sa110
9/12/05 12:59PM |
Excellent news on the possibility of an A9home version |
highlandcattle
9/12/05 1:00PM |
I don't get t how can it work on the A9 it has no Gforce2 inside??? |
Sawadee
9/12/05 1:02PM |
A nice speed increase all round. 
Good news for browser speed gains. |
adrianl (+4.2)
9/12/05 1:30PM |
In reply to highlandcattle:
much of the code is not specific to the GeForce2. In fact Geminus already includes driver code for GF2, SM501 (used in the A9home) and VIDC20 (RiscPC). |
egel
9/12/05 2:01PM |
With a ViewFinder-version too can everyone be happy. |
em2ac
9/12/05 4:58PM |
hmm, interesting, well if ya do get it to work on the RiscPC with even an ok speed increase, i'd buy it!
Cheers! |
AMS
9/12/05 7:57PM |
Good news that Geminus acceleration will be available on the Iyonix then. The nVIDIA is capable of a turn of speed and if code is available to make better use of it then all the better.
In reply to highlandcattle:
"I don't get t how can it work on the A9 it has no Gforce2 inside???"
Short answer - it *won't*, well not in the same way. Caching the bitmaps (even with A9's more limited graphics capabilities) may bring some useful speed improved - if nowhere near as much as with the Iyonix's nVIDIA. That having been said an improvement is, well, an improvement.
The cache I would think in the case of the A9 would simply be memory space taken out of it's main RAM - whereas on the Iyonix it's on the video card (so you won't be losing precious main RAM just for "caching" bitmaps on the Iyonix).
The take home point is "Geminus" is not a single thing - it's different things on different platforms. It's overall effect will be to improve speeds on whatever hardware it runs on. |
bucksboy
9/12/05 8:01PM |
Very pleased with this - the desktop seems brisker, and Firefox is transformed. Doing F12-Return on complex Draw and AW files shows they redraw in less than half the time. Putting two 8Mpix JPEGs on screen side by side and dragging one over the other is /much/ quicker and smoother. The only downside I've found so far is that Thump has to reload thumbnails for the images in each of my image directories, and this always takes a while the first time around (once done however it's very quick). Thanks again to Neil and his team. |
adrianl (+2.0)
9/12/05 8:06PM |
In reply to AMS:
The A9home does have some off-screen memory that can be used for cacheing sprites and window contents, albeit only a total of 8MB video memory versus the 64MB on the nVIDIA cards (32MB on older cards).
I intend to add support for extra cache in main RAM too, because the A9home has limited memory (and even the nVIDIA card's memory can seem limited in high res screens) and because the RiscPC version will require it. |
jms
10/12/05 6:35AM |
In reply to adrianl:
As well as cacheing sprites, are there any other improvements that could be incorporated in the RiscPC version? |
adrianl
10/12/05 4:44PM |
Sprite cacheing is something that's unlikely to be useful on the RiscPC because there's no hardware capable of copying cached sprites to the screen quickly. I think you mean redraw cacheing, which remembers the contents of windows so that they can be redrawn quickly.
This does make a big difference for slower apps on the RiscPC, with all processors from ARM610 through to StrongARM.
Some other improvements may be attainable using some of the Geminus code but the OS code is a better match to the hardware of the RiscPC and earlier machines than it is to the IYONIX pc (it hasn't been updated substantially), so any gains will likely be smaller. |
ROHC
10/12/05 10:22PM |
In reply to adrianl:
This rocks! 
Nice to see that others will benefit too...
& more speedyness to come?
|
adrianl
10/12/05 11:53PM |
There's still a good bit of room for improvement, yes. I've actually disabled a couple of features in the current release because I'm not yet convinced of their stability. |
EasyKees
11/12/05 1:24AM |
When can we try a A9Home version ? ( DEMO or FULL ) |
adrianl
11/12/05 4:16AM |
I can't make promises on that score, I'm afraid. It shouldn't be very much extra work to make the redraw cacheing work properly on the RiscPC again (that is where it started out) and thus on the A9home*, but there is also the small matter of a DVD player that I want to finish 
* the code does actually work on both, but it needs improvement and some small corrections. |
lproven
11/12/05 2:16PM |
Does make you wonder what the original RO5 driver on the Iyonix actually /does/ with its GeForce 2MX, tho', doesn't it? Does it take this super-powerful 2D and 3D accelerator with a local processor, 2 texturing pipleines and 2 pixel pipleines, buckets of RAM and the ability to draw 700 million textured pixels a second or 20 million polygons, including texture-mapping, shading them, lighting them and transforming them... and treat it as a dumb unaccelerated framebuffer? |
ROHC
11/12/05 3:29PM |
In reply to lproven:
I think that waht the RO5 driver does is just make the card work!
Given that nVIDIA drivers are closed source it must have been one hell of a battle for Tematic to have got as far as they did in the time that they did...
In reply to adrianl:
Hey does this mean we'll get filer speedyness too?
I understand that there are bits of the Xscale that RO5 isn't yet optimised for... |
highlandcattle
11/12/05 6:06PM |
Can somebody plz explain this a bit to me..
So tematic writes a driver that makes the card simply work.
What is it then that Simon wilson ported over from BeOS?
and what Adrian in essence wrote is simply a better driver that chaches sprites and stuff??
Pardon my Ignorance |
adrianl (+0.1)
11/12/05 6:28PM |
In reply to ROHC:
Filer speediness? Explain please.
In reply to highlandcattle:
Tematic/Castle wrote a driver that makes the card work but also does rectangle copying and solid filling in hardware (you'd find the machine unusably slow otherwise)
Simon ported the 3D driver and OpenGL-compatible library. There are presently no 3D programs for the IYONIX pc/RISC OS capable of using this 
Geminus uses the 2D hardware acceleration more extensively with the aim of making the desktop more responsive and doing so without requiring applications to be modified.
Geminus does still require the Tematic/Castle driver for setting up the screen and defining/moving the mouse pointer but does everything else itself. |
leeshep
12/12/05 2:34PM |
I've been using the beta of this software on my machine for a while now, it really does make a noticable difference to the speed of the desktop (redraws ect) , the beta has now expired and I really miss it already, I'm going to buy the full version and maybe even invest in a second monitor and graphics card |
DS1
12/12/05 2:47PM |
I've got a ViewFinder in my RPC which allows me to have 16m colours at 1600x1280. Set the same colour depth and resolution on my A9 and window drags are noticably slower and jerkier than on the RPC. Geminus for the A9 will be a very welcome improvement, and if it can also be made to work with/alongside ViewFinder, I'll be as happy as a fly on ... let's not go there |
highlandcattle
12/12/05 5:44PM |
thats perhaps because the A9 has a graphics chip designed for mobile phones and PDA's while your viewfinde cart has a graphics cart from 7 years ago which is still more powefull then the one in the A9 |
bucksboy
12/12/05 6:58PM |
In reply to DS1:
I'm not sure there's much more to come from the VF; John Kortink did a lot of development of the acceleration aspects of the s/ware during the life of the project. The main problem with it was not the card or the s/ware but the speed of the podule bus (easily demonstrated by running the !Tumble demo with VF switched on and off). |
DS1
12/12/05 7:36PM |
In reply to bucksboy:
yeah, but it's a brilliant piece of kit nonetheless. I couldn't now live without my ViewFinder, it's just too useful when I'm editing photo's. I'm expecting eventually to transfer much of this work onto the A9. At the various shows the a9 people have pointed out that graphics acceleration hasn't been switched on yet, so when it is that'll be at least as good, and probably a lot better I suspect. Plus I can cart it around much easier than my RPC! |
gvrace
13/12/05 2:02PM |
Does anyone know if the screen acceleration part of geminus includes non desktop software or is it only multitasking desktop apps. |
adrianl
13/12/05 4:24PM |
Sprite plotting and cacheing will work outside the desktop too, although many apps that run outside the desktop will just manipulate the screen directly and thus can't be readily accelerated.
Obviously redraw cacheing is specfic to the desktop environment. |
gvrace
13/12/05 7:03PM |
Okay so screen acceleration is mainly geared towards desktop apps and as I understand it also SpriteOp commands outside desktop ?
So how compatible is geminus screen acceleration with other newer GeForce cards ?
Is there any SWI's that can be accessed or is it all 'behind the scenes' and hidden. I would expect there not to be any SWI's or it would be more useful in a non desktop environment. Can anyone clarify this. |
mrchocky
13/12/05 7:30PM |
"RO5 isn't yet optimised for...". Rather meaningless, sadly. Since "optimised" means the best code ever for the system, that's rarely going to be the case except in very small cases, and hard to prove. So you could say the same thing about any OS/hardware combination, and that's also why compiler engineers prefer the term "code improvement". So, you could certainly improve the code in both RO5 and Adjust for the processors it runs on, but the question is of effort and the kinds of performance improvement you would see from that effort - often a case of diminishing returns.
|
adrianl
13/12/05 10:18PM |
The code is believed to work on later nVIDIA cards, although - as with everything nVIDIA-related - I have no documentation. Geminus does have a SWI interface but it's not yet complete and is thus not documented publically. Originally it was intended only to allow applications such as Cino and PCITV to commandeer extra screens for their own use, but it would be nice to extend the API to make the acceleration directly available to applications where there is no suitable OS API. |
| |
The Iyonix desktop graphics acceleration package Geminus has been 