@wegster

Quote:

PCI-X and PCI-express currently target different segments.

Factor in the release of Intel's E7525 (Tumwater) and nVidia?s nForce 4 Pro i.e. PCI-E for servers and workstations.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Wegster

You chop and change quite a bit, but that doesn't matter. The major problem we have with agreement is the point of view we are looking from. You say don't do in until we need it, I say do it now so it's ready when we need it.And if history tells us anything it is that you end up needing things long before you thought you would.
I also believe you are over conservative with your estimations and suppositions.For example, one of the reasons for re-writing OS4.0 in C was to make it much easier, simpler and less time consuming to port, and hence easily portable to new hardware!
Perhaps rogue could gtive us a time unti it would take....months or years.

As for PCIe, maybe AGP will be enough for now, but other considerations aside, I would rather have a couple standard PCIe slots, than PCI and a seperate dedicated AGP port.Thats just double handling to me.

FInally I have never said anything about dual G5 motherboard as needed, only that we allow for that option for when OS 4.x gets to that point of readiness.

The plain truth of the matter is that G4's are not competative with current x86 hardware, and if we do not make the move, or at the very least start planning the move to better hardware now, in two years we will have lost much of the catching up we have done.

Of course to do any of this I think we really need to leverage the linux market and really push it so we can get some commercial sales so we can GET the money to do these things.Remember I also talked about a G3/G4 board as well....which I believe the Artica S supports.

In two years I guess we shall see who was closer to the mark anyway.

_________________

@DrBombcrater

Quote:

And I'd point out that 100% of the Athlon 64 systems on the market

It's less than 100 percent.

Quote:

are still using AGP and PCI, but nobody regards them as out-dated or incapable.

Some integrated "PCI" device on AMD motherboards (e.g. A7N8X) are PCI-66.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@wegster
It is a simple side step 'thread distributed multiprocessing', of using normal sequential programing techniques and what is already established by thread based multitasking.

Parallel processing requires everything to be written for that purpose and completely different programing pardigm. Like a lot of good ideas, it is too grandiose to work well in a practical sense.

The multirocessor where two CPUs occupy the same slot, operate directly through pipes, is on the cards for future development, from what I understand this is sub-thread processing and uses cycles to distribute between the two.. I also understand, perhaps incorrectly there are inherent problems with scaling this idea past two processors.

Directling threads to different processors, might better be concievied as directing threads to different computers which just happen to be on the PCI-bus. It is a side step but a powerful one which niether requires new programing pardigms or extensive memoryt protection, shared registers.

Also this only makes sense because PPC computers on a PCI already exist from the same Teron design company.

Quote:

Let's assume that what you mean is to add multi-processing support to AOS, along with possibly a native (kernel space implemented as opposed to a 'pretend' thread implemented as a user process) threading model. At this point, with a good scheduler and priority model (which the Amiga has always had since the start for single CPU), the kernel can farm out some of it's processes (and threads) and distribute them properly across multiple CPUs. However....any running user process is very likely to be bound to execute on a single CPU.

Yes this is what I had mind and something that the Amiga was doing anyhow. The nidea would be minimal change to what we already have. As for a user process it depends how this is wrapped up. Wrapping aspects of the user interface can in effect deliver program threads to a "kernal" to distribute -- the user being none the wiser iun most cases.

Also launching threads from within combined programs is not a hard achievement as the central exe merely waits for returning values (keeping the sequential nature of presnt programing and gaining beneits from load sharing).

In short I am thinking of nothing more than an elobration of the present scheduler. As for the speed, as I am not contemplating micro co-ordination, but Macro-cordination, the waits will be small enough., way faster than ethernet speeds.

The idea here is that users are firing-off series of complex processes, but more importantly adding a lot of CPU wieght. Basically I am not talking of instensive tasks as running a multitude of tasks which belong to many different "applications" and sub-applications.

Quote:

If instead, you mean to treat each board as a seperate almost standalone computer...then I'm not sure what the benefit is? Clusters routinely have special dedicated switches for communicating between nodes that are 10GBit or I believe higher, dedicated only to cluster traffic, to deal with the latency issues in communication between the nodes.

I thionk you may be looking at benefits in an intense application sense, rather than as a host of applications environment. Losts of real world tasks, 3D in actual production, etc, are more usufully thought of in terms of distributed tasks in a multi-computer setting. Assembling m,any disjointed activities into a single cordinated application work well withg this, robotics, factory controls, real flow controls of complex sub-processes.

On a strict CPU race comparison I am sure my suggestion comes in as a very poor second, as a way of milking performance from standard passively cooled CPUs in practicial everyday ways that do not require specially written programs -- it has some virtues.

Plus I might add it is doable without too many changes at the moment while other "parallel processing" models are not (at least beyond a dual processor). At this mom ent I do not favoyr chacsing faster processors for perfpoormance, but looking for means to dramatically increase performance by using slower CPUs in innovative ways.

Of course I could be on the wrong bent altogether.

_________________
Greg Schofield, Perth Australia

@Hammer
Quote:

Factor in the release of Intel's E7525 (Tumwater) and nVidia?s nForce 4 Pro i.e. PCI-E for servers and workstations.

I saw the specs on the reference (presumably?) E7525 board design, and some of the chipset specs, but wasn't able to tell if it was able to handle more than the single PCIE x16?

The board itself if obviously nice- all of the 64 bit PCI slots are on seperte busses, which will certainly speed things up when needed. The reference board had 1 PCIE x16 and 1 PCIE x4 slot. I'd take one

My previous comment you were referring back to was before I realized that PCIE was in fact directed as a complete PCI replacement. I since did more 'homework' and saw that it's set to eventually phase out the current PCI 32-bit//64-bit, PCI-X etc...I'm still curuious as to what will use PCIE slots besides graphics cards, followed by 'the usual server hardware'? in other words, is there something in the pipeline that will need PCIE x16 bandwidth on a workstation that is _not_ a graphics card?

_________________
Are we not done with the same silly arguments and flames yet??!

@IonMane
Quote:

You chop and change quite a bit, but that doesn't matter. The major problem we have with agreement is the point of view we are looking from. You say don't do in until we need it, I say do it now so it's ready when we need it.And if history tells us anything it is that you end up needing things long before you thought you would. I also believe you are over conservative with your estimations and suppositions.For example, one of the reasons for re-writing OS4.0 in C was to make it much easier, simpler and less time consuming to port, and hence easily portable to new hardware! Perhaps rogue could gtive us a time unti it would take....months or years.

Not really (change quite a bit). It's mostly looking at the existing market, with the $ I'm guessing is involved, both in the production and investment, versus my estimates of what certain companies have or are willing to invest.....versus 'what I'd like to see'

As you said, additional $$ really _needs_ to come from somewhere, and the Linux market could be a prime target, but I see little that would tell me that EyeTech has made any progress on that front, unlike some people's least favorite 'other PPC platform.'

Look at what the cost of entry into the outdated (comparitively) G3 and G4 A1s costs users now. I expect the uA1 is fine for AOS4, but it isn't much for inspiring many people to purchase one to run Linux on by any means, especially with the memory constraints. The uA1 will _not_ have Linux users of any kind purchasing it at the current cost, unless they have more $ than sense. An XC style, Pegasos 2 style, or 'next gen A1' might...if it sees reasonable performance under Linux, at a halfway competetive price.

Anyways, I'm all for a G5, but if that G5 were to cost me $2k, there's no way I'd touch it..._that_ may be where/why you're thinking I'm changing my mind? I'm not sure where the dual G5 came from?

Likewise, I think PCIE would be a good move, and if the Amiga platform didn't get a new motherboard for some time, driver updates could allow newer cards to be supported, and someone would presumably come out with some CPU upgrades....so it would be in better shape than now. The problem here is, I see it all about tradeoffs- need $$$ to get more modern stuff made, but I don't think there's that kind of money in EyeTech/AIncs pockets that's required to get the technology, plus make it halfway reasonably priced.

I'm not entirely sure what the uA1-C's market is, aside from yet another set top or media center style box, or kiosk computer, but it certainly isn't the Linux market..I suppose it's _possible_ the uA1-I could possibly be made into a blade, but that would require either a good deal with a well-funded company, or again, more money than I expect EyeTech and company have in their pockets.

So...I'm not entirely too sure what we're 'disagreeing' on here? I'd rather see a PCI-Express system, with perhaps dual G4 ability or single G5, DDR RAM, etc...but if a board came out in a year that was a 'fixed XE' with AGP4x-8X, USB2 and firewire, DDR, I think it would be worthwhile, and possibly enough to ensure the $ for the follow-on board that would hopefully be closer to 'entirely up to date'....

_________________
Are we not done with the same silly arguments and flames yet??!

@Rogue

Quote:

Actually, it's Jedi Knight III: Jedi Academy. Never played KOTOR 1, I only have a Duron 850 and they never cared to release a demo so I dunno how it runs

When not chasing for crazy FPS and resolution, KOTOR (RPG type game) should run OK on A64@800Mhz (forced PowerNow) with Geforce FX5900U or ATI R9700 (laptop).

In my POV, Bioware?s KOTOR-I story plot and characterisations are better than SW-EU stuff (post-ROTJ except for T. Zhan?s books) i.e. it captured the key elements of the original trilogy. This can be also indicated by numerous fanfics sites specific to KOTOR. It's also captured Gamespy's RPG "Game of the Year" for 2003.

Quote:

OTOH, Neverwinter Nights - Hordes of the Underdark played just nice...

A Bioware game... I haven't tried Neverwinter Nights.? I recall you have an OS to finish first???

PS: My priority in games is for characterisation and story plot not pretty graphics/ tech demos (e.g. DOOM3, Farcry).

Last edited by Hammer on 27-Jan-2005 at 06:03 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@mlehto

Quote:

Toaster is multi-channel card loaded with bunch of quality sw, ready to broadcast from TV-stations. Calculators are ... calculators. I doubt, that their quality is enought anything more than edit something to tape for personal use.

Keep in mind that, "3D Edit" manipulates videos within the digital domain i.e. it's like Sony's Vega 5.0 with GpGPU's computational performance for RT effects, hence my calculator comments. Also, one has to keep in mind the digital broadcasts (MPEG2 SDTV/HDTV).

**NVIDIA?s NV2A (XBOX1) has theoretical 80 GFlops.

Last edited by Hammer on 27-Jan-2005 at 06:14 AM.
Last edited by Hammer on 27-Jan-2005 at 05:20 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@wegster

Quote:

is there something in the pipeline that will need PCIE x16 bandwidth on a workstation that is _not_ a graphics card?

Clearspeed?s 96 math co-processors on a chip. This product can process 64bit double precision math at ~50Gflops @ 250Mhz. Targeted for science and financial simulation markets. The product is mainly aimed for Xeon and Opteron servers**. PCI-E version is expected sometime in 2005. It?s roughly a rival to the new GpGPU non-gamming applications (32bit single precision floating point calculators).

With two Clearspeed cards, you have 100Gflops of 64bit double precision floating point math calculator on a low end server.

Last edited by Hammer on 27-Jan-2005 at 05:46 AM.
Last edited by Hammer on 27-Jan-2005 at 05:45 AM.
Last edited by Hammer on 27-Jan-2005 at 05:38 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@wegster

You go for one "open-ended" x4 PCIe and one PCI-E 16X slot for semi-SLI setup i.e. a hack on nForce4 Ultra chipset. Upto 80 percent of a full SLI setup.

Reference.
http://anandtech.com/cpuchipsets/showdoc.aspx?i=2322

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@wegster

I believe the two boards are aimed at company and industrial markets as, like you said, kiosk systems, perhaps cash registers, industrial controllers and that kind of thing. For those customers a larger production run could then be made thereby reducing the cost of the board not only for them, but also for us(as I would expect a couple hundred added onto the order for our use). The problem is that I have seen nor heard nothing from eyetech about pushing these board into the market. These G3's could easily be a linux workstation or an office terminal.At the moment I think these boards are mainly "demo" models.

I would have expected at least some kind of announcement that Eyetech has partnered with one of the linux PPC distributors aiming at a specific purpose/solution....but mayhap that is all behind the scenes.

512MB DDR400 SDRAM (PC3200) - 2x256
? 80GB Serial ATA - 7200rpm
? 8x SuperDrive (CD-RW/DVD-R)
? ATI Radeon 9800 XT w/256MB DDR SDRAM
? PCI-X Gigabit Ethernet Card
? Apple Keyboard + Apple Mouse
? Mac OS X [default]

Estimated Build Time:
6-8 days
Free Shipping
Amount ex GST A$ 2,926.36
GST A$ 292.64
Subtotal A$ 3,219.00

When you consider that this works out to roughly $2200US, does not have PCIe, only has one PCI-X (and that is for a gigabit ethernet card), nor some of the other options we have discussed, $2000 is not too bad really (assuming you were talking US dollars) and this only has 1 G5 1.8 Ghz CPU.
With that one developer board we could exceed these specifications and come in with hardware that is top level in time so we are no longer chasing, but competing.

However, this really is for the future, not for the XC. For that I would like to see a board similar to the current one, but with some kind of high bandwidth riser card capable of of at least having AGP x8 and 4 useable PCI cards. a single board that is highly configureable so that it can suit a very board group of requirements.

_________________

@Englyst

Quote:
Correction: PAL D1 = 720x576. And you're right about the Amiga Toaster card. reason was that it used composite connections and quality was never up to broadcast quality.

Thanks . This is prob, because resolutions vary from source to source .


Quote:
I've edited with a DPS PAR on the PC and the signal compression was horrendous. You could easily spot the compression artifacts But when it was released it was awesome for it's time.


This time it is impossible to blame pc-quality, since amiga PAR input is same card. Ie. it is same ISA-card.

It depends of speed of disks, q-factory from PAR software and mostly quality of source. When I recorded different sources from clients, I got sometimes very bad quality of results, when I got low quality svhs/beta tapes and artifacts were reality. But normally if source was enought well recorded, I didn't get any artifacts. Same with computergenerated anims, since they are allmost noise-free. PAR packing algorithm is hitted by noise very easily in recordind.


When you use VT4, do you have single processor mobo and what are you processor specs ? Sounds nice, with 5 to 6 layers you can do most things

Yes, this is what I heard, that VT4 is true beast at its price. Both in quality and features.


Quite reasonable prices for HDV. I don't want to think, what HDTV will cost. Anyway too much .


Miikka

@Hammer

Quote:
Keep in mind that, "3D Edit" manipulates videos within the digital domain i.e. it's like Sony's Vega 5.0 with GpGPU's computational performance for RT effects, hence my calculator comments. Also, one has to keep in mind the digital broadcasts (MPEG2 SDTV/HDTV).


Edited: I talk about SVGA cards like X800 or so...

I keep it ... and it is not a problem. As far as video is edited inside of computer, there is no problem. But when you throw it out via s-video or composite connector, analogic world step in. And this is keypoint. In ordinary PAL/NTSC signal is possible to drive it via s-video (wich is not practically very high-end) or composite (wich is worse). Signal dynamics (wideness ... ) is not so high. But in practice these toy-electronics cannot do it enought well plus they lacks some features, if you read what I wrote before.

When you throw HDTV signal out via some converter, it is even more difficult, because dynamics is more higher.

I can accept these cards to preview only via more or less ordinary video-monitor, not more . It even reduces costs, since good reference monitors are awfully expencive


If you think, that using MPEG1/2 streams to output help some, it doesn't make any difference, if you get them out via analogic output. It makes situation even worse, because if you use source with artifacts and noise, they usually get worse after D/A conversion.


They can say anything in adverts, but it is not needed to believe everything


If you try to find a way, how to solve these problems, you have to use external device/s, wich converts signal to analogic. Then you have different solutions. Use SVGA to PAL/NTSC converter, or use FW card and some quality device, wich make job for you.

I put this allso, because it gives some explanation about signal. Forget digital tech, because if you use anything analogic in your chain, things get totally different from view of output.

If and when you use SVGA/computer generated graphics, you have to take care some things. SVGA signal dynamics is higher than PAL/NTSC, and you have to use colour limitter (try IFX... :) ), wich cut ~20% away from each separate colour. Ie. no over 200 RGB values (somewhere there...) . Other way, colour range is reduced from 0-255 to 0-200 ... usually it is maintained so, that it is reduced to somewhere 10-210 from 0-255.

Another probs are, how to keep things in sync. There is not SMTPE connector in these cards itself, so it is impossible to keep them in sync with other devices, with sound and so on.

Since this site is amiga related, I dare to say this. It is allmost ridiculous to say, that these toys can reach anything broadcast quality. They can be suitable for very good editing, but not output. Or why there is equipments around, wich cost fortune, if you can take same from pc-store less than 1500e ?? Because video-editing studio owners are so stupid and are ready to pay for it ??

If you want to use X800 to broadcasting, feel free to do so. I'm not correct person to talk about this, because I'm bit out, what is happening just now, ie. formats, singnals today and so on.

Key point is not GFlops or TFlops, Key point is D/A conversion, where it happens and what is quality. I'm not checked video D/A:s, but generally 16bit medium quality D/A or A/D converters, wich are made for sound, have accuracy something 16 bit -+4 bit. Studio quality converters for sound (24bit +-0.5bit, last bit is correct mostly every second time, that is 0.5 bit) are very expencive. Sound signal dynamics is relativelly low. Videosignal dynamics is very high. So they are more expensive in video world.

I really have to work some ...


Miikka

Last edited by mlehto on 27-Jan-2005 at 12:44 PM.
Last edited by mlehto on 27-Jan-2005 at 12:42 PM.

@mlehto

Quote:

When you use VT4, do you have single processor mobo and what are you processor specs ? Sounds nice, with 5 to 6 layers you can do most things

Tyan Thunder i7505 motherboard with onboard Dual U320 SCSI controller with 3 Maxtor Atlas 72GB U320 SCSI disks in a software RAID-0 configuration (180 MB/s), and 2 Intel Xeon 2.4 GHz processors, 2 GB RAM,, GeForce TI-4800SE with 2 19" Iiyama monitors connected with a combined resolution of 3200x1200x32. 3 Maxtor 250 GB OneTouch drives connected via Firewire as backup drives.

That's the config I use.

_________________
A1200 with DKB Cobra 68030 + CF · A2000 with SCSI2SD · A4000 '040 with Mediator and Voodoo 5

@Englyst

Thanks It sounds quite nice ... and "only" two processors, not bad.

Do you feel, that bottleneck is processor speed or disk access ?? If you have any bottlenecks ... how many channels/layers you can handle with VT software ?? And what is max. number of channels from sw ??


Miikka