PlanetZ forums - scopeusers.com

Hi everybody.
Since I have been working on Pulsar V1.01 I was always wishing to have more power. So I bought another card and some time later a third one. After so many years I asked myself if the shark dsp wasn´t developed any further? so I searched the forum and the web and I found that the shark (not the mostly compatible tiger shark mentioned in earlier threads) is now in the third generation. Some versions work up to 300MHz provide more RAM, nine times the MFLOPs and are completely code and pin compatible with the ADSP-21065L (pulsar). All the sharks are made for distribution of one hardware design and supplying it with different DSPs to have multiple end products. The Price of the different 3rd generation shark vary between 15 - 25 $ (10Kpc OEM)
Now: could it be possible to alter the frequency/multiplier like doing it on a motherboard when upgrading a processor...
You guess what I´m suggesting?

<font size=-1>[ This Message was edited by: Rogurt on 2004-06-24 08:33 ]</font>

Maybe one day CWA will give us all a dream device and make a 15 DSP Tiger Shark PCI-X board

Hmmmmm smell the bandwidth...

If you go around changing the chips on your cards you can't go running to CWA for tech support.

If you get anything to work, let me know!!

Shayne

I know. But I do have a oooold Pulsar which I cant put into the comp because there are already 3 Cards in there. I thought about building a Pulsar Barebone but...
I´m trying to get considerations from guys with some knowledge on the materia. If it shows that this is not COMPLETE nonsense I would try to get some sharks and get the SRB altered by some guy I know and who does a lot of hardware developing.
Imagine what would happen if it really worked and I sent some pictures and benchmarks to CW...

By the way: think of the whole "turning ATI Radeon... into FireGL Graphics Card". Some things work even if you dont believe it on the first hand

I wouldn't touch the Pulsar Shark chips, it's not that simple: just replace the chips and hope for the best

Remember there's also an interface chip, IDs, power regulators designed for a specific power requirements, then DSP code might not be compatible 100%, then there will be timing issues , etc, etc, etc.

On 2004-06-24 13:23, Rogurt wrote:
By the way: think of the whole "turning ATI Radeon... into FireGL Graphics Card". Some things work even if you dont believe it on the first hand

yeah but that was just a matter of cracking the firmware/software, because the cards use the same hardware.

Yes that is what I meant before. The product sheet PDF of the analog Device Shark says explicitely that any of the sharks can be replaced by another type thus making it possible to have one hardware desing but selling two or more cards at different prices. Thats exactly what the story of the consumer/pro Graphics card is about.
(and the GPU of the consumer and pro graphic cards are NOT the same - more Renderpipelines, Hardwarelights etc.)

<font size=-1>[ This Message was edited by: Rogurt on 2004-06-24 15:06 ]</font>

... and the whole cracking thing was about a hardware lock that prohibits the use of the drivers for the pro version of the GPU (and some hardware features also). But thats the point. Different DSPs are sometimes compatible in a way that you can use the software written for the other one. Or vice versa: the same software can stay exactly the same even if the Processor changes

<font size=-1>[ This Message was edited by: Rogurt on 2004-06-24 15:04 ]</font>

right, the software instructions will remain the same, thus protecting the main costs in developement.

But the interface will almost certainly fail once you exchange the chips and clock the board higher.
Simplyfied there are timing constants in the code to anticipate what's ready at certain processing stages.
In other words the driving software has no idea about the 'increased' power and will provide it's instruction stream as usual.

Nevertheless your courage to fiddle with a quartett of millipedes deserves respect

cheers, Tom

On 2004-06-24 15:55, astroman wrote:

Nevertheless your courage to fiddle with a quartett of millipedes deserves respect

cheers, Tom

millipedes, that's a brilliant analogy

<font size=-1>[ This Message was edited by: BingoTheClowno on 2004-06-24 16:27 ]</font>

For anyone that knows:
do you think it's technically feasible for a next generation card to have upgradable DSP? Maybe on a daughterboard.

It would then be both cheap and simple to upgrade - best way to preserve your investment.

while speculations about technology are fun let's not forget that DSPs are't magic wands, but just (and simply) driven by software
in the match clockrate versus mind the latter always wins

you can see it on Pentiums and PowerPCs, the clockrate compensates either the inability of programmers or schedule pressure, depending with how much sarcasm one is willing to judge.

Apple distributes a program called Graphic Converter with it's latest machines, a well established piece of software for years (from the 68k CPU on). The performance under OSX is a joke and at best 10% of the original version.

On the Sharc side have a look at Transient Designer - a great tool burning close no nothing on the DSPs, but yielding perfect results.
Since there's a new reverb by STW I've checked what I have : the PT verb, Masterverb and the old Yamaha SPX90.
Guess what ? The Yamaha outperforms both in expessiveness of the sound. It colors and it's kind of noisy (of course) but it adds character and beefs up the sound - with a ridiculuous amount of processing 'power'.

I assume Warp69 added the modulation capabilities to his latest P100 release for good reasons
Imho we should keep our focus on ideas like that and not delaying everything to the next version of...

cheers, Tom

Well you´re right about the software issue. But over the years we stoped complaining about more and more hastily coded software. It was just no use arguing...
And if a software is already written who cares about optimizing the code (unless it just doesn´t sell -> Pulsar Ver1.01)

yeah, indeed - but those were the years when clockrates COULD be increased (at least somewhat) significantly.
What do you expect from 3 Gigs on ?
4-6 ok, maybe 8, but then - happy microwaving
and even current clockrates fail completely at supplying the CPU with proper instruction streams, resulting in empty running cores for most of the time

cheers, Tom

On 2004-06-25 14:01, astroman wrote:

and even current clockrates fail completely at supplying the CPU with proper instruction streams, resulting in empty running cores for most of the time

cheers, Tom

Astroman, do you have a link that details the above information? Are you refering to the Intel's CPUs instruction pipeline?

<font size=-1>[ This Message was edited by: BingoTheClowno on 2004-06-25 15:06 ]</font>

I picked it from YOUR reference to a company supplying boards with TigerSharcs

If I remember right they also offer boards with G4 CPUs and set up a test scenario to illustrate the differences in app design (since they want their customers to make the right choice).

Both a comparable CPU and a DSP solution (but both optimized like madness beyond 'regular' level) were compared in predicted (by chip capabilities) and measured values.

Bottom line : the Sharcs performed within expected range, but the G4 yielded only 20% of it's predicted power due to 'stuck' data between memory and CPU

This company had no reason to promote a certain brand, and used the results mainly to point out crucial parts of a design - so I consider it trustworthy.
A Pentium is almost certainly a complete magnitude lower in it's performance...

cheers, Tom
(I'll try to find that link back)

I remember now, I couldn't stop my uncontrollable drooling reflexes when I found this board:

http://www.transtech-dsp.com/tiger/index.asp

Or it could have been this company:
http://www.bittware.com

right the 2nd contains the link
http://www.bittware.com/corporate/artic ... partII.pdf
could be that I messed the 2 companies, looks like Bittware is only in Sharcs while Transtec uses PPCs too.

Nevertheless the effort put into the project in question can by no means be expected from 'consumer' programming - it's industrial hardcore at it's best.

8000 delivered FFTs on the G4 opposed to 64000 predicted ones was something to remember - and once again: since 'regular' developement is way below the level of sophistication demonstrated in this example one can expect even more dramatic losses of performance.

Makes native coding on Giga CPUs stand in a different light and me give a sh*t on benchmarks, data- and clockrates and all that blah...

cheers, Tom

<font size=-1>[ This Message was edited by: astroman on 2004-06-25 20:30 ]</font>

wow, nice thread.