At 06:03 PM 8/27/99 , Lord Isildur wrote:
>well, if we can get a few people and all the right expertise, i very much 
>want to produce or at least begin the design of a VAX compatible machine. 
>I still want to be ambitious and start with completely fresh designs  as 
>far as peripheral and so on connects go, and the like, and id like to 
>keep as true to the form and beauty of the VAX architecture as possible. 

Unless you can get pry AXE (the VAX validation tool) out of Compaq, it 
might be difficult to prove that it really is a VAX.  There are a whole 
lot of really obscure conditions that any implementation has to handle.

Think of doing a CALLS over a page boundary which really crosses a 
128 page boundary and the page of that page's PTE is not valid (paged out).

>As for web pages, i think perhaps int he beginnign we should just have 
>something more or less statuic that announces that were producing a VAX 
>clone machine designed to be 100% bit-compatible with a 'real' VAX 
>without infringing on any of DEC's technology ownerships, and that we aim 
>to produce these machines at competetive prices and performances on a par 
>with other modern machines. 
>I must say that though i want to see a completely native and custom 
>implementation, the first go at  it might be better done using somethign 
>like strongARM chips or something. There will probably be generous use of 
>gate arrays here. 
>  FPGA's are not cheap. One with several thousand gates will run you in 
>the realm of US$40-60 depending on speed. 
>I dont anticipate that the production cost of a machine we could make 
>would be below US$5000, and it will probably be more in the realm of US$7000
>and a very very primitive guess at what it woudl take. This doesnt 
>include what the develipment, prototyping, testing, and debugging would 
>cost. 
>I want to make the machine modular, with each module being a VAX nexus 
>like in the big VAXen (7x0's, 6000s, 8000s) and the interface being as 
>all nexi speak, nothing but virtual,paged, scatter/gather DMA. Each 
>device will have to manipulate VAX page tables and use a nontrivial 
>amount of circuitry. This might be best done with a microprocessor and 
>PROM, though i think a couple of gate arrays would be better or cleaner. 
>The cheap shot way to do it all would be to make a bus adapter which is a 
>nexus, and then hang devices off ot it which are not. Not as beautiful 
>and doesnt take full advantage of the true power of the VAX architecture, 
>which is that wonderfully equitable 'conference' of edevices, all of 
>which are euqally powerful citizens of a virtual memory system, all of 
>which manipulate page atbles and memory in the same way, some of which 
>are processors but all of which can be masters or slaves. 
>We might also consider using NVAX processors for our first machine. They 
>are still (very expensive i would believe) in existence and available in 
>_some_ quantity. I  dont know enough about this to be anywhere near 
>authoritative though. 

If I was doing this, I'd use an existing Pentium socket (such as socket 7 
or 370).  No support chips to design.  Relatively known problem.  But
will sacrifice some on performance.

>One thing i think we are gonna have to think about real hard is how on 
>earth would we get around the 512 MB memory limit!?? this modern world 
>wont tolerate it! 

The VAX 9000 already got around it since it used extended PTEs.  Anything
we do should use the same design.

>  i want to ship a VAX with 4 GB of fully populated address space which 
>will still support a program written for an 11/780!
>well, i want to be able to run anythign that will run on an 11/780, which 
>is _the_ reference machine for this, being the first VAX and all. :)

Do you want to boot VAX/VMS on it?  Will you write the SYSLOA for the machine?


-- 
Matt Thomas               Internet:   matt@3am-software.com
3am Software Foundry      WWW URL:    http://www.3am-software.com/bio/matt/
Cupertino, CA             Disclaimer: I avow all knowledge of this message