| Happy Birthday Arne! | page 1 2 3 4 5
|
|---|
|
|---|
Kyle Osborne
USA
| |
Posts 12
06 Jun 2011 23:50
| Asaf Ayoub wrote:
|
@Marcel
So 2 '060s can saturate all DDR memory ?
Then I started thinking what whould one pay for a triple processor system ?
|
Amiga OS isn't able to use multiple CPUs, as has been discussed in other threads. I don't know (kind of doubt) if it'd even be feasible to modify AROS to use them. Linux could, but I don't think you'd be buying a Natami to run Linux on.
| |
Jakob Eriksson
Sweden
| | (Moderator)
Posts 1097
07 Jun 2011 01:06
| You don't need any or much OS support to use an extra processor. You would use it as a co-processor, not to run OS processes on.
| |
Joe M
Norway
| |
Posts 500
07 Jun 2011 08:53
| Happy birthday to your brother, Annika! (Yeah, I know I'm a little late). I assume the sketch to the beautiful cake was designed in Deluxe Paint? ;) I hope the batch of Natami developer boards will be ready for Natami members soon so we can get extensive testing from several sources.
| |
Andrew / Grg
Poland
| |
Posts 15
07 Jun 2011 10:29
| | Im sure the team still needs more in depth testing with Bubble Bobble! |
I've made some quick test of executing 18 bubbles on emulated a500
EXTERNAL LINK
:)
| |
Ami_ Stuff
Poland
| |
Posts 6
07 Jun 2011 14:11
| It's a shame that this 133MHz thing have no FPU :(
| |
Matt Hey
USA
| |
Posts 737
07 Jun 2011 14:32
| ami_ stuff wrote:
|
It's a shame that this 133MHz thing have no FPU :(
|
Yea, for FFmpeg the N68050 should be able to outperform this Chinese 68060 by having a FPU, 64 bit multiply, and fast bit field instructions. I think 20+ fps will be possible for FFplay optimized for the N68050. I'll still take a Natami with processor board and a socket for a (Rev 6) 68060 for now though ;).
| |
Gunnar von Boehn
Germany
| | (Moderator)
Posts 5775
07 Jun 2011 15:00
| One small update :-D With tuned memsettings done by Chris.
| |
Wojtek P
Poland
| |
Posts 1597
07 Jun 2011 18:11
| How sysinfo measures chipram speed? 156 times A600 speed seems good result but may be misleading.
| |
Wojtek P
Poland
| |
Posts 1597
07 Jun 2011 18:11
| Marcel Verdaasdonk wrote:
|
Indeed these pictures are better then the one that you took before.
127MB/s is a fair score especially when you start comparing with the competition. ;)
The is still plenty of room to improvement IIRC the max would be somewhere near 400MB/s.(theoreticlimit)
Knowing this we can see there is still a long road ahead of us.
PS: I hoped Arne had a great birthday.
|
IMHO the problem is s-zorro bus in betweem that adds latency,
| |
Wojtek P
Poland
| |
Posts 1597
07 Jun 2011 18:15
| Gunnar von Boehn wrote:
|
One small update :-D
With tuned memsettings done by Chris.
|
Is it operating from these 4MB SRAM or from main memory with SRAM as L2 cache?
How large blocks are used.It seems like poor results. Whole ROM fits in onboard SRAM, 68060 can fetch longword/cycle. No idea why read from fastmem and from rom gives different results...
| |
Ayodele Stephenson
USA
| |
Posts 83
07 Jun 2011 20:05
| Wojtek P wrote:
|
Gunnar von Boehn wrote:
|
One small update :-D
With tuned memsettings done by Chris.
|
Is it operating from these 4MB SRAM or from main memory with SRAM as L2 cache?
How large blocks are used.
It seems like poor results. Whole ROM fits in onboard SRAM, 68060 can fetch longword/cycle.
No idea why read from fastmem and from rom gives different results...
|
Happy Birthday Arne!! Can someone compare these results with an A4000 040 and 060?
Would be nice to see some screen grabs with results from Syspeed. I think it would make the questions slightly more understandable and in context. Thank You.
| |
Marcel Verdaasdonk
Netherlands
| |
Posts 3991
07 Jun 2011 21:48
| Wojtek P wrote:
|
IMHO the problem is s-zorro bus in betweem that adds latency,
|
Latency are less interesting the longer your burst are.
| |
Matt Hey
USA
| |
Posts 737
08 Jun 2011 00:22
| Ayodele Stephenson wrote:
|
Can someone compare these results with an A4000 040 and 060?
|
Natami 060 120Mhz vs CSMK3 75Mhz (overclocked with 50ns SIMMs)
Cache Read 448.97 284.18
ReadROMb 61.49 35.15
ReadROMw 111.82 65.63
ReadROMl 148.62 85.64
ReadFastb 64.61 38.35
ReadFastw 89.77 53.04
ReadFastl 113.88 65.46
WriteFastb 51.16 30.97
WriteFastw 89.81 48.16
WriteFastl 100.00 48.19
Fast2Fastb 29.79 18.00
Fast2Fastw 44.65 26.45
Fast2Fastl 56.06 29.94
Fast2Fastm 52.15 27.57
Fast4Fast16 63.86 37.05
Note: The ReadROM values may be flawed as they shouldn't be any faster than ReadFast. All the Fast2Fast values should probably be double what they are. It's still useful for comparison which the Natami 060 board has almost 2x as fast of memory performance as the fastest old 060 accelerators. The only old accelerator with faster memory access than my overclocked CSMK3 would be an overclocked QuickPack 4060 with 50ns EDO SIMMs. My memory performance is 30-45% faster than a stock CSMK3. I think it's safe to say that the Natami 060 is the fastest 68k Amiga ever.
| |
Darren Eveland
Canada
| | (Natami Team)
Posts 50
08 Jun 2011 05:14
| Matt Hey wrote:
|
Ayodele Stephenson wrote:
|
Can someone compare these results with an A4000 040 and 060?
|
Natami 060 vs CSMK3 (overclocked with 50ns SIMMs)
Cache Read 448.97 284.18
ReadROMb 61.49 35.15
ReadROMw 111.82 65.63
ReadROMl 148.62 85.64
ReadFastb 64.61 38.35
ReadFastw 89.77 53.04
ReadFastl 113.88 65.46
WriteFastb 51.16 30.97
WriteFastw 89.81 48.16
WriteFastl 100.00 48.19
Fast2Fastb 29.79 18.00
Fast2Fastw 44.65 26.45
Fast2Fastl 56.06 29.94
Fast2Fastm 52.15 27.57
Fast4Fast16 63.86 37.05
Note: The ReadROM values may be flawed as they shouldn't be any faster than ReadFast. All the Fast2Fast values should probably be double what they are. It's still useful for comparison which the Natami 060 board has almost 2x as fast of memory performance as the fastest old 060 accelerators. The only old accelerator with faster memory access than my overclocked CSMK3 would be an overclocked QuickPack 4060 with 50ns EDO SIMMs. My memory performance is 30-45% faster than a stock CSMK3. I think it's safe to say that the Natami 060 is the fastest 68k Amiga ever.
|
I'm not sure where you are getting the Cache Read value? Unless the 448 is the timing and the second value of 284 is your MB/sec? What program did you use for those benchmarks?
I just did a test on my A4000/Cyberstorm PPC with 66Mhz 060, 60NS ram, and MAPROM option turned on (using SysSpeed 2.6) and got this:
EXTERNAL LINK
PS - @ Gunnar, can you please show the read and write chip speed test results? You know the ones directly below the Fast2Fast16 that you conspicuously omitted from the screenshot? ;) (as did I :) LOL!
But, seriously, I get really low chip mem benchmarks like 1-6.75 MB/sec ...Ok, well really low is "normal" for an A4000 ;)ps - how do I post an in-line image like you did?
| |
Matt Hey
USA
| |
Posts 737
08 Jun 2011 05:45
| Darren Eveland wrote:
|
I'm not sure where you are getting the Cache Read value? Unless the 448 is the timing and the second value of 284 is your MB/sec? What program did you use for those benchmarks? |
The 448 is the Natami 060 and the 284 is my CSMK3. This value increases a lot with the clock speed as the cache is in the CPU. My 060 is clocked at 75MHz.
Darren Eveland wrote:
|
I just did a test on my A4000/Cyberstorm PPC with 66Mhz 060, 60NS ram, and MAPROM option turned on (using SysSpeed 2.6) and got this:
EXTERNAL LINK 
PS - @ Gunnar, can you please show the read and write chip speed test results? You know the ones directly below the Fast2Fast16 that you conspicuously omitted from the screenshot? ;) (as did I :) LOL!
|
The chip ram speed should be the same as fast ram but it is never cached which will makes it appear slower in many cases. I kind of wondered if Gunnar left the results off on purpose.Darren Eveland wrote:
|
ps - how do I post an in-line image like you did?
|
Copy the picture to CLICK HERE and then use the img="picture" in brackets. The img tags I tried all ended up with CLICK HERE at the beginning though. Must be a feature of the web site :/.
| |
Gunnar von Boehn
Germany
| | (Moderator)
Posts 5775
08 Jun 2011 06:47
| Wojtek P wrote:
|
It seems like poor results.
|
Are you sure you understand what we talk about here?The results have a little room for improvement on the NATAMI,
but READ 114 MB, MEMCOPY of 127 MB is not poor. Are you aware that this memory result is higher than that a 600 MHz PowerPC Amiga?
| |
Mr Z
Netherlands
| |
Posts 5
08 Jun 2011 06:52
| Gunnar is it safe to say that the memory speed is now limited due to the 060 ?
Do you think that the softcore N68050 will be faster ?
| |
Gunnar von Boehn
Germany
| | (Moderator)
Posts 5775
08 Jun 2011 07:24
| Darren Eveland wrote:
|
I'm not sure where you are getting the Cache Read value? Unless the 448 is the timing and the second value of 284 is your MB/sec? What program did you use for those benchmarks?
|
I think he used Sysspeed and compared NATAMI with his AMIGA.
448 for the NATAMI and 284 for his AMIGA.
120Mhz 75Mhz 50Mhz?
NATAMI CSMK3 Darren 4000
Cache Read 448.97 284.18 256
ReadROMb 61.49 35.15 32
ReadROMw 111.82 65.63 62
ReadROMl 148.62 85.64 84
ReadFastb 64.61 38.35 36
ReadFastw 89.77 53.04 50
ReadFastl 113.88 65.46 62
WriteFastb 51.16 30.97 29
WriteFastw 89.81 48.16 50
WriteFastl 100.00 48.19 50
Fast2Fastb 29.79 18.00 16
Fast2Fastw 44.65 26.45 25
Fast2Fastl 56.06 29.94 30
Fast2Fastm 52.15 27.57 27
Fast4Fast16 63.86 37.05 34
| |
Gunnar von Boehn
Germany
| | (Moderator)
Posts 5775
08 Jun 2011 08:10
| Marcel Verdaasdonk wrote:
|
Indeed these pictures are better then the one that you took before.
127MB/s is a fair score especially when you start comparing with the competition. ;)
The is still plenty of room to improvement IIRC the max would be somewhere near 400MB/s.(theoreticlimit)
|
No, you can not reach 400MB/sec with the 68060.
You have to understand that a CPU has a bus protocol.
With a certain number of cycles per single access,
with a certain number of cycles per line burst,
with a certain number of cycles extra per cache miss,
with a certain ratio between CPU clockrate meaning that you might not have an extras cycle on a higher clocked CPU,
And last but not least to copy 1 MB of memory the CPU needs
to read 2 MB of memory and write 1 MB.
This means if your copy scores 64 MB/sec then this means that the CPU did actually move 190 MB/sec over the memory bus.
BTW the current memcopy SysSpeed results of the NATAMI are already a little better than a Pegasos 1 could do!
Wojtek P wrote:
|
Whole ROM fits in onboard SRAM, 68060 can fetch longword/cycle.
|
No, the 68060 can NOT fetch a LW per cycle.
If you read the 68060 manual ten you will better understand the memory protocol of the 68060.
But the 68060 is actually not bad. Its a very good CPU.
And the results here are not bad either.
Its true that there is still a little room for improvement.
Its also worth to understand that the memory code of Sysspeed is NOT the best possible code.
Its possible to write the code faster and to reach a little better results with the 68060.
With tuned CPU settings and a little tuned code you can get full 180MB/sec. This is about 50% better than a 600 MHz PPC!
| |
Gunnar von Boehn
Germany
| | (Moderator)
Posts 5775
08 Jun 2011 08:20
| Mr Z wrote:
|
Gunnar is it safe to say that the memory speed is now limited due to the 060 ?
Do you think that the softcore N68050 will be faster ?
|
Of course the 68060 does limit the results at some point.
Also when you increase the CPU clockrate you will always create effects that at some points you need to add a waitstate.
This means if your memory chips are for example able to run at 100MHz then the best memory score is able to achieve if you clock your CPU at 100Mhz.
If you clock your CPU at 120 MHz then there is a mismatch between memory and CPU clock and you loose some memory performance.
This means the pure memory score of the 120 CPU might be lower than the score of the CPU at 100Mhz.
Nevertheless in real live the 20% higher clocked CPU will give you the in total faster system.
We should not forget at what system we are looking here.
The NATAMI is a classic 68K AMIGA.
The NATAMI performance strength always was and is the chipset.
The 68K CPU is not its key performance asset.
The value of 68K CPU is compatibility and easy of programming it.
If you don't mind that the CPU is not the strongpoint of the classic AMIGA /NATAMI then these results are very promissing.
The current results show that even if the CPU is actually its weakpoint, the NATAMI/Classic AMIGA can compete and even beat PowerPC Amigas in memory intensive workloads done with the CPU.
I think this is a not bad.
| |
|
The team will post updates and news here