CoreMark and LINPACK benchmarks for Atari

All about the serious stuff.

Moderators: Mug UK, Zorro 2, Moderator Team

czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

CoreMark and LINPACK benchmarks for Atari

Post by czietz »

Lately, I have ported the well-known EEMBC’s CoreMark® and LINPACK benchmarks to the Atari. See below for download links and results. I consider the latter benchmark to be more of a gimmick; you can compare your FPU to a Cray supercomputer.

However, CoreMark is immensely valuable to gauge the speed (or speed-up in case of an accelerator) of a system. Some old benchmarks measure speed in terms of repeated execution of a single instruction. In contrast, CoreMark uses algorithmic building blocks and data structures that you would find in real applications. It's for this reason that CoreMark has become the de-facto benchmark for modern embedded processors. Since embedded systems are comparable to Atari machines in terms of computing power and memory size, CoreMark suits them well, too, and gives much more meaningful results for the overall speed.

CoreMark - runs on every 68k Atari:
Download: https://github.com/czietz/coremark/rele ... atari_port
Results for comparison: https://github.com/czietz/coremark/wiki/Results
Source code: https://github.com/czietz/coremark
Whitepaper by EEMBC about CoreMark design: https://www.eembc.org/techlit/articles/ ... epaper.pdf

LINPACK - needs at least 68020 + FPU:
Download, results, source code: https://github.com/czietz/linpack-atari/
Wikipedia article: https://en.wikipedia.org/wiki/LINPACK_benchmarks
User avatar
Cyprian
10 GOTO 10
10 GOTO 10
Posts: 2230
Joined: Fri Oct 04, 2002 11:23 am
Location: Warsaw, Poland

Re: CoreMark and LINPACK benchmarks for Atari

Post by Cyprian »

nice idea
especially the results are gathered together in one places
We now compare different Atari machines.

Below interesting figures for Firebee - Fast RAM vs ST RAM:

Code: Select all

Firebee FAST- (also ColdFire-) RAM:
Iterations/Sec   : 466.744457

Firebee FPGA-RAM (praktisch = ST RAM):
Iterations/Sec   : 151.630023
Mega ST 1 / 7800 / Portfolio / Lynx II / Jaguar / TT030 / Mega STe / 800 XL / 1040 STe / Falcon030 / 65 XE / 520 STm / SM124 / SC1435
DDD HDD / AT Speed C16 / TF536 / SDrive / PAK68/3 / Lynx Multi Card / LDW Super 2000 / XCA12 / SkunkBoard / CosmosEx / SatanDisk / UltraSatan / USB Floppy Drive Emulator / Eiffel / SIO2PC / Crazy Dots / PAM Net
Hatari / Steem SSE / Aranym / Saint
http://260ste.atari.org
Moulinaie
Captain Atari
Captain Atari
Posts: 391
Joined: Wed Feb 01, 2012 9:34 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by Moulinaie »

Thanks !

Works well on the TT.

Guillaume.
User avatar
mfro
Atari Super Hero
Atari Super Hero
Posts: 968
Joined: Thu Aug 02, 2012 10:33 am
Location: SW Germany

Re: CoreMark and LINPACK benchmarks for Atari

Post by mfro »

Cyprian wrote: Thu Feb 11, 2021 4:39 pm Below interesting figures for Firebee - Fast RAM vs ST RAM:
Note that this isn't the default setup, but my own config (all FPGA-RAM = ST RAM).

In the original setup, only the screen memory is (slow) FPGA RAM (not really slow but only accessible for the CPU through the FlexBus, mapped in using the ColdFire MMU), the rest of the first 14 M is ColdFire (= Fast/Alternate) RAM. Although slower, I rather like the former as it allows animation tricks flipping the screen base adress just as we all know it from the original. More like a TT (if you want to go fast, use Alternate RAM).
OL
Atari Super Hero
Atari Super Hero
Posts: 598
Joined: Fri Apr 01, 2005 6:59 am
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by OL »

czietz wrote: Thu Feb 11, 2021 4:23 pm Lately, I have ported the well-known EEMBC’s CoreMark® and LINPACK benchmarks to the Atari. See below for download links and results. I consider the latter benchmark to be more of a gimmick; you can compare your FPU to a Cray supercomputer.

However, CoreMark is immensely valuable to gauge the speed (or speed-up in case of an accelerator) of a system. Some old benchmarks measure speed in terms of repeated execution of a single instruction. In contrast, CoreMark uses algorithmic building blocks and data structures that you would find in real applications. It's for this reason that CoreMark has become the de-facto benchmark for modern embedded processors. Since embedded systems are comparable to Atari machines in terms of computing power and memory size, CoreMark suits them well, too, and gives much more meaningful results for the overall speed.

CoreMark - runs on every 68k Atari:
Download: https://github.com/czietz/coremark/rele ... atari_port
Results for comparison: https://github.com/czietz/coremark/wiki/Results
Source code: https://github.com/czietz/coremark
Whitepaper by EEMBC about CoreMark design: https://www.eembc.org/techlit/articles/ ... epaper.pdf

LINPACK - needs at least 68020 + FPU:
Download, results, source code: https://github.com/czietz/linpack-atari/
Wikipedia article: https://en.wikipedia.org/wiki/LINPACK_benchmarks
Linpack value on V4SA 85Mhz core january 2021: 8663

Olivier
OL
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

OL wrote: Thu Feb 11, 2021 7:27 pm Linpack value on V4SA 85Mhz core january 2021: 8663
Thank you, added to my list on GitHub.

PS: You can also run CoreMark on the V4SA, if you skip the launcher !RUNME.TOS (which is incompatible) but start COREMARK.TOS directly, e.g. from EmuCON.
OL
Atari Super Hero
Atari Super Hero
Posts: 598
Joined: Fri Apr 01, 2005 6:59 am
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by OL »

czietz wrote: Thu Feb 11, 2021 7:32 pm
OL wrote: Thu Feb 11, 2021 7:27 pm Linpack value on V4SA 85Mhz core january 2021: 8663
Thank you, added to my list on GitHub.

PS: You can also run CoreMark on the V4SA, if you skip the launcher !RUNME.TOS (which is incompatible) but start COREMARK.TOS directly, e.g. from EmuCON.
Runme run fine on V4 no problem.

Coremark result for V4SA 85Mhz core january 2021: 213.75


Note GCC 4-10 not give good code on 68K unfortunately compare to other processor and there is no 68020 version, I think it should

Olivier
OL
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

OL wrote: Thu Feb 11, 2021 10:22 pm and there is no 68020 version, I think it should
Building and releasing a new/different version would invalidate all previous results. Therefore, for results to be added to my table, one has to use the one original binary released by me.

However, the source code is available if you want to experiment with different compiler settings or compilers. Just please make absolutely sure to mention that any results gained in this way are not "official" scores. Also note that -- as per EEMBC rules -- you may only change the files core_portme.*. Of course, you're not allowed to change the actual benchmark code (e.g. to make it faster).
User avatar
Arne
Atari Super Hero
Atari Super Hero
Posts: 746
Joined: Thu Nov 01, 2007 10:01 am

Re: CoreMark and LINPACK benchmarks for Atari

Post by Arne »

HBS 240:
Image

HBS640T28:
Image
Image
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

Arne wrote: Fri May 14, 2021 1:55 pm HBS 240:
HBS640T28:
Thank you! Added to the results table.
User avatar
Arne
Atari Super Hero
Atari Super Hero
Posts: 746
Joined: Thu Nov 01, 2007 10:01 am

Re: CoreMark and LINPACK benchmarks for Atari

Post by Arne »

Two more results (actually it's more like 1.5):

Hypercache 030 (25MHz L1 I/D On and L2 I/D On):
Image

Hypercache 030 (32MHz L1 I/D On and L2 I/D On):
Image

I don't know if the 32MHz version was ever sold but swapping CPU and oscillator does the trick.
But 36MHz was too much for the HC030. But I don't complain. 8)
Image
User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4723
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by DarkLord »

Is Mint required to run Linpack?

On my 68030/68882 STacy, it 2 bombs with the setting at 200.

Thanks.
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520
User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4723
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by DarkLord »

Here are the coremark results for my Atari STacy, equipped with a Pak 68/3 board.

That's a 68030 running at 40mzh, a 68882 FPU running at 40mhz, 4 megs of ST RAM and TOS v3.06.

Looking at the other scores on the table, this feels off. It was beaten by slower running '030's and
'020 boards, if I'm reading the table right... Thanks.

P1010007.JPG
You do not have the required permissions to view the files attached to this post.
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

DarkLord wrote: Mon May 17, 2021 2:37 am Is Mint required to run Linpack?
MiNT is not required.
DarkLord wrote: Mon May 17, 2021 2:49 am Looking at the other scores on the table, this feels off. It was beaten by slower running '030's and
'020 boards, if I'm reading the table right... Thanks.
I assume this is the effect of not having fast RAM. If you look up the result for the TT (68030 @ 32 MHz) with only ST-RAM you find 8.7 it/s. Scaled by 40/32 MHz, this matches your result well.
User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4723
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by DarkLord »

Okay, thanks. Not sure about Linpack then. I can set it to something like "10" and it runs,
but seems to take forever. Anything near the 200 setting, 2 bomb crash (not fatal, back
to desktop).

As far as coremark goes, I guess you're right. I was looking at Arne's 25mhz and 32mhz '030
results being higher than my '030 at 40mhz. Perhaps he's using fast RAM and that makes
the difference.

Thanks again. :)
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520
User avatar
Arne
Atari Super Hero
Atari Super Hero
Posts: 746
Joined: Thu Nov 01, 2007 10:01 am

Re: CoreMark and LINPACK benchmarks for Atari

Post by Arne »

DarkLord wrote: Mon May 17, 2021 3:22 pm As far as coremark goes, I guess you're right. I was looking at Arne's 25mhz and 32mhz '030
results being higher than my '030 at 40mhz.
8O
Sure you switched on L1/L2?
HC030 does have 16KB L2 which is (according to the config ACC) split into I-, and D-cache.
DarkLord wrote: Mon May 17, 2021 3:22 pm Perhaps he's using fast RAM and that makes the difference.
HC030 does not have FastRAM! The SST Board can be populated with FastRAM but not the Hypercache 030.
It's a standard 8MHz board clock MegaST4 board w/o Blitter + HBS or HC030.
Image
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

Arne wrote: Mon May 17, 2021 4:06 pm
DarkLord wrote: Mon May 17, 2021 3:22 pm As far as coremark goes, I guess you're right. I was looking at Arne's 25mhz and 32mhz '030
results being higher than my '030 at 40mhz.
Sure you switched on L1/L2?
Iirc, the L2 cache is optional on the PAK68/3.
Arne wrote: Mon May 17, 2021 4:06 pm HC030 does have 16KB L2 which is (according to the config ACC) split into I-, and D-cache.
HC030 does not have FastRAM!
But it does have enough L2 cache to significantly accelerate CoreMark (or rather: the system in general).
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

DarkLord wrote: Mon May 17, 2021 3:22 pm As far as coremark goes, I guess you're right. I was looking at Arne's 25mhz and 32mhz '030
results being higher than my '030 at 40mhz.
The PAK68/3 page quotes Dhyrstone benchmark results: http://www.wrsonline.de/pak3.html. According to that table, a PAK68/3 running at 50 MHz is about 4.3 times as fast as a standard ST. Adding fast RAM more than doubles the speedup!
User avatar
Arne
Atari Super Hero
Atari Super Hero
Posts: 746
Joined: Thu Nov 01, 2007 10:01 am

Re: CoreMark and LINPACK benchmarks for Atari

Post by Arne »

czietz wrote: Mon May 17, 2021 4:15 pm Iirc, the L2 cache is optional on the PAK68/3.
Sure. But without L2 it's like driving with an attracted handbrake. A buddy once called the 256Bytes L1 I-Cache an extended prefetch queue.
I can put a 40MHz quartz on my PAK68/3 030 with L2 for comparison (but no FRAK).
Image
czietz
Hardware Guru
Hardware Guru
Posts: 1540
Joined: Tue May 24, 2016 6:47 pm

Re: CoreMark and LINPACK benchmarks for Atari

Post by czietz »

Arne wrote: Mon May 17, 2021 4:29 pm Sure. But without L2 it's like driving with an attracted handbrake.
I fully agree! A tool like MEMSPEED should be able to tell if there is L2 cache available (and activated):
http://atariftp.czietz.de/pub/atari/Uti ... mspeed.lzh
User avatar
Arne
Atari Super Hero
Atari Super Hero
Posts: 746
Joined: Thu Nov 01, 2007 10:01 am

Re: CoreMark and LINPACK benchmarks for Atari

Post by Arne »

czietz wrote: Mon May 17, 2021 4:36 pmA tool like MEMSPEED should be able to tell if there is L2 cache available (and activated)
I'd say: the owner should know. :D
Image
User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4723
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by DarkLord »

Kinda hard to open up the STacy just to double check but I did take pictures
before I closed everything up.

According to my pictures and the Pak PDF, J2 and J4 must be open to be
enabled and they are. That's CPU cache and SLC (second level cache).

I will download and run MEMSPEED when I get the chance though, just
out of curiosity. :)

Thanks.
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520
User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4723
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by DarkLord »

Okay, ran memspeed just for the fun of it...and it looks like the caches
are enabled...

P1010004.JPG
You do not have the required permissions to view the files attached to this post.
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520
User avatar
Arne
Atari Super Hero
Atari Super Hero
Posts: 746
Joined: Thu Nov 01, 2007 10:01 am

Re: CoreMark and LINPACK benchmarks for Atari

Post by Arne »

Here are my results:
Image

Image
Image
User avatar
DarkLord
Fuji Shaped Bastard
Fuji Shaped Bastard
Posts: 4723
Joined: Mon Aug 16, 2004 12:06 pm
Location: Prestonsburg, KY - USA
Contact:

Re: CoreMark and LINPACK benchmarks for Atari

Post by DarkLord »

So it's actually pretty close.

And it appears that your 40mhz Pak 68/3 is also beaten by the 32mhz '030 Hypercache board.

Okay, thanks for the comparisons, appreciated.
Welcome To DarkForce! http://www.darkforce.org "The Fuji Lives.!"
Atari SW/HW based BBS - Telnet:darkforce-bbs.dyndns.org 520
Post Reply

Return to “Applications”