Amiga.org
Amiga News and Community Announcements => Amiga News and Community Announcements => Amiga Hardware News => Topic started by: Kees on October 08, 2003, 09:36:49 PM
-
The specs of the Pegasos 2 have been made public at IBM's Site (http://www.developer.ibm.com/solutions/isv/igssg.nsf/list/bycompanyname/86256B7B0003EBBF86256DB7006F37A8?OpenDocument).
"Genesi's new hardware platform is called Pegasos. Genesi's engineers have created an open hardware platform based on the CHRP motherboard standard for the PowerPC and selected Open Firmware so that many operating systems can work easily on the Pegasos platform. "
Click on 'functional discription' to see the specs or read below.
MicroATX motherboard
Marvell Discovery II System Controller (MV64361)
VIA peripheral controller (VT8231)
CPU on daughter card
Supports PowerPC CPUs from IBM and Motorola
DDR RAM PC 2100 / DDR266
2X Ethernet ports (1x1Gbit and 1x10/100Mb)
3 X Firewire USB 1.1
2 X ATA 100
AC97 Audio (Line In / Line Out / Headphone Out)
SPDIF Out
AGP x1 port
3 X PCI slots
Riser connector for 1st PCI slot
-
Excellent bit of publicity.
:pint:
-
What I wanna know is what speed is the AGP port? :-?
-
Um, looks like 1x speed. I assume we'll have a clarification or comment (ie. damage control) soon from BBRV.
USB 2.0 would have been nice.
-
I'd ssume that it is at least a 2x-port. ( I doubt that P2 will have worse specifications than it's predecessor)
-
It's 1x speed. (Edit: or at least looks very like it)
Some interesting statements on Ann. BBRV puts a nice point across that 1x is good enough and that the bus doesn't really matter any more. However, somebody also says that ATI's cards only work on 2x...
Whabang, the reason Peg-1 had AGP 2x was Articia. However, bPlan dropped this chip, labelling it with the nice title "a piece of buggy crap" and went for Marvell instead. This has no AGP and so it had to be added with logic. This would be why the AGP in Peg-2 is 1x.
-
As posted on ANN:
Unless there has been some horrible miss-communication yes it's a x1 (speed) AGP slot.
However there is always a twist in the tail...
This AGP x1 is no slower than the AGP x2 in the Pegasos 1 for the simple fact that the Articia cannot read from memory at AGP x2 speeds, in fact the fastest we've measured is lower than AGP x1, The G4 is faster but not significantly so.
However if you know anything about graphics or the history of AGP slots you'll also know it's unlikely to make any difference whatsoever.
AGP was originally invented because video memory at the time was highly expensive and the card firms couldn't more than 4MB on a card as a result. As games became more complex they needed more textures and these would not fit into 4MB so AGP was added as a fast path to main memory to fetch textures. Fast forward to now and video memory is dirt cheap in comparison and the textures are held on the card with essentially no need for AGP.
I doubt it makes any difference in FPS at all as the most intensive part of the graphics pipeline, rasterisation (i.e. drawing) is done on the card itself, this doesn't go anywhere near the AGP bus.
The only application I know of which actually requires huge bandwidth to a graphics card is Raw HDTV playback at 140MB / Second, AGP x1 can handle that but I doubt many Pegasos owners have the large SCSI array required to play it...
--
Regarding the other specs there's nothing anyone should be surprised about, changing anything else other than the NorthBridge would have delayed the Pegasos II by months and probably put the cost up.
--
Nicholas Blachford
Genesi France.
-
From ATI's site:
RAGE 128 PRO, RADEON, and newer products are keyed "Universal AGP" and may operate in AGP 1.0 and AGP 2.0 compliant motherboards. RAGE 128 PRO, RADEON, and newer products are keyed "Universal AGP" and may operate in AGP 1.0 and AGP 2.0 compliant motherboards.
That can be found here (http://www.ati.com/support/faq/agpchart.html).
Remember that AGP2.0 is not 2x. I can't find anywhere that says Radeon won't work on 1x.
Radeons will continue to work on Pegasos-2, and the bus will be faster on Peg-2 than it was on Peg-1, despite the stated 2x bus.
-
There are a couple of things regarding the AGP port that no-one seems to have picked up on:
First, what is the signalling voltage used on the AGP port? Real AGP 1x/v1.0 ports use 3.3v, which is incompatible with most new video cards as they need 1.5v. (this is the opposite of the voltage issue that stops AGP 2x cards working in 8x slots).
Second, is it really an AGP port or just a 66MHz PCI bus slot modified with a AGP connector and some kludged logic to cope with AGP cards? If it is then AGP features like DMA, Sidebanding and FastWrites will not work, or will be shaky. Anyone who remembers the AGP problems on the old MVP3 and Aladdin 5 PC chipsets will know what kludged AGP port logic can do to a system's stability.
If both of these issues affect the Peg2's AGP port then it'll never run anything more advanced than a Voodoo 3 reliably :-(
I hope Genesi can offer some reassurance than the Peg2 can run AGP 4x and 8x cards. If they can (and if OS4 gets ported to it) I'd buy one like a shot. :-)
-
Um, looks like 1x speed. I assume we'll have a clarification or comment (ie. damage control) soon from BBRV.
As "damage control" it'd be sufficient to say what can we expect compared to a Pegasos I. What is better (and if there is anything) what is worse.
The news item appeared pretty unexpectedly, and quite late in the night (european time), and that gave place to the wildest speculations.
-
AGP was originally invented because video memory at the time was highly expensive and the card firms couldn't more than 4MB on a card as a result. As games became more complex they needed more textures and these would not fit into 4MB so AGP was added as a fast path to main memory to fetch textures. Fast forward to now and video memory is dirt cheap in comparison and the textures are held on the card with essentially no need for AGP.
It does make a difference when the game exceeds the Video card’s memory i.e. UT2003 (details at least @ 1024x768 + “holy $hit” mode + 16 players) with 2019 custom map (very detailed Blade Runner style city map). This scenario exceeds NV25’s 64MB @DDR512 ram. AGP Graphic Aperture was set to 256MB (from 128MB). Main memory was PC3200 1GB(dual channel) with DDR400 FSB.
With Amiga’s games market, this issue may not be as important (due to the game titles with that kind of requirements does not exist). It would be a completely waste of time and money IF the AGP8X doesn’t boost some aspects of the heavy games titles.
-
Well, speed and the number of ports are two different things, but anyway the performance comes out of the GFX card. It would be the absolutely technical overkill from a component requirement (and cost) to have more. Today's graphics cards do not use the AGP bus to move texture data to the graphics core. The textures are stored inside the graphics card RAM. The only data that is transferred to the card are the coordinates for the graphic core. For this task the current bus speed (AGPx1 = PCI66) is ok. This bus can transfer 266MB/s, that is about 40M coordinates for textured triangles. With a estimated size of 20 pixel per triangle you can draw 800M pixel/s on the screen. With a resolution of 1600x1200 this is about 400 frames per second. You will NOT find a monitor that is capable of doing this.
The initial idea of the AGP bus was using the main memory as graphics RAM to save some cost. As today's performance expectations are going much behind the limit of this approach, the cards are using their own local memory.
More fun news coming from IBM soon...;-)
R&B :-)
-
Will it have the same Graphic Card compatibility as the Pegasos I? I suppose a simple YES wil settle half of the worrying.
EDIT
By the way, out of interest:
1) Does Peg2 allow me to use various Radeon's as in Peg1, or will there be problems?
2) Will drivers for more modern Radeon's be implemented, or are we stuck to 9000 and less?
As Hooligan summarized nicely the uncertain issues raised on ANN.
-
Today's graphics cards do not use the AGP bus to move texture data to the graphics core.
The need for "AGP Graphics Aperure Size" indicates otherwise...
References
http://www.tweak3d.net/articles/aperture-size/
http://support.gateway.com/s/MOTHERBD/shared/agpadap.shtml
With a estimated size of 20 pixel per triangle you can draw 800M pixel/s on the screen
That’s about Geforce 2 GTS level i.e. +1G pixel/s claim to fame.
-
After a first look to the Specs I am a bit disappointed by the external interfaces...
"AC97 Audio (Line In / Line Out / Headphone Out)"
Not even a Microphone-In? Thats strange! There must be more input-possibilities!
"SPDIF Out "
No SPDIF In? Why?
regards selco, http://Selco.da.ru
-
This bus can transfer 266MB/s, that is about 40M coordinates for textured triangles. With a estimated size of 20 pixel per triangle you can draw 800M pixel/s on the screen
Guess that rules out me using my 3000M pixel/s and 70M polygon capable Radeon 9700pro then :-(
-
>Guess that rules out me using my 3000M pixel/s
>and 70M polygon capable Radeon 9700pro then
No, the fact that there are no drivers for the Radeon
9600+ rules you out.
PS: The currently supported graphics cards for
MorphOS have been tested in a AGP x1 environment
and are working.
-
PS: The currently supported graphics cards for
MorphOS have been tested in a AGP x1 environment
and are working.
Now that quite settles the question. Thanks for the clarification.
-
Well, speed and the number of ports are two different things, but anyway the performance comes out of the GFX card. It would be the absolutely technical overkill from a component requirement (and cost) to have more. Today's graphics cards do not use the AGP bus to move texture data to the graphics core. The textures are stored inside the graphics card RAM. The only data that is transferred to the card are the coordinates for the graphic core. For this task the current bus speed (AGPx1 = PCI66) is ok. This bus can transfer 266MB/s, that is about 40M coordinates for textured triangles. With a estimated size of 20 pixel per triangle you can draw 800M pixel/s on the screen. With a resolution of 1600x1200 this is about 400 frames per second. You will NOT find a monitor that is capable of doing this.
This is only half the truth. You forget that you actually have to get the textures uploaded to the graphics card memory. As Hammer points out there are many games today that use many high-res textures in a scene. Add to that that many newer games use mulitexturing (i.e. several layers of textures on the same polygon to achive e.g lighting and bumpmapping effects). So there's a whole lot of texture data that needs to be transfered to the memory for a complex 3D-scene.
-
So there's a whole lot of texture data that needs to be transfered to the memory for a complex 3D-scene.
And where are they loaded from ?
HD, CDROM ?
The best hard drives (SCSI Ultra 320 15000 RP) hardly support 80 MB/s. AGP * 1 is 256 MB/s.
Texture may be compressed so you could eventualy get higher speed than your CD/HD. Still 256 MB offers a good margin.
-
If Genesi uses the MV64361 controller then why do the
only have one Gb ethernet port, and why no PCI-X slots?
Or does the MV64361 only have one PCI-X controller?
And wasn't it supposed to have a 200Mhz fsb and a DDR400 controller?
I'm guessing that this is due to cost considerations,
which is nice in a way but it has put it off my list of desireable hardware.
The sad part (for me) is that I was looking forward to buying it if
Genesi could get OS4 ported to it, now I guess that I'll
have to wait untill the A1-light is released to get a nice
speced and priced mobo
(or the peg-III if OS4 will be available and the price is right).
------------------------------------------------------------------------------------------------------------------
From: http://www.marvell.com/products/communication/discoveryII/MV64360.jsp
The Discovery II MV64361 controller offers a 72-bit DDR
memory controller with a 183 MHz clock rate (366 MHz data rate),
on-board 2 Megabits SRAM, dual 32-bit PCI/PCI-X interfaces,
PCI bridge and arbiter, two 10/100/1000 Mbps Ethernet controllers,
two multi-protocol serial channels, and TWSI and interrupt controllers.
Key Features
High-performance controller for PowerPC-based communications systems
* 64-bit 133 MHz 60X/MPC CPU bus interface
* 72-bit (64-bit with 8-bit ECC) 183 MHz DDR SDRAM controller
* Dual 64-bit 66 MHz PCI /133 MHz PCI-X interfaces (for MV64360)
= >* Dual 32-bit 66 Mhz PC/133 Mhz PCI-X Interfaces (for MV64361)
* Single 64-bit 66 Mhz PCI/133 MHz PCI-X interface (for MV64362)
* Advanced internal crossbar fabric
* 32-bit 133 MHz peripheral device bus interface
Integrated 2 Megabits SRAM (MV64360 and MV64361 only)
Advanced communications unit
* One 10/100/1000 Mbps Ethernet controllers with packet filtering and priority queuing (for MV64362)
= >* Two 10/100/1000 Mbps Ethernet controllers with packet filtering and priority queuing (for MV64361)
* Three 10/100/1000 Mbps Ethernet controllers with packet filtering and priority queuing (for MV64360)
* Two Multi-Protocol Serial Controllers (MPSC)
64-bit PowerPC CPU bus interface
* Motorola MPC750, 755, 74xx and IBM PPC750, 750CXe, 750FX processor support
* 60x and MPX bus protocol support
* 133 MHz CPU bus frequency (2.5v or 3.3v configuration)
* Configurable cache coherency
* Supports split-read transactions with out-of-order completion
-
>And wasn't it supposed to have a 200Mhz fsb and a
>DDR400 controller?
That's Díscovery III (MV6446x).
-
If Genesi uses the MV64361 controller then why do the only have one Gb ethernet port
There's only enough space on the back for a certain number of slots and sockets. There's the original VIA Rhine ethernet device there for compatibility with everything else.
and why no PCI-X slots?
Do you know where to buy any?
If the answer to any of those is "no", then this is why you do not need PCI-X slots, and why they are not provided. There is also the issue that PCI-X slots are longer than PCI slots, and there isn't enough space on the board. Anyway. You don't need them.
~~
As for other stuff: AGP 1x is fine.
There is no game on the planet that does a forced AGP transfer for texturing for every frame, unless it is starved of video memory. With a 64MB card, this doesn't happen very often.
The same can be said of vertex buffer objects, which means that certain transformable vertices (i.e. raw data to be manipulated by the tnl engine) can be stored in video memory anyway, saving transfers for whole swathes of vertex data.
A PCI Radeon or GeForce 4 card performs identically to the AGP versions, apart from initial texture loading (maybe at the beginning of a level.. that progress bar will stay for longer) and in starved situations.
Quite rare these days unless you're running 32bit 1600x1200, 32bit Z-Buffer, 8-bit Stencil buffer, 4xFSAA, and have 60MB of raw texture data.
There are texture compression techniques which can be employed, Z buffer compression techniques, and many other things, which means most AGP-capable cards rarely touch the AGP bus because in fact it is highly discouraged :)
AGP 8x and so on is a pandering to the onboard chip manufacturer, where video RAM *is* system RAM and is *always* transferred across the AGP bus. If you noticed, these chips allocate some pitiful 16MB (4, 8 are not uncommon either), and are also very common on most motherboards anyway.
With 133MHz DDR (i.e. DDR266, PC2100) memory you can saturate your AGP bus completely with those rates. Now that most PC motherboards come with DDR400 memory capability, dual channel options, AGP becomes a slow tortoise to the memory's rabbit.
Slowly but surely, it gets there. Since upping to 8x is easy and gives Intel and SiS and VIA something to increase the speed of their commodity internal graphics architectures, they did it.
That doesn't mean it makes a great deal of difference to your Radeon. As such, it makes barely a difference to the Pegasos.
-
Do you know where to buy any?
Yes.
There is also the issue that PCI-X slots are longer than PCI slots, and there isn't enough space on the board. Anyway. You don't need them.
The space requirements I can understand and respect but why do you feel that you can tell me what I need and do not?
Anyway good luck to you with the sales of the peg-II
and the design work for the peg-III(please try and get OS4 ported to it). :-)
-
Poster: bbrv Date: 2003/10/9 4:45:10
Question for you BBRV: If AGP 1x is good enough, why not have the gfx onboard? I realize that your pressed for time and doing minimal differences between Peg1 and Peg2 for cost/time savings, but could future versions of the Peg2 have gfx onboard? Atleast that would end the bitching coming from the Red Trolls for awhile. ;)
Dammy
-
@dammy:
If AGP 1x is good enough, why not have the gfx onboard?
I guess Neko's reply explained that. Onboard gfx would share the system RAM, making all GPU <-> RAM transfers happen across that "slow" AGP bus.
That's one reason, at least with regards to AGP speed.
Edit:
I understand there are onboard video chipsets that have built-in video RAM. How much built-in RAM is common?
Are there any mobos with onboard video that also have physically separate and fast onboard video RAM?
-
8x is faster than 1x....its the purpose of the whole numbering system, to give you a quick idea about the bus capacity.
but bbrv is correct, its not the issue. there are few drivers for cards....there are few games to run on those cards....
people are going to run into many issues before they ever run into an issue about an agp bus bottleneck...
the most important thing, was having an AGP slot, because PCI cards are getting rare and limit your choices even more.
-
Poster: Seehund Date: 2003/10/9 10:28:42
I guess Neko's reply explained that. Onboard gfx would share the system RAM, making all GPU <-> RAM transfers happen across that "slow" AGP bus.
That's one reason, at least with regards to AGP speed.
But how many games are out there for MOS or Linux that actually run slow because of the shared bus? I really don't see that as a big minus when it could be freeing up a slot.
Dammy
-
@Dammy
I see what you mean.
Personally, in an either-or situation, I'd prefer an AGP slot over on-board graphics. That's just my personal preference, but since the PegII is supposed to be a "geek" mobo I guess it's also a general preference on its intended market. Pick and mix. It's got most things people need/expect onboard already, so 3 PCI slots on a micro-ATX mobo sounds OK to me.
-
@Neko
AGP 8x and so on is a pandering to the onboard chip manufacturer, where video RAM *is* system RAM and is *always* transferred across the AGP bus. If you noticed, these chips allocate some pitiful 16MB (4, 8 are not uncommon either), and are also very common on most motherboards anyway.
I'm no motherboard engineer but it seems to me that the nVidia nForce 2 motherboards with onboard GeForce 4 graphics don't work this way. It uses the systembus for graphics and uses separate memory controllers for the graphics subsystem and the CPU which both can use the bus at the same time. They're also not limited to 16Mb graphics memory. My AOpen board lets me set the amount of memory to reserve for graphics in the BIOS. The default setting is 64MB, but I can set it to 128Mb at least. I only have 256Mb in that machine, so I can't go higher because there would be no system memory left so I don't know what the actual upper limit is, if there is one.
-
AND, now MorphOS too!!!! (https://www.developer.ibm.com/solutions/isv/admin.nsf/allsolutionsunid/2D3016E6E1BE730886256DB900575117?opendocument&s=8E4ABAE68A428F3D86256DBA0058119D&id=408097)
:-) :-o :-D 8-) :-D ;-)
-
Poster: bbrv Date: 2003/10/9 12:29:51
AND, now MorphOS too!!!!
Congrats!
Dammy
-
Hi Dammy, thanks!
What is really good is that when other operating systems run on the Pegasos we can submit them for the logo too for their packaging, websites, etc.
(http://ascii24.com/news/i/mrkt/article/2002/06/26/thumbnail/thumb220x67-images689100.jpg)
All the more reason to get up sites like THIS (http://odc.pegasosppc.com/) one for OpenBSD ready for the Pegasos. BTW, there will be an OpenBSD distro ready for the Pegasos II launch also.
MorphOS, PegXLin, and OpenBSD for the Pegasos II!
:-D
R&B
-
P.S. Think the AROS Team is more interested now?!
Please! :-D
...or even OS4?!?!?! 8-)
-
Poster: bbrv Date: 2003/10/9 13:10:29
P.S. Think the AROS Team is more interested now?!
Please!
AROS has one Dev for PPC, and from the last I talked to him, he's seperated from his Peg for a few weeks due to school. The rest of the main devs are also college students so developing slows down during the school year. Pity funding was so skimpy this summer for teamAROS bounties, I was hoping some decent cash would have been an incentive to get things done prior to school starting again when developing takes back seat to passing tough college courses.
Dammy
-
Hmmm, I was expecting, considering the hype, that the spec would make me say "That kicks ass compared to the Peg 1" what I got was "thats so undewhelming"
As for the rest of the comments....welcome to the mutal admiration society for the hopelessly biased:)
-
Well.. Pega 2 spces are about the same I predicted on local User group
magazine last spring. So it's more kind of evolution.. Not Revolution.
-
Good news! :)
When do we get the pics?
-
I'm no motherboard engineer but it seems to me that the nVidia nForce 2 motherboards with onboard GeForce 4 graphics don't work this way
nVidia's solution is a custom chip that has the same capabilities of the GeForce 4, and uses a custom bus controller (some Hypertransport variant) to do memory access. This is why it's so "fast".
What I was talking about was your plain ordinary common-or-garden embedded chipset - like an Intel i845 or VIA CL266 with that "Chrome" graphics. Not a fully fledged GPU, but just a quick fix to get video onto the motherboard - not high end, not really suitable for games (even the i845 plays Quake III poorly.. it's bolstered by the fact that the minimum speed CPU to use with it is some 2.4GHz)
The Intel variants of these chips used to - I don't know if they still do - come with the option of a special "cache" AGP card which was basically faster memory so that it didn't have to shuffle around using slow system RAM (this was in the days of PC133 SDRAM, perhaps DDR & huge FSB speeds have negated the need for it).
As it stands, anyway, you can't get around the fact that you don't need AGP faster than 1x with a decent graphics card. With an 8MB Permedia card, you're going to be in trouble. With a Radeon 8500 at 64MB average memory, you won't.
As for Eric_Z, what on earth are you going to use PCI-X for? I'm telling you that you don't need PCI-X cards in a Pegasos because that's not the market (huge servers and high-performance dual processor workstations, like the PowerMac G5). Most people can't afford PCI-X cards, if they even exist for the purposes people think they want them for, and don't need the added performance.
-
Like I tried to say before, I can understand that you have to
make design desitions based on what will attract the larges
customer base and bring in the largest profit.
This does not, however, mean that ALL people will think of the
lack of PCI-X as good thing. Even thought (in this techno-nerd communety)
most people will undoubtedly not care,
there will be people who do, ie me.
Why am I interested in PCI-X you ask, well it's mostly due
to some crazy "server - modding" ideas that I had.
And before anybody point out the obvious, one word,
dual boot.
http://www.dssnetworks.com/v3/gigabit_pci_6162.asp
http://www.emulex.com/products/fc/index.html
-
@neko
Thank you for your insightful answers, you seem to know your motherboard electronics. =)
As it stands, anyway, you can't get around the fact that you don't need AGP faster than 1x with a decent graphics card. With an 8MB Permedia card, you're going to be in trouble. With a Radeon 8500 at 64MB average memory, you won't.
Well, the whole thing is largely academic anyway, since MorphOS (or any of the other OSes is supports) in the foreseeable future is unlikely to get any games that are at such a level where this could be an issue anyway. It took Valve more than 5 years to create Half-Life 2, so well probably see Pegasos III, IV and V before such a game would become available for MorpOS.
-
They do, try to overclock an NForce2 Northbridge with the GF4 chipset at the same rates that the
same chip without the gfx chipset can be overclock
and see what happens. The gfx chip will be choked.
If you try to push the system RAM, the on board gfx system will fail to work.