Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Documentation about the Vampire hardware

PowerPC In FPGA - How Much Sense Does It Make?page  1 2 3 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
14 Dec 2019 09:16


We got ask if putting a PPC into an FPGA makes sense.

To answer this I think it makes good sense to explain a few things about FPGA.

Some years ago I participated in a project at IBM
which goal was to port the PowerPC-460 CPU from ASIC to FPGA.

As you know the PowerPC 460 is the CPU used in the SAM systems.

The 460 CPU cost as ASIC a few dollars and reaches as ASIC over 1 GigaHerz.

At IBM we used an very high-end FPGA which cost over $5000 per FPGA.
In this FPGA the maximum clockrate for the Soft-460 was 210 MHz.

What do we clearly see here:
- FPGA by design will always reach a lower clockrate than ASIC.
- The used FPGA reaches only 20% speed of the ASIC but had 200 times higher cost! 

If we would have used a more reasonable priced FPGA e.g. for $100 per unit - the speed of the 460-CPU would have been much lower - more like 80 MHz.

Part of the FPGA IBM team did also run various benchmarks on the PPC 460 Core and compared the PPC with 2 other FPGA cores.
One of the 3 benchmarked cores was APOLLO Core.
APOLLO did win this benchmark shootout by far.

It was clearly visible that if running in the same FPGA - APOLLO was much faster then the PPC and the other RISC chip which competed against it.

So its very clear:
Putting a PPC as "accelerator" in an FPGA makes no sense.
As its not faster than APOLLO 68080 but much slower.

A PPC inside an FPGA can run PPC software but a lower speed than existing ASICs.
Roughly ~ 10 times slower than existing ASICs.



Richard Gatineau

Posts 60
14 Dec 2019 10:07


Does an APOLLO ASIC make sens? (for IBM or others, for Vampire V4/V5)


Michal Pietal

Posts 236
14 Dec 2019 10:11


Putting PPC in FPGA - makes no sense.

Putting PPC into a V4SA daughterboard card - could be reasoned, possibly.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
14 Dec 2019 10:11


Richard Gatineau wrote:

Does an APOLLO ASIC make sens? (for IBM or others, for Vampire V4/V5)

The answer is complex.

Today that we use an FPGA is great as it allows us to upgrade and improve the system continuously.

We can update and improve SAGA, we can add better 3D acceleration to APOLLO etc.

There are several topics we are working on and developing all these improvement is a lot of effort and will take many many month to come.

Doing an ASIC run is significant work and also risk.
I think for us this only makes sense after all improvements and upgrades we want to do are done, and 100% tested and polished.
So maybe in 1-2 years time.

Then I believe we will be ready for the next big step.


Richard Gatineau

Posts 60
14 Dec 2019 10:16


Thank you.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
14 Dec 2019 10:17


Michal Pietal wrote:

  Putting PPC in FPGA - makes no sense.
 
  Putting PPC into a V4SA daughterboard card - could be reasoned, possibly.
 

 
Yes Michal, you are 100% right.
 
A PPC in FPGA will be just to slow for reasonable OS4 or anything like this.
I'm pretty sure no one will like running OS 4 on a PPC460 @ 80MHz, right? It will crawl.
 
Adding a daughter CPU board with e.g. 800 MHz PPC will make a lot more sense.
 
The problem with PPC is that its dying technology.
In the 90th this was totally different - there PPC was emerging technology, there you could get good priced, reasonable PPC chips.
 
But this is not the case today anymore. Today all company are abandoning PPC technology.
Its getting very difficult to find good PPC chips today allowing you to build a CPU card.
 


Wawa T

Posts 695
14 Dec 2019 18:54


whoever wants to run ppc code like os4 or morphos or aros ppc can simply acquire a sam460 or run in under qemu on a pc. im not sure why it should be demanded that vampire is capable of such things. keep it simple stupid as much as possible.


Wawa T

Posts 695
14 Dec 2019 18:56


it also makes me wonder that such a stupid ideas as ppc in fpga are being repetitively discussed on sites as aw.net. likely just because m68k users have fpga systems so os4 users want to have them too just for the sake of it.


Kyle Blake
(Needs Verification)
Posts 108/ 1
14 Dec 2019 20:27


OS4 users don't get to have FPGA. They had "what is X" on their motherboard for ten years and did nothing with it.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
14 Dec 2019 20:47


wawa t wrote:

it also makes me wonder that such a stupid ideas as ppc in fpga are being repetitively discussed on sites as aw.net.

I think this is not surprising
If you have no background knowledge about FPGAs
than its easy to think that an  FPGA is a magic wonder weapon
allowing you to make a CPU super fast, make a GFX chipset that beats latest NVIDIA and in addition also mines bitcoins for you - making you a billionaire in no time.

FPGA knowledge is rare island knowledge - I do not blame anyone for not beeing an FPGA expert.



Vojin Vidanovic
(Needs Verification)
Posts 1916/ 1
14 Dec 2019 20:48


Kyle Blake wrote:

  OS4 users don't get to have FPGA. They had "what is X" on their motherboard for ten years and did nothing with it.
 

 
  I have one PCI-e slot less modified to fancy name. Poor os4 Dev tools, mailed chip manufacturer, not interrsted to port Linux x86 tools to ppc.
 

  Have one prototype Blank card made wrong (will be on ebay 2025) and I recall sentinel autologger debuger card shortly on sale and Aron website.
  I recall one os4 depot led blink tool that uses it.

  Not a Bad Idea, Bad execution. Would swap for cyclone3-5 anyday
 


Wawa T

Posts 695
15 Dec 2019 02:11


Gunnar von Boehn wrote:

I do not blame anyone for not beeing an FPGA expert.

by now you must be aware people like me lack almost any engineering knowledge, yet it is not outside my grasp, to inform oneself and understand the concepts and understand what these tech can achieve. thats a question of attitude. in particular a kind of attitude described by "hoping and dreaming" and not bending a finger, i may add.


Marian Nowicki
(Needs Verification)
Posts 22/ 1
16 Dec 2019 07:58


PowerPC In FPGA - How Much Sense Does It Make?
Someone may for example ask why use vampire when anbody can buy better PC for 5 EURO or mac for MOS for 10 EURO.
You see it wrong. These questions don't make sense.
PowerPC is something that used in Amiga for more than twenty years.
People want to use something that they use when they are young, that's ok. 
Even if today it is useless, so what?
What is important is it was in their Amiga twenty years ago.



David Pesce

Posts 12
16 Dec 2019 08:17


Few people were using PPC and most of the time it was only used as coprocessor for decoding mpeg2, mp3 ...
You can do that with the actual vampire. If you want more power, you could simply write an arexx network service to launch task on distant servers. So you could use a MacG4, an X86 or an ARM.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
16 Dec 2019 08:35


Marian Nowicki wrote:

PowerPC is something that used in Amiga for more than twenty years.
People want to use something that they use when they are young,

 
But doesn't this make as much sense as asking for a "wooden leg"?
 
Lets get back here and look at this from engineering point of view:
 
- 68K is the original CPU of the AMIGA.
Choosing 68K as CPU for Amiga was clever:
- the 68K is very elegant and very easy to program.
68K allows programmers to write good programs.
 
Why was PowerPC chosen?
PowerPC was not chosen because its an elegant architecture.
Writing code is clumsy on PowerPC.
You need on PowerPC always many instruction to do the same work.
This leads to big programs which are real hard to oversee,
this lower fun on developing and increases risk of bugs.
PowerPC is harder to program and clock by clock much lower performing than a good 68K.
 
PowerPC was clearly a 2nd choice, and only taken because they could get no fast 68K CPU on the market.
 
You can think of the topic like if you loose your leg in an accident - someone gives you a wooden leg as replacement.
 
 


Rob M

Posts 60
16 Dec 2019 10:09


Vojin Vidanovic wrote:

Kyle Blake wrote:

    OS4 users don't get to have FPGA. They had "what is X" on their motherboard for ten years and did nothing with it.
   

   
    I have one PCI-e slot less modified to fancy name. Poor os4 Dev tools, mailed chip manufacturer, not interrsted to port Linux x86 tools to ppc.
   
 
 
 
    Have one prototype Blank card made wrong (will be on ebay 2025) and I recall sentinel autologger debuger card shortly on sale and Aron website.
    I recall one os4 depot led blink tool that uses it.
 
    Not a Bad Idea, Bad execution. Would swap for cyclone3-5 anyday
   

The best development tools and support in the world wouldn't fix the major problem with the XMOS subsystem.  That is that the userbase is simply too small to contain many people with the knowledge, skills and a clear idea of a project they think it would be ideally suited to.

If you think it would be any different with an FPGA you only have to look at the number of projects that use the Lattice FPGA of the Sam440 or Sam460.



Marian Nowicki
(Needs Verification)
Posts 22/ 1
16 Dec 2019 15:42


PowerPC was chosen because there was no faster 68k.
Motorola switch to PowerPC because was unable to compete alone with Intel and need to cooperate with Apple and IBM.
It is history. It was more than 25 years Ago.
Amiga Tech announce switch to PowerPC in 1995 and PowerPC cards for  Amiga were avaible in 1997.
It is also history. If people want to use PowerPC in Amiga it is ok.
After all Amiga is hobby, so why not?



Olaf Schoenweiss

Posts 690
16 Dec 2019 17:24


no problem...

who implements the PPC in FPGA?

You?

Gunnar certainly not


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
16 Dec 2019 18:14


Marian Nowicki wrote:

Motorola switch to PowerPC because was unable to compete alone with 

 
You get history here a little wrong, my friend. :-)

Motorola was not competing with INTEL.
 
Motorola once had a very dominant position in 2 major markets - in both markets Intel played no major role .
 
a) Mototola was dominant in Arcade and home computer Market
 
b) Motorolas high price market was UNIX servers.
Motorola sold here 68030 chips and later 68040.
 
Motorolas main problem was their development cycles were long,
that 68040 was delayed, and did not reach the expected clockrates.
 
Their customers got disappointed by the slow progress of Motorola.
The new RISC chip idea allowed, shorter and cheaper development cycles.
And mostly of all Motorolas customers started their own RISC chip developments, SUN (SPARC), HP (PARISC), MIPS ...
 
The RISC chips were clear a trade-off.
 
- Not as programmer friendly,
- relative weak on Integer performance.
+ But easy to build, fast to build, which short development cycles.
+ Much easier to pipeline, therefore easier to get high clock.
+ Easier to SuperScalar. Much much easier to verify = saves a lot $$
+ And with the shorter development cycles the RISC chips were able to quicker benefit from new ASIC processes = were able to earlier benefit of higher clock.
 
Motorola lost all their big customer this way.
And decided to give up on the development of complex CISC chip, and did design their own RISC the 88K.
 
After their 88K was designed, working and doing actually well,
IBM showed up and "persuaded" Motorola to trash their own 88K
and to be IBM new partner in crime = producing the PowerPC
This is how history went.
 


Wawa T

Posts 695
16 Dec 2019 20:00


@gunnar

thx for the insights:)

posts 43page  1 2 3