[Info-vax] Alpha Personal Workstation question

[Paul Sture]

On Fri, 10 Aug 2012 13:35:00 -0600, Keith Parris wrote:

> On 8/9/2012 5:10 PM, Michael Kraemer wrote:
>> Well, considering all the fuss back then about the Alpha being a 64-bit
>> CPU and thus superior to all others, it would have been a much better
>> marketing argument if already entry systems would have allowed to use
>> RAM in excess of 4GB.
> 
> I recall Alpha being promoted because it had the fastest clock rate, but
> it seems to me there were other chips that were 64-bit at the time.

I recall seeing early 1990s benchmarks comparing the Alpha with HP and 
other manufacturers' kit.

In early 1997 the latest greatest Intel offering was 200 MHz (I was 
advised to go for a 2 cpu 133 MHz solution instead on cost grounds and 
was not disappointed).  If I have interpreted the serial number of my
PWS 600au correctly, that was made in early 1997.

DEC could ride on Intel's marketing of clock speeds.  It was certainly 
easy to impress folks who had never heard of the Alpha by quoting clock 
rates.

>> Of course system design was limited by RAM availability and pricing,
> 
> Exactly. Memory wasn't anywhere near as cheap in the early-to-mid '90s
> as today. So to find even an 8 GB system would have been rare, and
> excessively expensive for a single individual's engineering workstation.

Even the servers were limited, and there was a trade off between maximum 
memory and the number of processors you could physically fit.

http://www.icc4it.co.uk/servers/hp_alphaserver/alphaserver-8200.php#tab1

"The AlphaServer 8200 5/300 supports up to 6 processors with 12 GB 
memory, two processors with 28 GB of memory, and any combination between 
these maximums, with 1.2 GB/sec (peak) of I/O."

>> but this means that there was not much differentiation between 32bit
>> and 64bit systems at the bottom line.
> 
> 64 bits applied to several things:
> - Physical address, as you mentioned - Virtual address (here Alpha could
> have way more than 4 GB)
> - Register size (allowing 64-bit arithmetic operations instead of having
> to break them up into multiple 32-bit operations)

Which meant you could compile and test on a workstation and then run on a 
server which did have more memory.

-- 
Paul Sture