 seconds
C Converted Double Precision Whetstones: 654.5 MIPS


All tests done.

@Hans

Great thread! My A1222 showed virtually identical results, except for the Whetstone results:


One other point is that the RageMem numbers may be off; the video bandwidth numbers, for example, are absurdly low. I suspect RageMem may be hitting the FPU emulation layer. In which case the numbers both are and are not correct. They would be correct in so far as it shows the impact of floating-point emulation. They would be incorrect in terms of reflecting actual HW bandwidth.

I'll go ahead and and try and verify/disprove that hypothesis shortly.

-- eliyahu

@eliyahu
Quote:No, CPU access to VRAM is extremely slow on all systems. Both the read and write results of the A1222 are about 1/3 of the X5000/40 ones.

@Hans

As a quick test I added RageMem to the LTE blacklist, and wow, was it slow in finishing. That alone tells me that FPU emulation is impacting the numbers; and with LTE disable the results were substantially worse as expected:


So I wonder if we can get Crisot to build us a version using unsigned integers rather than floats, if possible, for an A1222 test. That should be possible since for some CPUs he already does that. This doesn't matter for the other benchmarks, since they are supposed to measure workload performance, and if FPU emulation kills the result, so be it. But for something that should be an indicator of relative HW bandwidth, we should take the emulation issue out of the equation if possible for a more accurate comparison. IMO, of course.

-- eliyahu

@Hans

... and lastly, here are the full results from my X5000/20 system:


Hope this helps!

-- eliyahu

@eliyahu
Quote:Not possible since 32 bit CPUs don't support 64 bit integer loads/stores, 64 bit accesses are only supported for double (and 128 bit accesses for vector on some CPUs).
64 bit C integer loads/stores have to be executed as two independent 32 bit assembler loads/stores, and you have the results for 32 bit integer accesses in the READ32/WRITE32 results already.

A1-X1000


CPUBench v1.0


RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: P.A. Semi PWRficient PA6T-1682M B1 @ 1800 Mhz
Caches Sizes: L1: 64 KB - L2: 2048 KB - L3: none
Cache Line: 64

---> CPU <---
MAX MIPS: 3084

---> L1 <---
READ32: 6851 MB/Sec
READ64: 13610 MB/Sec
WRITE32: 6851 MB/Sec
WRITE64: 13688 MB/Sec

---> L2 <---
READ32: 3273 MB/Sec
READ64: 5007 MB/Sec
WRITE32: 2532 MB/Sec
WRITE64: 4093 MB/Sec

---> RAM <---
READ32: 2842 MB/Sec
READ64: 4006 MB/Sec
WRITE32: 2732 MB/Sec
WRITE64: 3392 MB/Sec
WRITE: 350 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 132 MB/Sec
WRITE: 161 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 1763.99 MB/sec
2 K Element : 1766.06 MB/sec
4 K Element : 1768.30 MB/sec
8 K Element : 1769.26 MB/sec
16 K Element : 1753.39 MB/sec
32 K Element : 1532.57 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 192.00 BogoMips


Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends

Duration in seconds: 38.8
Microseconds for one run through Dhrystone: 0.8
Dhrystones per Second: 1289483.0
Dhrystone MIPS (DMIPS) 733



Quicksort:
Elaborating quicksort of 1000 numbers repeated for 10 times
Total time taken by CPU: 4.09


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.001163 0.001163 1425.0
10000 2261 19997 0.001420 0.001469 1383.4
20000 4202 39989 0.002841 0.002747 1498.7
40000 7836 79999 0.005682 0.005875 1418.5
80000 14683 160001 0.011364 0.012360 1364.1
160000 27607 319993 0.022727 0.031433 1084.4
320000 52073 639997 0.045454 0.065308 1054.7
640000 98609 1279997 0.090908 0.137939 1008.6
1280000 187133 2559989 0.181816 0.275879 1018.1
2560000 356243 5119997 0.363633 0.576172 983.7
5120000 679460 10239989 0.727265 1.591797 718.2
10240000 1299068 20479999 1.454531 3.710938 621.2
20480000 2488465 40960001 2.909062 8.007812 580.3

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.001795 (sec)
High MIPS = 1498.7
Low MIPS = 580.3



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 8.2 seconds
C Converted Double Precision Whetstones: 611.6 MIPS


All tests done.

X5000/20:

RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Freescale P5020 (E5500 core) 1.2 @ 1995 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 64

---> CPU <---
MAX MIPS: 3988

---> L1 <---
READ32: 7532 MB/Sec
READ64: 15041 MB/Sec
WRITE32: 7533 MB/Sec
WRITE64: 15045 MB/Sec

---> L2 <---
READ32: 4286 MB/Sec
READ64: 7708 MB/Sec
WRITE32: 5020 MB/Sec
WRITE64: 8813 MB/Sec

---> RAM <---
READ32: 732 MB/Sec
READ64: 1341 MB/Sec
WRITE32: 1748 MB/Sec
WRITE64: 1754 MB/Sec
WRITE: 2320 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 22 MB/Sec
WRITE: 535 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 3720.51 MB/sec
2 K Element : 3696.03 MB/sec
4 K Element : 3682.75 MB/sec
8 K Element : 3761.76 MB/sec
16 K Element : 3316.98 MB/sec
32 K Element : 3272.27 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 320.00 BogoMips



Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends

Final values of the variables used in the benchmark:

Int_Glob: 5
should be: 5
Bool_Glob: 1
should be: 1
Ch_1_Glob: A
should be: A
Ch_2_Glob: B
should be: B
Arr_1_Glob[8]: 7
should be: 7
Arr_2_Glob[8][7]: 50000010
should be: Number_Of_Runs + 10
Ptr_Glob->
Ptr_Comp: 1405485704
should be: (implementation-dependent)
Discr: 0
should be: 0
Enum_Comp: 2
should be: 2
Int_Comp: 17
should be: 17
Str_Comp: DHRYSTONE PROGRAM, SOME STRING
should be: DHRYSTONE PROGRAM, SOME STRING
Next_Ptr_Glob->
Ptr_Comp: 1405485704
should be: (implementation-dependent), same as above
Discr: 0
should be: 0
Enum_Comp: 1
should be: 1
Int_Comp: 18
should be: 18
Str_Comp: DHRYSTONE PROGRAM, SOME STRING
should be: DHRYSTONE PROGRAM, SOME STRING
Int_1_Loc: 5
should be: 5
Int_2_Loc: 13
should be: 13
Int_3_Loc: 7
should be: 7
Enum_Loc: 1
should be: 1
Str_1_Loc: DHRYSTONE PROGRAM, 1'ST STRING
should be: DHRYSTONE PROGRAM, 1'ST STRING
Str_2_Loc: DHRYSTONE PROGRAM, 2'ND STRING
should be: DHRYSTONE PROGRAM, 2'ND STRING

Duration in seconds: 26.0
Microseconds for one run through Dhrystone: 0.5
Dhrystones per Second: 1923210.4
Dhrystone MIPS (DMIPS) 1094



Quicksort:
Elaborating quicksort of 1000 numbers repeated for 10 times
Unsorted array:
<clip>
Sorted array:
<clip>
Total time taken by CPU: 2.76


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.000877 0.000877 1889.7
10000 2261 19997 0.001071 0.001068 1902.2
20000 4202 39989 0.002142 0.001945 2115.8
40000 7836 79999 0.004285 0.004730 1761.7
80000 14683 160001 0.008569 0.009308 1811.3
160000 27607 319993 0.017138 0.018921 1801.6
320000 52073 639997 0.034277 0.040283 1710.0
640000 98609 1279997 0.068554 0.100098 1389.9
1280000 187133 2559989 0.137107 0.275879 1018.1
2560000 356243 5119997 0.274215 1.098633 515.9
5120000 679460 10239989 0.548430 3.554688 321.6
10240000 1299068 20479999 1.096859 8.789062 262.3
20480000 2488465 40960001 2.193719 19.804688 234.6

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.002590 (sec)
High MIPS = 2115.8
Low MIPS = 234.6



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 8.5 seconds
C Converted Double Precision Whetstones: 590.7 MIPS


All tests done.

@Hans

System: QEMU 8.0.0 Windows 10 x64 Ryzen 5600

@Hans

SAM460ex

@derfs

System: WinUAE Windows x64 Ryzen 5600

SAM440Flex 800MHz

CPUBench v1.0


RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: AMCC PPC440EP 1.3 @ 799 Mhz
Caches Sizes: L1: 32 KB - L2: none - L3: none
Cache Line: 32

---> CPU <---
MAX MIPS: 1599

---> L1 <---
READ32: 3023 MB/Sec
READ64: 6027 MB/Sec
WRITE32: 3024 MB/Sec
WRITE64: 6026 MB/Sec

---> RAM <---
READ32: 320 MB/Sec
READ64: 320 MB/Sec
WRITE32: 200 MB/Sec
WRITE64: 200 MB/Sec
WRITE: 887 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 27 MB/Sec
WRITE: 39 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 1185.90 MB/sec
2 K Element : 1183.31 MB/sec
4 K Element : 1183.96 MB/sec
8 K Element : 1171.92 MB/sec
16 K Element : 375.57 MB/sec
32 K Element : 298.67 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 128.00 BogoMips



Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends



Duration in seconds: 71.9
Microseconds for one run through Dhrystone: 1.4
Dhrystones per Second: 695362.4
Dhrystone MIPS (DMIPS) 395



Quicksort:

Total time taken by CPU: 7.68


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.002708 0.002708 612.1
10000 2261 19997 0.003307 0.003433 591.8
20000 4202 39989 0.006613 0.006866 599.5
40000 7836 79999 0.013226 0.023804 350.1
80000 14683 160001 0.026453 0.117188 143.9
160000 27607 319993 0.052906 0.284424 119.8
320000 52073 639997 0.105811 0.662842 103.9
640000 98609 1279997 0.211622 1.425781 97.6
1280000 187133 2559989 0.423245 3.173828 88.5
2560000 356243 5119997 0.846489 7.001953 80.9
5120000 679460 10239989 1.692979 14.960938 76.4
10240000 1299068 20479999 3.385957 31.406250 73.4
20480000 2488465 40960001 6.771914 68.437500 67.9

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.014933 (sec)
High MIPS = 612.1
Low MIPS = 67.9



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 72.5 seconds
C Converted Double Precision Whetstones: 68.9 MIPS

@Hans

First of all, thanks for the great compilation and the script which works very well.

Here now also my result under Qemu, the hardware they please take from my signature.

CPUBench v1.0


RageMem:

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Motorola MPC 7447/7457 Apollo 1.2 @ 1533 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 32

---> CPU <---
MAX MIPS: 9845

---> L1 <---
READ32: 5848 MB/Sec
READ64: 8829 MB/Sec
WRITE32: 6469 MB/Sec
WRITE64: 10205 MB/Sec

---> L2 <---
READ32: 5812 MB/Sec
READ64: 8822 MB/Sec
WRITE32: 6452 MB/Sec
WRITE64: 10234 MB/Sec

---> RAM <---
READ32: 5172 MB/Sec
READ64: 7506 MB/Sec
WRITE32: 5408 MB/Sec
WRITE64: 10191 MB/Sec
WRITE: 2943 MB/Sec (Tricky)

---> VIDEO BUS <---
READ: 8767 MB/Sec
WRITE: 10102 MB/Sec



SortBench:
-------------------------------------------------------------
SORTBENCH 1.1 (Gunnar von Boehn)
Its a CPU benchmark that stresses CPU, DCache and branch prediction.
-------------------------------------------------------------
1 K Element : 3091.48 MB/sec
2 K Element : 3103.53 MB/sec
4 K Element : 3110.91 MB/sec
8 K Element : 3102.89 MB/sec
16 K Element : 3110.96 MB/sec
32 K Element : 3109.41 MB/sec


BogoMIPS:
Calibrating delay loop..

Ok - 656.00 BogoMips



Dhrystone:
Dhrystone Benchmark, Version 2.1 (Language: C)
Program compiled without 'register' attribute

Execution starts, 50000000 runs through Dhrystone
Execution ends

Duration in seconds: 28.1
Microseconds for one run through Dhrystone: 0.6
Dhrystones per Second: 1778409.9
Dhrystone MIPS (DMIPS) 1012



Quicksort:
Elaborating quicksort of 1000 numbers repeated for 10 times
Unsorted array:

Total time taken by CPU: 2.44


Sieve:

Sieve of Eratosthenes (Scaled to 10 Iterations)
Version 1.2, 03 April 1992

Array Size Number Last Prime Linear RunTime MIPS
(Bytes) of Primes Time(sec) (Sec)
8191 1899 16381 0.000684 0.000684 2425.1
10000 2261 19997 0.000835 0.000879 2311.8
20000 4202 39989 0.001669 0.001758 2341.8
40000 7836 79999 0.003339 0.003515 2370.4
80000 14683 160001 0.006677 0.007813 2157.9
160000 27607 319993 0.013354 0.015624 2181.8
320000 52073 639997 0.026709 0.031252 2204.1
640000 98609 1279997 0.053418 0.056248 2473.5
1280000 187133 2559989 0.106836 0.125008 2246.8
2560000 356243 5119997 0.213671 0.250015 2267.0
5120000 679460 10239989 0.427343 0.499954 2286.8
10240000 1299068 20479999 0.854686 1.399994 1646.7
20480000 2488465 40960001 1.709372 4.400024 1056.1

Relative to 10 Iterations and the 8191 Array Size:
Average RunTime = 0.000865 (sec)
High MIPS = 2473.5
Low MIPS = 1056.1



Whetstone:

Please wait...

Loops: 50000 Iterations: 1 Duration: 9.6 seconds
C Converted Double Precision Whetstones: 523.2 MIPS


All tests done.

Now can I please get a 32 bit driver for AmigaOs4.1 so we can move on to the GPU benchmark ? Don't take it seriously it's just a joke...

Thanks again for the script and merging the CPU benchmarks.

I am somewhat pleasantly surprised that these results are so different.....es shows once again how strong this ARM Cpu is, but also shows its weaknesses in some areas.

---> VIDEOBUS <---
READ: 8767 MB/S.
WRITE: 10102 MB/sec.

That impressed me the most, how does that come about, maybe it's because the main memory as well as the garficard memory are used as shared memory.

With the benchmark you can imagine how well AmigaOs4.1 with Qemu runs on my machine. That is the point why I chose this emulation, it gives me a good feeling to be able to use AmigaOs4.1 under it. AmigaOs4.1 runs albeit somewhat limited as my 2nd operating system thanks to Qemu.

Numbers can be manipulated, with videos it becomes very difficult, if they would like I can also create a video about it again.

@eliyahu

Yes, the FPU emulation gets in the way of some of the copy benchmarks. There's clearly something wrong on my system with the Whetstone benchmark. Maybe I haven't kept my system up-to-date enough...


@joerg
Quote:
You're clearly unfamiliar with the unusual P1022 CPU in the A1222. It doesn't have a proper PowerPC FPU, but does have 64-bit GPRs which can be used via its Stream-Processing-Engine (SPE). The RadeonHD & RX drivers both have SPE optimized copy routines in them.

Hans

@Maijestro

Quote:
Yes, the results are interesting. I am impressed with the ARM numbers. Derfs' Ryzen numbers are also significantly better than my laptop.

@all
Has anyone tried Qemu on a Raspberry Pi 4 or a PineBook Pro? I'd be interested to see what a weaker ARM CPU can do (one that doesn't have Apple's M1 extensions to aid with x86 emulation).

Hans

@Maijestro
Quote:Because you are testing mainboard RAM accesses, the part of it used for the SM501 emulation, not gfx card RAM accesses.
The RageMem L1/L2 cache results compared to RAM ones are of course wrong as well, your host CPU has much larger L1 and L2 caches than the emulated 32/512 KB.

Unlike QEmu UAE seems to support direct access to the gfx card, for example compare derfs QEmu results (11034/5330 MB/s) with the UAE (51/49 MB/s) ones.

Quote:To get usable gfx speed you need an AmigaOS gfx driver for QEmu, just like the uaegfx driver used in UAE.
Best way for implementing a QEmu gfx driver would probably be one using Cairo and/or OpenGL, that way it would work with any host gfx card.
The currently used SM501/2 emulation is the worst possible way, because of it's strange 4 byte 24 bit modes neither supported by AmigaOS nor host gfx cards, and even emulating a Permedia 2 or a Voodoo 3/5 would be much better than that.

@Hans
I tried the Qemu 8 under Raspberry Pi 400 some time ago and the feeling I had was that it was slower than my classic PPC A1200. But I will try to run it again and get results based on your script.