|
|---|
Louis Dias
USA
| |
Posts 217
28 Mar 2012 19:51
| Marcel Verdaasdonk wrote:
|
Thierry again something i have said for the past 3 years making the CPU and stand alone project means that both projects could make progress independent and on the CPU side would attain a greater market support.
More 68K systems mean more programmers + code for similar systems.
And thus hopefully means more consumers. (which creates a virtuous loop)
Besides that because of the split, the chipset project members can spend more time in optimal solutions with more a abstract loop on the 68K then knowing every nitty gritty detail.(this is a double edged sword)
Besides that a split could gain interest from 68K enthusiasts for the CPU. (Atari fans, and etc)
|
Yes, my "commercialization" comment was meant to infer that they are going to be selling the core separately. The "CORE - IBM Wants Dhrystone Results" thread in the TEAM forum that showed up a week ago made me ponder this back then. It seems Gunnar did keep himself quite busy once he stopped posting publicly. ;)
| |
Asaf Ayoub
United Kingdom
| |
Posts 332
28 Mar 2012 19:55
| nice results. I wonder if Dr. Edward L. Hepler (Commodore Hombre project), EXTERNAL LINK would create some dicounted asics ;-) Link to vlsi-concepts.com
| |
Wawa Tk
Germany
| |
Posts 581
28 Mar 2012 22:47
| finally good to see some charts again. best thing would be though news that it runs inside natami fpga and some real world tests. likely it is not ready for that soon..
| |
Golgoth 27
France
| |
Posts 185
28 Mar 2012 22:57
| Samuel D Crow wrote:
|
Correction: This is the future home of where some real numbers will be when they exist. These are currently placeholders. (Oops.)
|
So for now it's not real numbers ? Or just expectations ?
| |
Marcel Verdaasdonk
Netherlands
| |
Posts 3991
28 Mar 2012 23:43
| Golgoth 27 wrote:
|
Samuel D Crow wrote:
|
Correction: This is the future home of where some real numbers will be when they exist. These are currently placeholders. (Oops.)
|
So for now it's not real numbers ? Or just expectations ?
|
They used a later generation Altera FPGA then used on the Natami to experiment with, so the results cannot be compared sadly.
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 11:42
| Marcel Verdaasdonk wrote:
| | They used a later generation Altera FPGA then used on the Natami to experiment with, so the results cannot be compared sadly. |
Are you saying that the Apollo figures could even improve when run on the same FPGA as the cited competetors (NIOS and POWER)?
| |
Marcel Verdaasdonk
Netherlands
| |
Posts 3991
29 Mar 2012 11:51
| That is not what i a saying, Nixus. What i am saying is compare apples with apples and not bananas. ;)
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 12:41
| Marcel Verdaasdonk wrote:
| | That is not what i a saying, Nixus. |
?SYNTAX ERROR Who is the first "They" in your comment that I didn't understand?
| |
Marcel Verdaasdonk
Netherlands
| |
Posts 3991
29 Mar 2012 12:49
| 'They' refers to those working on the N68 core.
Read the link above it posted the use of a Stratix and the Natami has a Cyclone (IIRC) as it's main FPGA.
IBM softcore performance study convirmed again the performance lead of the APPOLLO-Core.
IBM compared three cores. The POWERPC 440 core, the NIOS Risc CPU, and APOLLO-CPU.
The study was conducted in simulation and on a PLDA FPGA PCIe card using a ALTERA Stratix4 230C2.
|
so compare apples with apples please.
| |
Ayodele Stephenson
USA
| |
Posts 83
29 Mar 2012 12:55
| Good to see progress in both cores!!
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 12:58
| Marcel Verdaasdonk wrote:
|
'They' refers to those working on the N68 core. |
So apart from the "apples" thing you are saying that there was an "unfair" advantage for the N68k?
Read the link above it posted the use of a Stratix and the Natami has a Cyclone (IIRC) as it's main FPGA.
IBM softcore performance study convirmed again the performance lead of the APPOLLO-Core.
IBM compared three cores. The POWERPC 440 core, the NIOS Risc CPU, and APOLLO-CPU.
The study was conducted in simulation and on a PLDA FPGA PCIe card using a ALTERA Stratix4 230C2.
|
|
Well, I'm no FPGA guy, I designed real chips. However, I understand the above quote to say that all three cores were tested on the Stratix4. So it wouldn't really matter whether the Stratix is newer or older technology than the FPGA used in the Natami.
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 13:13
| I found some official documentation for the POWERPC 440 including the soft macro: https://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/F72367F770327F8A87256E63006CB7EC/$file/PowerPC440_Nov2006.pdf The performance of the soft macro is the same as that of the hard macros when looking at Dhrystones per MHz. The soft core has a much lower clock rate, of course. IBM gives 533 and 667 MHz for the hard macros and 300 to 350 MHz for the softcore. I haven't really followed the line of PowerPC processors but I guess that a 300 MHz 440 should be at least at the same speed as a 603e or 604. It doesn't seem to have an FPU, though. Looking for some performance figures, I found that in 2005 half of the top ten of the world's fastest computers used the PowerPC 440 at 700 MHz. If the N68k compares to this softcore in just some way, I'm optimistic that we will get unprecedented computing speed in an Amiga... :)
| |
Marcel Verdaasdonk
Netherlands
| |
Posts 3991
29 Mar 2012 13:19
| Nixus Again compare the same two things.
The test results were obtained using a different FPGA then Embedded in the Natami.
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 14:09
| Marcel Verdaasdonk wrote:
| | The test results were obtained using a different FPGA then Embedded in the Natami. |
Are you reading what I write? To me the quote from the Apollo web-site sounds very much as if all three soft cores were run on the very same FPGA (which is different from that currently used by the Natami). There is no reason a final commercial Natami couldn't come with a more recent FPGA.
| |
Golgoth 27
France
| |
Posts 185
29 Mar 2012 14:21
| As i understood it, the Appollo comparative vs the two other chips don't refer to real test numbers, as Samuel told it, they are "placeholder" (translated by "fictive" by my english/french dictionnary). To avoid the incomprehensions, i would like a clear answer about the numbers of the Apollo (i miss the N050 and N070 ;-) ). Are they based on a true simulation, or a projection, or just totally arbitrary ? In any cases, keep up the great work !
| |
André Jernung
Sweden
| | (MX-Board Owner)
Posts 988
29 Mar 2012 14:32
| Nixus Minimax wrote:
|
Marcel Verdaasdonk wrote:
| | The test results were obtained using a different FPGA then Embedded in the Natami. |
Are you reading what I write? To me the quote from the Apollo web-site sounds very much as if all three soft cores were run on the very same FPGA (which is different from that currently used by the Natami). There is no reason a final commercial Natami couldn't come with a more recent FPGA.
|
An Altera Stratix 4 230 with a C2 rating costs from about 6700 USD up to about 9000 USD per chip. Would you build a Natami based on it?
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 14:37
| André Jernung wrote:
|
Nixus Minimax wrote:
| | There is no reason a final commercial Natami couldn't come with a more recent FPGA. |
An Altera Stratix 4 230 with a C2 rating costs from about 6700 USD up to about 9000 USD per chip. Would you build a Natami based on it? |
No, I wouldn't. Did I say that the "more recent FPGA" had to be an "Altera Stratix 4 230 with a C2 rating"? No, I didn't. The most important point seems to be that the numbers do tell a lot if all three softcores are run on the same FPGA, no matter what FPGA that may be. Provided that the numbers aren't all fictional and the simulation and test runs have really taken place, of course.
| |
André Jernung
Sweden
| | (MX-Board Owner)
Posts 988
29 Mar 2012 14:42
| Golgoth 27 wrote:
|
To avoid the incomprehensions, i would like a clear answer about the numbers of the Apollo (i miss the N050 and N070 ;-) ). Are they based on a true simulation, or a projection, or just totally arbitrary ?
|
You would have to ask the person who made the page, Gunnar von Boehn. I would guess that some of the numbers are tested and some predicted. The page certainly was not meant to be spread to the public in it's current state, that is for sure. But unfortunately Samuel got a bit confused and posted it. If we are a bit more patient, we might hear some information from the people who did the majority of the work on the core, Jens Kuenzer (Deep Sub Micron) and Christoph Hoehne.
| |
André Jernung
Sweden
| | (MX-Board Owner)
Posts 988
29 Mar 2012 15:06
| Nixus Minimax wrote:
|
No, I wouldn't. Did I say that the "more recent FPGA" had to be an "Altera Stratix 4 230 with a C2 rating"? No, I didn't.
|
Sorry, I misread your post. But we are already pushing the limits of price with the current FPGA. If anything, it would be wise to stick with the same chip and get a lower system end price when the next chip generation arrives and the current Cyclone-4 series gets cheaper.
But yes, the cores were tested in the same environment afaik.
| |
Nixus Minimax
Germany
| |
Posts 275
29 Mar 2012 15:23
| André Jernung wrote:
| | But we are already pushing the limits of price with the current FPGA. If anything, it would be wise to stick with the same chip and get a lower system end price when the next chip generation arrives and the current Cyclone-4 series gets cheaper. |
Do you happen to know the price of the FPGA used in the Natami? But yes, the cores were tested in the same environment afaik.
|
This is some more speculation so people not liking speculation should ignore this: if the performance of the N68k is really quite a bit better than that of other commercially available softcores, this would mean that a good hard macro of the processor should do quite well performance-wise. Even when compared to (low-end) present day technology. Perhaps I'm so thrilled by all this because I always thought that the CPU would be the weakest link.
| |
|
Welcome to the Natami lounge.