Post Reply 
For those interested in a Prime "Benchmark"...
11-28-2017, 11:30 PM (This post was last modified: 11-28-2017 11:31 PM by zeno333.)
Post: #1
For those interested in a Prime "Benchmark"...
I found a web page that mentions the speed of the CPU in the HP prime in MIPS, and given that and using the clock of the Prime at 400 MHZ, the Prime is a 440 MIPS machine...pretty fast. Too bad there is not a program for the Prime equivalent to the LINPACK that tests FLOPS to compare it using Floating Point Operations.
Find all posts by this user
Quote this message in a reply
11-29-2017, 05:04 AM
Post: #2
RE: For those interested in a Prime "Benchmark"...
(11-28-2017 11:30 PM)zeno333 Wrote:  I found a web page that mentions the speed of the CPU in the HP prime in MIPS, and given that and using the clock of the Prime at 400 MHZ, the Prime is a 440 MIPS machine...pretty fast. Too bad there is not a program for the Prime equivalent to the LINPACK that tests FLOPS to compare it using Floating Point Operations.

Don't believe everything you read. The CPU in the Prime can at most and only under certain specific circumstances execute 1 instruction per cycle. Therefore at 400 MHz the theoretical limit is 400 MIPS. In practice it is lower, since the 1 instruction per cycle is for many instructions (not all) that must be unrelated (in this context related means one instruction uses as argument the result from the previous one), and in a sequence of more than 3 instructions (since there's 3 pipelines).
If you hit a branch, or a multiple store, multiplication or any instruction that needs more cycles you'll end up taking 3 to 5 cycles per instruction.
Find all posts by this user
Quote this message in a reply
11-29-2017, 05:54 AM
Post: #3
RE: For those interested in a Prime "Benchmark"...
Hello,

Althrough mips usually refers to millions instructions per seconds, a lot of the early mips mesurements programs that I have seen in the past did assume that some operations, like multiplications were not CPU instructions and took multiple instructions cycles to execute.

Similary, all these programs (even the more modern onces) assume that a statement like this one:
a= b+c<<d;
is 3 instructions.

However the Prime CPU does have a 32*32=64 bit multiplication instruction (which will take between 1 and 3 cycles to execute) and can execute an add + shift with result stored somewhere else in one single instructions (an intel chip would need 3 of them).

All this together might explain why the 400Mhz chip might clock 440 mips on some benchmarks.

Cyrille

Although I work for the HP calculator group, the views and opinions I post here are my own. I do not speak for HP.
Find all posts by this user
Quote this message in a reply
11-29-2017, 01:38 PM
Post: #4
RE: For those interested in a Prime "Benchmark"...
(11-29-2017 05:04 AM)Claudio L. Wrote:  
(11-28-2017 11:30 PM)zeno333 Wrote:  I found a web page that mentions the speed of the CPU in the HP prime in MIPS, and given that and using the clock of the Prime at 400 MHZ, the Prime is a 440 MIPS machine...pretty fast. Too bad there is not a program for the Prime equivalent to the LINPACK that tests FLOPS to compare it using Floating Point Operations.

Don't believe everything you read. The CPU in the Prime can at most and only under certain specific circumstances execute 1 instruction per cycle.

While that's the case for the Prime, it isn't the case for most desktop microprocessors these days. Most can execute more than one instruction per cycle. Try powering a handheld device with those, however. The battery life is terrible!

Tom L
Cui bono?
Find all posts by this user
Quote this message in a reply
11-29-2017, 09:01 PM
Post: #5
RE: For those interested in a Prime "Benchmark"...
(11-29-2017 01:38 PM)toml_12953 Wrote:  
(11-29-2017 05:04 AM)Claudio L. Wrote:  Don't believe everything you read. The CPU in the Prime can at most and only under certain specific circumstances execute 1 instruction per cycle.

While that's the case for the Prime, it isn't the case for most desktop microprocessors these days. Most can execute more than one instruction per cycle. Try powering a handheld device with those, however. The battery life is terrible!

Very true...I use a PC with a Ryzen 5-1400 CPU which runs at 3.2 GHZ or 3.4 GHZ in it's Turbo Mode and it benchmarks using the "IntelBurn" testing program at 33.6 GFLOPS.
Find all posts by this user
Quote this message in a reply
11-29-2017, 09:16 PM (This post was last modified: 11-29-2017 09:20 PM by StephenG1CMZ.)
Post: #6
RE: For those interested in a Prime "Benchmark"...
I have been working for some time on a Whetstone benchmark, but have been unable to locate any on-line specification of the correct outputs, without which it is difficult to have confidence in the accuracy of my implementation or to discover any difference in accuracy between the PPL and PC versions.
http://www.hpmuseum.org/forum/thread-817...=Whetstone
http://www.hpmuseum.org/forum/thread-959...l#pid84382
Can anyone suggest a suitable source?

Stephen Lewkowicz (G1CMZ)
https://my.numworks.com/python/steveg1cmz
Visit this user's website Find all posts by this user
Quote this message in a reply
12-04-2017, 07:57 AM (This post was last modified: 12-04-2017 08:08 AM by StephenG1CMZ.)
Post: #7
RE: For those interested in a Prime "Benchmark"...
I have now uploaded my implementation of a translation of the original Algol benchmark into PPL to the software library, although I still haven't found test results for that benchmark on-line.
http://www.hpmuseum.org/forum/thread-9627.html
This should be seen as a "Yes the Prime can run the benchmark", rather than a real measure of what speed the Prime is capable of achieving.
I am sure there are plenty of optimisations that could be done.

Stephen Lewkowicz (G1CMZ)
https://my.numworks.com/python/steveg1cmz
Visit this user's website Find all posts by this user
Quote this message in a reply
Post Reply 




User(s) browsing this thread: 1 Guest(s)