| 68K=Rules / PPC =Shit | page 1 2 3 4 5 6 7
|
|---|
|
|---|
| | 
Saladriel Amrael
Posts 166
18 Sep 2017 11:32
| Gunnar, it's all cool and clear after your explanation.
But can I suggesto you one thing? If you tell all the story since the first post you make it all clear since the beginning and everybody can read the results with the proper mindset since the first moment.
That would be a better way to communicate (IMHO).By the way, these results look quite impressive, and would be interesting to also have a comparison with some WinUAE users, both JIT and non JIT.
| |
| | 
Anthony Jacques
Posts 11
18 Sep 2017 11:37
| Markus Horbach wrote:
|
Nice results, but as always, I am a bit sceptical if the term "world record" is used. I always get the feeling the marketing department is talking. So I did the math for the results.
|
100% agree. Mis-representing benchmarking results isn't doing you any favours. Having been lurking for a while, I've noticed this as a trend. Gunnar von Boehn wrote:
|
But we have to mind one thing here.
BUSTEST is the de facto standard memory benchmark.
Many AMIGA accelerators were advertised using this benchmark.
So we run the standard AMIGA tool to measure memory operations here.
This tool is an 68k program - as the vast majority of AMIGA programs are of course 68k programs.
|
I notice however that this isn't the same approach you use when it comes to features that are improved by using AMMX. Eg, in "Is Vampire Faster Than Classic PPC Cards?" CLICK HERE ) presumably the answer is "no" unless using the new AMMX optimised software. How does it perform using a classic, non 68080 optimised build of the software? Picking and choosing how you prefer to benchmark based upon how to make the Vampire look the best in the area you've just improved isn't helpful, especially if you're not making it clear when this is done. Perhaps it would be best to leave benchmarking to someone independent of the team? Or take a more rigourous approach to benchmarking as a whole, presenting all the facts, and accepting constructive critism of your testing methodology? Let your product speak for itself in fair tests. I'm sure it'll still look good. Unfair / incomplete / misleading / inconsistent tests however are making you look bad.
| |
| | 
Gunnar von Boehn
(Apollo Team Member)
Posts 6233
18 Sep 2017 11:46
| thellier alain wrote:
|
Buut if you run (say) Lightwave 68k on an NG Amiga it will not use only the 68k emulation but also call the OS .libraries that are PPC and that use the NG hardware so the average speed will be better that if all was done 100% inside the 68k emulation (with the slow mem access that you describe well)
Alain Thellier
|
Yes, ALAIN you are of course 100% correct here.
The POWERPC has the option to execute some programs or OS libraries in native PPC code. But how much CPU power has the PowerPC.
As you know MegaHerz is misleading. Lets have some fun and make a proper comparison. In the BLUE corner
The POWERPC 750 FX, a marvel of IBM engineering. In the RED corner
the 68080 a modern, Super Scalar, 64Bit CISC CPU. The POWERPC is a RISC CPU and by design can not do Memory Operations as part of a real instruction. Therefore the RISC CPU needs to use several instruction for common CISC instructions. For a ADD (mem),Reg Operation - the PPC needs 2 instruction For a ADD Reg,(mem) Operation - the PPC needs 3 instruction So by design the PowerPC needs more instructions than the 68K to do the same work. Round1) 2:1 FOR APOLLO Lets look at the CPU units. The IBM PPC 750-FX has two 32bit ALUs.
It can at maximum do 2 operations per cycle.
It has no own unit to do Memory address calculations
but will use an block an ALU when doing a memory operation. The APOLLO Core is a full 64bit machine.
APOLLO has two full 64bit ALUs.
And APOLLU has in addition to this 2 extra EA UNITs.
APOLLO can do memory address calculations for free. So 2 units versus 4 Units Round2) 2:1 FOR APOLLO If you want to process data, like copying GFX data, rendering sprites etc. then the width of the ALU and the width of the registers define how much work you can do. IBM 750
32bit registers
2x 32Bit ALU APOLLO 68080
64bit registers
2x 64BIT ALU Round3) 2:1 FOR APOLLO Now if you want to process media data - like Videos for example.
Then this boils down to processing lots of 8bit or 16bit calculations. APOLLO has AMMX for this.
And can do 4 x 16bit operations in 1 ALU.
So 4 times as many operations as the PowerPC per cycle. Round4) 4:1 FOR APOLLO As you can easily see,
the 68080 can do in average a lot more work than the PowerPC per cycle. And for case where AMMX can be used the 68080 @80 MHz
has in fact very similar computing power as the PowerPC 750@800 MHz With proper usage of AMMX you can bring the 68080 on the same computing level as a 600-800 MHz POWERPC G3.
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6233
18 Sep 2017 12:00
| Gentlemen,
If you want to compare other systems.
Then please mind to use the correct parameters
BUSTEST FAST size=8M
And I would suggest that we in this thread focus on PPC and 68K.
I doubt anyone has a working PI emulator running, but It would surprise me if the PI can even compete with PPCs.
I would propose to leave the x86 comparison for another thread
| |
| | 
Vojin Vidanovic
Posts 770
18 Sep 2017 13:05
| Gunnar von Boehn wrote:
|
And I would suggest that we in this thread focus on PPC and 68K. |
A1-x1000 CPU PA Semi PA6T-1682M 1800Mhz, 2MB L2, 64K L1 CACHE
Total RAM: 2 GiB External Bus (FSB) Speed: 900 MHz
Radeon HD 7700 2GB DDR5
PCIGraphics.card 53.15
RadeonHD.chip 2.21
Exec.library version: 53.89
Graphics.library version: 54.226
Intuition.library version: 54.26
Picasso96API.library version: 54.18
Rtg.library version: 54.90
Gfx2Bench results EXTERNAL LINK
------------------------------------------------------------
STREAM version $Revision: 5.7 $
-------------------------------------------------------------Function Rate (MB/s) Avg time Min time Max time
Copy: 4090.5074 0.0079 0.0078 0.0081
Scale: 3795.5355 0.0087 0.0084 0.0090
Add: 3865.6437 0.0137 0.0124 0.0176
Triad: 3786.9682 0.0130 0.0127 0.0133
.RAM Disk:> bustest size=8m fast
BusSpeedTest 0.19 (mlelstv) Buffer: 8388608 Bytes, Alignment: 32768
========================================================================
memtype addr op cycle calib bandwidth
fast $5E298000 readw 13.1 ns normal 152.4 * 10^6 byte/s
fast $5E298000 readl 14.0 ns normal 285.6 * 10^6 byte/s
fast $5E298000 readm 17.9 ns normal 223.4 * 10^6 byte/s
fast $5E298000 writew 12.7 ns normal 157.0 * 10^6 byte/s
fast $5E298000 writel 14.1 ns normal 283.7 * 10^6 byte/s
fast $5E298000 writem 17.2 ns normal 232.2 * 10^6 byte/sCongrats on Vamp v2 results, hoping to see v4 a bit faster with interleaved DDR3 outlassing this Kingston DDR2
| |
| | 
Sean Sk
Posts 488
18 Sep 2017 14:10
| Steve Ferrell wrote:
| Why do you keep referring to the Vampire as "emulation"? The Vampire is no more an emulator than an EEPROM is an emulator.
|
Steve, he wasn't. Samuel was referring to 68k emulation on Amiga PPC computers.
| |
| |
Vojin Vidanovic
Posts 770
18 Sep 2017 14:18
| sean sk wrote:
|
Steve Ferrell wrote:
| Why do you keep referring to the Vampire as "emulation"? The Vampire is no more an emulator than an EEPROM is an emulator.
|
Steve, he wasn't. Samuel was referring to 68k emulation on Amiga PPC computers.
|
Aka Petunia, which I believe emulates 020 no FPU or so.
Even its really fast on PA Semi, it doesnt help memory performance.
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6233
18 Sep 2017 14:18
| Thanks Vojin Your result is included in the table
| |
| |
Vojin Vidanovic
Posts 770
18 Sep 2017 14:33
| Gunnar von Boehn wrote:
|
Thanks Vojin
Your result is included in the table
|
Reason to turn machine on. Dream of having AmigaOS enjoyable, but with decent browser and office is still unfinished. Little of hardwares power is used, Radeon HD 7700 performs poor even in 2D with most recent video and enhancer v1.2 drivers. Only refesh is video player Emotion, and that and Odyssey is worthy of backporting.
Hey, I have best PPC result! That US Army secret weapon runner and G5 Class Laptop Power Efficient CPU has some magic! Until some MOS on G5 squashes it :-)
Still monolith cant do the work even it used to do in Linux where more of SSD/dual core/fast RAM power is used because SI card has no proper video drivers on PPC32 Linux ... sigh.
But x1000 case was used in recent haircut commercial, to hide that new x5000 case is cheaper and uglier. And its presented as x5000. So now monolith case prostitutes.
Now, waiting for Standalone v4, keep on the good work on cores!
| |
| | 
Hbarra 2Pi
Posts 8
18 Sep 2017 15:28
| Bravo Vojin. Totally agree with your arguments.
I'm looking forward to the V4 as soon as possible.
This project is more than exciting.
And hopefully thousands of future Vampire AMIGA users will know about this feeling. A new real begining.
| |
| | 
Kresimir Lukin
Posts 65
18 Sep 2017 18:00
| I agree as well.
Also I don't see point in buying x5000, Amiga OS4 lacks software, PPC Linux is behind x86-64 Linux. Classic Amiga software runs as well and maybe better on UAE on PC.
Aros runs great on x86 if you like NG Amiga.Price of x5000 is another point where x86-64 is better option.
Only this project can return Amiga back in life, together with Aros, as that is only one developing Amiga 68k OS at this moment.
| |
| | 
Steve Ferrell
Posts 424
18 Sep 2017 18:28
| ExiE CZEX wrote:
|
Steve Ferrell wrote:
|
And the benchmarks aren't marketing hype. As I stated earlier, for those who are considering the purchase of an OS4 system versus an OS3 Vampire or other 68K acclerator, these benchmarks are invaluable.
|
Of course it is part of the marketing. And these synthetic benchmarks are hardy invaluable. Don't get me wrong, it is really nice that Vamp has so fast mem access, but it says nothing about real power of the machine.
Do you really believe people make their decisions on one test and not thinking about real power, available software, future possibilities etc.?
Steve Ferrell wrote:
|
They're all very well aware of UAE and other emulation solutions. Comparisons outside the scope above are pointless as no one here is looking for a Pi/G5/PC/UAE solution.
|
So why compare Vamp with MOS running on Mac hardware?
|
I think English is getting in the way. In English, "invaluable" means "very valuable".
And what's wrong with marketing? Gunnar IS in the business of selling the Vampire and the Apollo core? Should he just stop all marketing and not post any info or benchmarks? That would be ludicrous. And I don't see any other Amiga hardware manufacturers posting benchmarks at all. Why is that? Are they afraid of being exposed for selling vastly overpriced and under performing hardware?
And I disagree, his benchmarks say quite a lot about the power of the Vampire, if you choose not to see it then you're just being obstinate.
And no, I don't think people make their decisions on just one test. For me, this test and other benchmarks posted here are very helpful and it has helped me make a final decision against buying any more PPC hardware. I can't justify the cost of a PPC system when it's outclassed by an FPGA.
Why would I want to run MOS on Mac hardware when I can use my existing 68K software on a Vampire which runs circles around that Mac hardware? Maybe because I want the most bang for my buck and have a lot invested in classic software.
| |
| |
Anthony Jacques
Posts 11
18 Sep 2017 21:43
| Steve Ferrell wrote:
|
And what's wrong with marketing? Gunnar IS in the business of selling the Vampire and the Apollo core? Should he just stop all marketing and not post any info or benchmarks? That would be ludicrous. |
That is actually the reason why he should be being more careful. Claims like these on a commercial website can count as advertising and therefore legally should not be misleading. There are rules about comparative claims such as these, that the advertiser is usually much much more careful with the wording of claims that is is "the best" or when comparing to other products.
I'm sure that in reality nobody is going to complain to the ASA or other equivalent bodies. Also here it is more just the lack of caveats and detail that are the issues, but that doesn't mean it is OK to repeatedly post such bold claims with misleading details which are only highlighted when queried.
As has been said before, post honest, transparent benchmarks without the unnecessary digs, from what I've seen, it's good enough that it's unnecessary and just gives a bad impression to those who see the flaws in the tests.
| |
| |
Steve Ferrell
Posts 424
18 Sep 2017 22:32
| Anthony Jacques wrote:
|
I'm sure that in reality nobody is going to complain to the ASA or other equivalent bodies. Also here it is more just the lack of caveats and detail that are the issues, but that doesn't mean it is OK to repeatedly post such bold claims with misleading details which are only highlighted when queried.
|
Yes, but it seems as those who are complaining most loudly have financial interests in A-EON, and/or Hyperion. It's interesting that they want to point out all the "flaws" in Gunnar's benchmarks yet it's nearly impossible to find benchmarking data on A-Eon's PPC offerings. At least Gunnar is posting claims. What little data I can find on the PPC chips used in the latest OS4 systems puts their performance on par with PC's from the early 2000's.
You won't see A-Eon posting any official benchmarks for the A1222 or the X5000 anytime soon because the data would certainly impact their sales in a negative way. The only data I can even find for the X1000 was posted by users who were disappointed by its performance.
| |
| | 
Mo Retro
Posts 241
18 Sep 2017 22:43
| I don't get it at all about that negativism.
If they get no info then they complain. If they get info then they look for arguments to discredit it.
Sorry but a lot of energy gets wasted way, energy that the Team can use positively on getting on with the project.
Come-on guys let's move on and give the kids a break.
| |
| | 
John William
Posts 566
18 Sep 2017 22:52
| Mo Retro wrote:
|
I don't get it at all about that negativism.
If they get no info then they complain. If they get info then they look for arguments to discredit it.
Sorry but a lot of energy gets wasted way, energy that the Team can use positively on getting on with the project.
Come-on guys let's move on and give the kids a break.
|
The Amiga community flourish in negativity, trolling, war and destroying anything that could benefit them. We have the self destruct program executed.
I bet you if Commodore was still alive and making new custom Amiga and continued Amiga there will be 60% of the Amiga's community complain, bash and attempt to destroy or jeopardize any potential of advancement and will stick with OCS kickstart 1.3 512 KB CHIP RAM. Any attempt of advancing ahead will result in why? we already have a PCs? We can just use an emulator? on and on.
| |
| |
Sean Sk
Posts 488
19 Sep 2017 00:15
| Anthony Jacques wrote:
|
post honest, transparent benchmarks without the unnecessary digs, from what I've seen, it's good enough that it's unnecessary and just gives a bad impression to those who see the flaws in the tests.
|
As indicated earlier by Gunnar, this is being done in fun. If there are people here taking this all too seriously then they need to step away from the computer for awhile, take a deep breath, clear their minds, step outside and spend some time with family and friends. I don't get why people get so intense over this.
I'm sure this hobby is populated by middle-aged men going through mid-life crises'. :)
| |
| |
John William
Posts 566
19 Sep 2017 00:32
| sean sk wrote:
|
Anthony Jacques wrote:
|
post honest, transparent benchmarks without the unnecessary digs, from what I've seen, it's good enough that it's unnecessary and just gives a bad impression to those who see the flaws in the tests.
|
As indicated earlier by Gunnar, this is being done in fun. If there are people here taking this all too seriously then they need to step away from the computer for awhile, take a deep breath, clear their minds, step outside and spend some time with family and friends. I don't get why people get so intense over this.
I'm sure this hobby is populated by middle-aged men going through mid-life crises'. :)
|
HEY!!! -mumbles quietly- I am middle aged man and I am fine about my age!!
| |
| | 
Samuel Devulder
Posts 248
19 Sep 2017 00:48
| sean sk wrote:
|
As indicated earlier by Gunnar, this is being done in fun.
|
Oh! if it was kind of a joke, then there's nothing to complain about. Sure!
But I wonder if all this was more to exemplify Vampire-power or to bash amiga-PPC for their lack of it. Better concentrate on vampire results with impressive h264 decoding in Riva EXTERNAL LINK (or MPEG if h264 is not yet ready. Oh? already done. I see... EXTERNAL LINK ). This might require a hard-fpu though *sigh*
| |
| |
John William
Posts 566
19 Sep 2017 01:11
| Samuel Devulder wrote:
|
sean sk wrote:
|
As indicated earlier by Gunnar, this is being done in fun.
|
Oh! if it was kind of a joke, then there's nothing to complain about. Sure!
But I wonder if all this was more to exemplify Vampire-power or to bash amiga-PPC for their lack of it. Better concentrate on vampire results with impressive h264 decoding in Riva EXTERNAL LINK (or MPEG if h264 is not yet ready. Oh? already done. I see... EXTERNAL LINK ). This might require a hard-fpu though *sigh*
|
You are wrong. There is never need for FPU to watch movies.
| |
|
Performance and Benchmark Results!