Lies, Damned Lies and Benchmarks.

Using, learning, programming and modding the Gigatron and anything related.
Forum rules
Be nice. No drama.
veekoo
Posts: 70
Joined: 07 Jun 2021, 07:07

Re: Lies, Damned Lies and Benchmarks.

Post by veekoo »

Just run all the different sieve versions.
Sieve0: 41.1 seconds normal screen mode
Sieve0: 28.67 seconds fast screen mode <- Ratio 2.05

Sieve1: 39.22 seconds normal screen mode
Sieve1: 27.57 seconds fast screen mode <- Ratio 1.97

Sieve2: 29.72 seconds normal screen mode
Sieve2: 20.87 seconds fast screen mode <- Ratio 1.49

To me it seems very fast for whats inside the machine. Good job engineering this device! ;)
lb3361
Posts: 188
Joined: 17 Feb 2021, 23:07

Re: Lies, Damned Lies and Benchmarks.

Post by lb3361 »

veekoo wrote: 01 May 2022, 12:09 Sieve0: 41.1 seconds normal screen mode
Sieve0: 28.67 seconds fast screen mode <- Ratio 2.05

Sieve1: 39.22 seconds normal screen mode
Sieve1: 27.57 seconds fast screen mode <- Ratio 1.97

Sieve2: 29.72 seconds normal screen mode
Sieve2: 20.87 seconds fast screen mode <- Ratio 1.49
I am really surprised by these changing ratios. First, I would have expected ratios closer to 3. Second, I am surprised to see the Sieve2 ratio (using memset) so different. At67?

Also, did you use the same compilation option -map=64k,hionly as https://github.com/lb3361/gigatron-lcc/ ... e/Makefile ? That shouldn't impact the ratios but make the code a bit faster by avoiding long jumps between screen holes..
veekoo
Posts: 70
Joined: 07 Jun 2021, 07:07

Re: Lies, Damned Lies and Benchmarks.

Post by veekoo »

lb3361 wrote: 01 May 2022, 14:37 Also, did you use the same compilation option -map=64k,hionly as https://github.com/lb3361/gigatron-lcc/ ... e/Makefile ? That shouldn't impact the ratios but make the code a bit faster by avoiding long jumps between screen holes..
I only changed -rom=dev, otherwise it was compiled same way

Btw. I am fiddling something with Linux-X86-PC and using the same sieve programs. I have trouble making good load to the machine. The parameters are mystery to me.

If I use these:
#define size 8190
#define sizepl 8191
iterations 10

I get:
10 iterations

1899 primes

Took 0.000584 seconds to execute

Your machine is 23972.602740 times faster than Z80(PL/I)-Sieve-benchmark
Your machine is 3253.424658 times faster than Intel 8086(Asm)(8 MHz)-Sieve-benchmark
Your machine is 1917.808219 times faster than Motorola 68000(Asm)(8 MHz)-Sieve-benchmark
Your machine is 188.356164 times faster than Cray1 supercomputer(Fortran)-Sieve-benchmark
Your machine is 35.102740 times faster than RPi modelB(C)-Sieve-benchmark
Your machine is 1.529110 times faster than Linux-X86-PC/Ryzen7-1700X(C)-Sieve-benchmark
Your machine is 1.215753 times faster than Linux-X86-PC/i7-4770K(C)-Sieve-benchmark
Your machine is 0.996575 times faster than Linux-X86-PC/i7-8850H(C)-Sieve-benchmark
Last edited by veekoo on 03 May 2022, 06:36, edited 5 times in total.
lb3361
Posts: 188
Joined: 17 Feb 2021, 23:07

Re: Lies, Damned Lies and Benchmarks.

Post by lb3361 »

That's probably correct. Modern computers are really fast, especially when they can work entirely from the memory cache...
veekoo
Posts: 70
Joined: 07 Jun 2021, 07:07

Re: Lies, Damned Lies and Benchmarks.

Post by veekoo »

This is little bit OT, but did everybody knew that Cray 1 supercomputer 1970's used only ICs and had 200 000 gates. 1990's they started to use microprocessors. Thanks to sieve now we can compare all the machines. Cray 1 supercomputer(Fortran)-Sieve-benchmark 0.110 seconds. Thats 190 times faster than GigatronTTL.
walter
Site Admin
Posts: 137
Joined: 13 May 2018, 08:00

Re: Lies, Damned Lies and Benchmarks.

Post by walter »

Nice. 190 times faster and 200.000 times more expensive :)

Still, the Cray-1 is better looking than the Gigatron. There's in in the Heinz Nixdorf museum in Paderborn Germany.
lb3361
Posts: 188
Joined: 17 Feb 2021, 23:07

Re: Lies, Damned Lies and Benchmarks.

Post by lb3361 »

What would have been the price of Gigatron in the 70s? ICs were not as cheap, especially SRAM. Still a good deal compared to a Cray1...
veekoo
Posts: 70
Joined: 07 Jun 2021, 07:07

Re: Lies, Damned Lies and Benchmarks.

Post by veekoo »

To get real world load to todays computers I modified sieve to "toughsieve"

#define size 80000000
#define sizepl 80000001

I get:
10 iterations

8974457 primes

Took 11.419000 seconds to execute

Your machine is 2.395744 times faster than Win10-X86-PC/i3-4000M(C)-ToughSieve-benchmark
Your machine is 2.082351 times faster than Linux-X86-PC/Ryzen7-1700X(C)-ToughSieve-benchmark
Your machine is 1.981259 times faster than Win10-X86-PC/Ryzen7-1700X(C)-ToughSieve-benchmark
Your machine is 1.114204 times faster than Linux-X86-PC/i7-8850H(C)-ToughSieve-benchmark
Your machine is 1.000350 times faster than Win10-X86-PC/i7-8850H(C)-ToughSieve-benchmark
Last edited by veekoo on 03 May 2022, 06:35, edited 1 time in total.
monsonite
Posts: 96
Joined: 17 May 2018, 07:17

Re: Lies, Damned Lies and Benchmarks.

Post by monsonite »

lb3361 wrote: 02 May 2022, 11:14 What would have been the price of Gigatron in the 70s? ICs were not as cheap, especially SRAM. Still a good deal compared to a Cray1...
I have found various magazine adverts in online archives for TTL ICs from the 1970s.

Almost all of the TTL ICs used in the Gigatron were available by 1975. Typically they cost between $1 and $2 back then, but $1 in 1975 is the equivalent of $5.34 in 2022 prices.

EPROM and RAM were still very expensive, and had slow access times. EPROM was about 450nS and RAM about 250nS. That would slow the Gigatron clock down to about 1MHz.

You would probably only afford about 2K words of ROM and 2K of RAM. A video screen would have been out of the question, unless you used a character ROM or a serial terminal interface.

Custom made PCBs were unheard of back then - so you would probably have used wirewrap sockets on a big prototyping board.

Probably the best comparison is the Apple I from 1976 which was $666.66 in kit form, about $3000 equivalent today.

We are very lucky that we can buy a Gigatron kit today for about $100!
at67
Posts: 486
Joined: 14 May 2018, 08:29

Re: Lies, Damned Lies and Benchmarks.

Post by at67 »

Messing around with Mode 1975 and new instructions recently added to ROMvX0, we have a new leader board champ!
Post Reply