[Info-vax] BASIC compiler in the hobbyist distribution

Stephen Hoffman seaohveh at hoffmanlabs.invalid
Sun May 31 11:20:14 EDT 2015


On 2015-05-30 22:46:54 +0000, David Froble said:

> JF Mezei wrote:
>> On 15-05-30 13:34, David Froble wrote:
>> 
>>> The performance difference isn't that great.  Unless you require single 
>>> processor performance, you can usually purchase 2 AMD chips for the 
>>> cost of one Intel chip.
>> 
>> 
>> Sockets on motherboards are expensive, and having multiple sockets has 
>> performance disadvantages with regards to memory access and especially 
>> shared cache between cores on the same chip.
> 
> I'm not everybody.  What I can say is that I doubt I'll ever be able to 
> use that many cores.  Single socket is fine for me.  I think I'm 
> talking for all my customers.  But, never say "never" ....

Where JF is referencing here is that the physical socket, the 
associated components, and the chips with necessary coordination in 
aggregate add to the cost of the box, and can — because you're now 
going off-chip to access the other socket — your performance can suffer 
in comparison to a one-socket multi-core.  What used to be a two-socket 
Xeon (or two-socket AMD, for David) is increasingly now workable with a 
one-socket box.

> 
>> AMD had its 15 secodns of fame when it pushed intel to produce a 64 bit 
>> 8086. But since then, they seem to have languished in the CPU market.
> 
> Seriously, go look at benchmarks.  How large is the gap?  When someone 
> wins at Lemans by a car length, how big is that gap over 24 hours?
> 
> Then look at the prices.  On price / performance, AMD is usually far 
> ahead.  True, they have to keep prices low.  But isn't that what we've 
> all asked for for many years?
> 
> Just an example:
> 
> AMD Opteron 6320 Abu Dhabi 2.8GHz 8MB L2 Cache 16MB L3 Cache   $290
> 
> Intel Xeon E5-2630 v3 Haswell-EP 2.4GHz 8 x 256KB L2 Cache 20MB L3 
> Cache LGA 2011-3 85W BX80644E52630V3 Server Processor  $680
> 
> Two 8 core chips.  Neither top of the line.  But the Intel chip is over 
> twice the cost of the AMD chip.  Do you really think it has more than 
> twice the performance?
> 
> No, I didn't spend the time to get actual benchmark figures.

Based on what I'm seeing from a quick look around, the one-socket Intel 
is more than twice as fast as a one-socket AMD Abu Dhabi and is 
somewhat faster than the two-socket version, but sorting that out is 
not at all clear.  Without access to the costs for the differences 
between the one-socket and the two-socket boards, I don't know which 
configuration would be cheaper to build and sell and buy here.  The 
aggregate cost of the Intel chip and the one-socket board design is 
likely a little more expensive in aggregate than the brace of AMD chips 
and the two-socket board, but that's a guess entirely based on the 
likely AMD pricing delta.

FWIW and as David is probably aware, AMD has been hemorrhaging cash in 
some recent years, and has been restructuring and has made very 
substantial cuts in R&D and other lines in response to their recent 
losses 
<http://www.investopedia.com/stock-analysis/051515/3-key-challenges-amds-turnaround-amd.aspx>, 
and they spun off their manufacturing as GlobalFoundaries — like ARM, 
AMD is now a chip-design house.   This current financial and product 
predicament means AMD can presently only compete on price, and at the 
low-end, and which depresses the AMD margins.

The POWER and SPARC servers — and Itanium for that matter, at least 
through Kittson — do appear to be providing a price umbrella that helps 
support the higher-end Intel x86-64 product pricing, but I digress.

Now whether AMD can get back into the x86-64 server market with newer 
designs, and where there's potentially more money to be made there than 
in gazillion-unit-volume IoT and the low-end desktop and commodity 
x86-64?  Or via their ATI chips, as have recently found their way into 
the newest MacBook Pro products?  Donno.

AMD also bought an ARM server vendor a few years ago, which probably 
led to 
<http://www.anandtech.com/show/8362/amds-big-bet-on-arm-powered-servers-a1100-revealed> 
at al.  Not that there haven't been some detractors there, too 
<http://www.fool.com/investing/general/2014/01/29/dont-buy-amds-arm-server-hype.aspx>. 


It'd be interesting to have some more serious high-performance 
competition for Intel, as right now that competition looks to be 
somewhere between lacking — and the POWER and SPARC sales trends are 
headed in the wrong direction — and asymmetric competition for Intel 
isn't there yet — folks using ARM designs in increasingly huge numbers, 
but those look to be displacing low-end, low-performing x86 designs 
based mostly on price and the existing ARM installed base, and that's 
probably mostly based on Intel's preferred margins for x86 — not that 
Intel hasn't been dropping their low-end prices 
<http://www.computerworld.com/article/2896588/intels-surround-and-conquer-iot-strategy.html> 
or (2013) 
<http://www.extremetech.com/computing/166365-intels-new-quark-cpu-core-is-the-start-of-a-new-arm-like-foundry-model> 
or low-end <https://en.wikipedia.org/wiki/Intel_Edison> 
<https://en.wikipedia.org/wiki/Intel_Galileo> boards.  There does not 
seem to be much direct competition for Intel x86-64 from AMD for the 
high-end, though.  At least not until and unless AMD gets their 
high-end offerings sorted.  (But then how big is the x86-64 high-end 
market these days?  HPE folks report 80% of their server sales are 
two-socket boxes, and (my speculation) that trend may well be headed 
toward one-socket designs with the proliferation of cores and with GPUs 
configured for those that need that sort of computing.)

As for OpenVMS servers on x86-64 post-port, I'd tend to expect we'll 
see specific HPE ProLiant Xeon configurations supported, specific 
Nemonix Xeon configurations, and (maybe) a few other configurations 
with third-party testing and support for OpenVMS.   I can hope that 
there might be a few Apple hardware and VM-based configurations, too.  
But something akin to 
<https://en.wikipedia.org/wiki/Apple–Intel_architecture#Dont_Steal_Mac_OS_X.kext> 
wouldn't be a huge surprise as a business model; a "DSVMS" hardware 
tie-in.

Interesting times.


-- 
Pure Personal Opinion | HoffmanLabs LLC




More information about the Info-vax mailing list