|
---|
| | Gunnar von Boehn (Apollo Team Member) Posts 6254 23 Apr 2016 20:27
| Last month the max performance of the Vampire was 140 PTS With continuous tuning on the of the APOLLO core, we were able to improve this by 33 points. The new maximum performance is 173 PTS. CARD Total-Score ---------------------------------------------------------------------------- ACA620 68020 @16 4 * FPGA ARCADE TG68 @ 9 ** ACA1231 68030 @41 11 *** Blizzard 68030 @50 11 *** AMIGA4000 68040 @25 13 *** AMIGA4000 68040 @40 25 ****** MK2 68060 @66 65 **************** VAMPIRE2 Silver4-Pre 173 *******************************************
Detailed Benchmark result: CLICK HERE
| |
| | Roger Andre Lassen
Posts 150 23 Apr 2016 21:37
| Fantastic :-)
| |
| | Gregthe Canuck
Posts 274 23 Apr 2016 23:36
| Excellent! That is more than 20% improvement in one release.
| |
| | Mr-Z EdgeOfPanic
Posts 189 24 Apr 2016 06:35
| Good work!
| |
| | Ulf Andersson
Posts 29 24 Apr 2016 15:18
| That is REALLY cool! .. Apollo Core is the BEST! .. Keep up the good work! :)
| |
| | John Heritage
Posts 112 25 Apr 2016 19:17
| 1 point = 68000 @ ~7.1 mhz?
| |
| | Gunnar von Boehn (Apollo Team Member) Posts 6254 25 Apr 2016 19:43
| John Heritage wrote:
| 1 point = 68000 @ ~7.1 mhz? |
Minibench does a number of different tests. - There are number of test measuring the raw CPU calculation speed. E.g test which count the MIPS of different instructions. - There are tests measuring the speed of different memory accessing EA-modes - There are tests measuring the speed/overhead of workloops - There are test measuring the speed/overhead of calling subroutines - There are test measuring the benefit/effectiveness of your branch prediction. - There are test measuring the speed of your main memory and of your caches. The whole minibench does 107 individual tests. The score is the calculated average. About 10 tests use instructions which the 68000 does not support. Therefore the benchmark needs 68020 or higher to run. I assume an 68000 will score between 0.5 and 1.0
| |
| | Kaosmaster RJ
Posts 3 26 Apr 2016 11:19
| Hi Gunner, "theoretically" if it would be possible to compare these performance against a PPC amiga CPU, in your mind what score it would perform more or less a PPC CPU ? How far is a Vampire from a PPC... thanks, Franco
| |
| | Mr-Z EdgeOfPanic
Posts 189 26 Apr 2016 18:08
| kaosmaster RJ wrote:
| Hi Gunner, "theoretically" if it would be possible to compare these performance against a PPC amiga CPU, in your mind what score it would perform more or less a PPC CPU ? How far is a Vampire from a PPC... thanks, Franco
|
http://amiga.resource.cx/perf/sysspeed.pl?id=cyberstormppc#cyberstormppc According to the benchmarks above a 604e 233Mhz does just under 317 MIPS. Vampire X15 turbo does 131 MIPS. Concluding from that vampire will be able to beat a 604e ppc when the Altera Arria 10 will be used.
| |
| | Gunnar von Boehn (Apollo Team Member) Posts 6254 26 Apr 2016 18:20
| kaosmaster RJ wrote:
| Hi Gunner, "theoretically" if it would be possible to compare these performance against a PPC amiga CPU, in your mind what score it would perform more or less a PPC CPU ? |
Hi Franko, There are many different PPC. 603/604/5002/750/440/460/... 1) Most PPC can per clock max do 2 instructions. APOLLO can per clock do up to 4 instructions. 2) As you know, 68K instructions are typically stronger than PPC instructions. There is sometimes 1=1 ratio But also often 1=3 ratio This means PPC needs often 3 instruction to do the same work as 1 68K-instruction. Sometimes PPC even needs a lot more than 3 instruction to do the same as 1 68K. 3) All the Above PPC are 32bit only. So they can at most do 2x32bit. While APOLLO is 64bit. As you can see, APOLLO can do per clock a lot more processing. 4) APOLLO can do a number of advanced features - Like instruction fusing. Which none of the PPC can do. 5) APOLLO has more advanced memory prefetcher and better memory controller = better memory performance. kaosmaster RJ wrote:
| How far is a Vampire from a PPC... thanks, Franco |
You can assume that Vampire is in the sum already faster than PowerUp PPC. Would be interesting the run minibench on PPC and then to compare.
| |
| | Gunnar von Boehn (Apollo Team Member) Posts 6254 26 Apr 2016 18:32
| Mr-Z EdgeOfPanic wrote:
| kaosmaster RJ wrote:
| Hi Gunner, "theoretically" if it would be possible to compare these performance against a PPC amiga CPU, in your mind what score it would perform more or less a PPC CPU ? How far is a Vampire from a PPC... thanks, Franco |
http://amiga.resource.cx/perf/sysspeed.pl?id=cyberstormppc#cyberstormppc According to the benchmarks above a 604e 233Mhz does just under 317 MIPS. Vampire X15 turbo does 131 MIPS. |
No No, you misunderstand what you see!The Vampire today reaches an average of 180 points. The PPC you quote reaches in 1 test a peak of 317. The Peak of the Vampire is today already over 400 in some tests. The are less complex and more complex tests. The PowerPC will also in many test score below 100. Mind that Minibench does not just print the highest number. Minibench does over 100 test - also very complex and slow test like doing DIVIDES. Minibench averages all the scores. Mr-Z EdgeOfPanic wrote:
| Concluding from that vampire will be able to beat a 604e ppc when the Altera Arria 10 will be used. |
For a fair comparision you need to run all 100 test also on PPC And then calc the average - and then compare. Probably VAMPIRE 2 - is already faster than 603 PowerPC We should measure against 604 - would be interesting to see if Vampire beat it today also. The Apollo in ARRIA should performance wise be combative to a G3 PowerPC system.
| |
| | Mr-Z EdgeOfPanic
Posts 189 27 Apr 2016 06:26
| Thanks for the explanation Gunnar, I already had some doubts about that 604e benchmark. It would be very interesting indeed to see how Vampire in it's current state compares to 604e ppc. So let's run a minibench test :)
| |
|