~Yes I am a Kiwi, No, I did not appear as an extra in 'Lord of the Rings'~ 1x AmigaOne X5000 2.0GHz 2gM RadeonR9280X AOS4.x 3x AmigaOne X1000 1.8GHz 2gM RadeonHD7970 AOS4.x
~Yes I am a Kiwi, No, I did not appear as an extra in 'Lord of the Rings'~ 1x AmigaOne X5000 2.0GHz 2gM RadeonR9280X AOS4.x 3x AmigaOne X1000 1.8GHz 2gM RadeonHD7970 AOS4.x
~Yes I am a Kiwi, No, I did not appear as an extra in 'Lord of the Rings'~ 1x AmigaOne X5000 2.0GHz 2gM RadeonR9280X AOS4.x 3x AmigaOne X1000 1.8GHz 2gM RadeonHD7970 AOS4.x
Do we really need boards with slow cpu's ? If they want more interested people do a more modern board for god sake. Dont devs listen when people says... its slow... ill buy if there is something faster then A1G4Xe etc etc.
I believe the more important today than CPU power, is the PCI buss be changed to PCIe and a CPU slot in the board.
If any company make a board with a CPU with the 604e Power it is ok, if it have the option of a CPU slot to the people use more one or 2 CPU in the bard plus the original, and if it have a actual PCIe interface to the people use actual and cheap GPUs.
like the classic amiga, the 68020 from a1200 is very slow, the people buyed it think in put a 68040 or 68060 in the past.
I believe the more important today than CPU power, is the PCI buss be changed to PCIe and a CPU slot in the board.
I agree that we need to be able to use modern gfx cards, a very important point.
Quote:
If any company make a board with a CPU with the 604e Power it is ok, if it have the option of a CPU slot to the people use more one or 2 CPU in the bard plus the original, and if it have a actual PCIe interface to the people use actual and cheap GPUs. like the classic amiga, the 68020 from a1200 is very slow, the people buyed it think in put a 68040 or 68060 in the past.
Cpu slots are a nice idea, but they are expensive to make and if you sell for example 1000 basis boards with slower CPU how many do you think can afford to upgrade? If only 300 will buy a CPU card/accelerator it would be way to expensive.
I agree on the gfx but for CPU (and IMHO even RAM) developers have to make a decision and just solder the bloody thing onboard to keep prices as low as possible.
for example, if a company make a board like the sam440, with this CPU power. it is Ok for 90% of the users. but if it have plus a CPU slot, it will be fantastic, becouse the PPL will can instal powerfull CPU to professional use, like render or video edition.
It can be a PCIex16 slot., for example, a board with 3 PCIex16, 1 you use for CPU up, 2 to GPU, or 2 for CPU and 1 to gPU.
it will be fantastic, and I believe not much more expencive than one actual sam440.
The CPU slot can use until a x86 base CPU, it is not primary, the OS need just know how to work with the secundary CPU, like the IBM worstations using X80i Cell CPU more opteron CPU together.
The problem with adding a connector between the motherboard and the CPU is that it spoils the direct connection between the CPU and everything else on the motherboard. It immediately limits the speed of data transfers and hence CPU clock speeds. Look at the connector used on the old A4000 A3640 CPU and compare it with the connector on the A1 XE CPU modules.
The A1 XE motherboard-connector-CPU chain is pushing it to get up to a Gig (it's rated at 800 MHz). Whatever connection mechanism is used, there is a substantial cost increase (the engineering work required to design and debug all those critical signal paths) and ultimately a much higher cost to the customer.
I'm very pro-modular. I'd do things similar to your proposal. For how I'd go about it though, it wouldn't be a slot per se, mostly due to connector style of the standard I like. It'd be a module, but lying parallel to motherboard in my ideal design, much like accelerators for A3000/A4000. Not because I want it that way, but because a popular standard I think suitable just already is that way.
One benefit of modular is that you do the hard stuff once, on the module. The motherboard then becomes much simpler, and thus cheaper. And easier to pop out a few different form factors, make an AGP and a MiniITX and a format-X motherboard much cheaper and faster than making complete motherboard designs individually containing everything that could go onto a common module. Granted, if there's only one module, then there's only one performance level. Maketwomodules and 3 carriers, and then you have six configurations but only have to do the hard stuff twice...
The connector does not have anything running at 800MHz or 1GHz. A1XE memory is 133MHz is it not? The CPU to northbridge bus I think topped out around 200MHz, and I don't think anything on A1XE goes that high.
The 800MHz or 1GHz etc. are derived on the CPU module (inside the CPU?) by multiplying the northbridge bus clock frequency by some selectable number.
You could put a faster CPU on an A1XE module, nobody ever did due to cost and market size. Eyetech slowed down their 1GHz chips due to insufficient heatsink choice, a bigger heatsink would have allowed 1GHz operation. Perhaps some PCB design issue as well, but a good design would not be limited by the PCB for this. No idea if Eyetech/MAI/whoever did a bad job or not, it may be perfectly fine in this context.
the supercomputer that I post a link use 2 PCIex8 links to the boards tranfer data.
and if you get a look in one usual today video card, one HD 4890 for example process more than 1.6 Teraflop of computer power with a PCIeX16 slot linking to the PC CPU and more than 125gb/s
It is a CPU if you thing about it, just dedicated to process graphics.
In the same way a CPU to other use can be linked too using the PCIe conection to the board CPU.
It is just question to make a driver to the OS can see and use this new card.
The CPU connector may not be running at 200MHz but it is rated at 200MHz.
Think of it this way - if you compare pure copper with very old, very impure steel wire, neither has a clock rate but you'd be hard pushed to get anything like as fast a reliable transmission of data with the old steel as with the new copper, would you? In the same way, when you're routing a mobo for super-fast speeds, you need to take timing and trace lengths into consideration. Look at a DDR module on a PC mobo - you'll often find the traces aren't straight, but are bent in little loops and spirals. This is to keep the signal paths lengths the same between all the signals between the northbridge and the RAM.
The A1 CPU slot was never designed for >200MHz speeds, if you get more than that it's just a lucky bonus, but we can't rely on it.
[Edit: The CPU slot rated at 200MHz, I put 800MHz originally but of course it's clock-quadrupled. Same applies though!]
And this is good comparison ;) For Mulder some people also created a hoax to make him believe it even more. This is exactly the same... Make believe the hoax ;)