[Info-vax] BOINC for VMS

[Jan-Erik Soderholm]

David Froble wrote 2012-04-04 18:28:
> Johnny Billquist wrote:
>> On 2012-04-03 19.10, David Froble wrote:
>>
>> [...]
>>
>>> Myself, (which probably doesn't mean much), I feel that the OoO concept
>>> in Alpha and Power provides a method to get faster processing out of a
>>> set of instructions, if, IF, you can coordinate it well enough. How to
>>> do this? Simple, one word, "money". Takes lots of smart people trying
>>> different ideas, adopting those that work, and setting aside those that
>>> don't work, "today". DEC didn't, or couldn't, continue to fund Alpha. So
>>> far I haven't heard that IBM has taken that route.
>>>
>>> Process shrinks and tons of cache have made the itanic much faster than
>>> the Alphas. That, and some small inter-core communication improvements
>>> are all Intel has been able to manage.
>>
>> I basically agree with all you write. Signals speeds and distances are
>> the limiting factors today, and signal speed is not going to change much.
>> Size are also getting close to their lower limits, which leaves more
>> clever execution.
>>
>> I'm curious though, is really the Itanium that much faster than the Alpha?
>>
>> Johnny
>
>  From my limited exposure, I believe it is.
>
> We have a large number of customers who have moved from PDP-11 to VAX to
> Alpha, and now Itanium. But the CPU is not the only changes over the years.
> Alphas with plenty of memory, and the cacheing that was implemented in
> somewhere around VMS V7.3 were very capable. Some of that was due to the
> cacheing of data. If we would disable the data cache, the systems would
> become slow. The difference was not small, it was major. With a significant
> part of the database in cache, especially the keys, data access was greatly
> enhanced.
>
> So, was the Alphas faster than the VAXs? The cache and larger memory was a
> large part of it, but yes, the later Alphas were faster. We did some CPU
> intensive benchmarks to get an idea how much faster, but I cannot remember
> the numbers, that was 12-13 years ago.
>
> The itaniums that we've put in at the customer sites all have lots of
> memory. Even more of the database ends up sitting in memory. Regardless, we
> can measure greater CPU performance, in the limited testing we've done. I'm
> not the one to answer some of the details, but, I believe EV7 did some
> things that were and are still better than the itanic can do, but that was
> the inter-CPU communications in a multi-processor (like 32 CPUs) system.
> But even there, it isn't the CPU architecture, it's the other stuff on the
> chip, which probably could be done on an itanium chip.
>
> I do remember Bill Todd putting the question to one of the Alpha chip
> engineers, "what is more important for speed, architecture design or
> process (die size)", and the immediate response was "process". So even back
> then, it was basically stated that process shrinks would do more for speed
> than architecture. So what was EV7? 110 nano or even larger? What is the
> current die size for the itanic, and then there is that hugh on-chip cache ...

As I wrote in another port, but quote here for clarity :

Fastest released Alpha was EV7z, 1.3 GHz on a 180 nm process in 2004.
Current IA64 is Tukwila, 1.33-1.73 GHz on a 65 nm process in 2009.

So yes. EV7 was "110 nano or even larger"... :-)

180/65 is aprox 2.8, or 2.8*2.8 = close to *8 times* if we speak
actualy die area difference.

It would surprice a lot if not Alpha whould had performed better on
65 nm then on 180 nm !

EV8 was planned to go up to 2.0 GHz using 125 nm geometry.
IA65 tops at 1.75 using 65...

See tables on :

http://en.wikipedia.org/wiki/DEC_Alpha
http://en.wikipedia.org/wiki/Itanium

Jan-Erik.



>
> So all things being equal, perhaps Alpha would be better than itanium, but,
> all things are not equal. It's a moot point at this point in time. There
> are not and most likely will not be any 35 nanometer Alphas.