y-cruncher

July 21, 2024)
Windows: Version 0.8.5 Build 9543 (Released: July 21, 2024)
Linux      : Version 0.8.5 Build 9543 (Released: July 21, 2024)
Latest Releases: (July 21, 2024)
y-cruncher v0.8.5.9543.zip
y-cruncher v0.8.5.9543-static.tar.xz
33.7 MB
y-cruncher v0.8.5.9543-dynamic.tar.xz
Core i3 8121U
Intel Cannon Lake
2/4
~2.5 - 3.2 GHz (stock)
8 GB @ 2400 MT/s
v0.8.5 (18-CNL)
x64 AVX512-VBMI
1.951
0.490
4.279
1.083
(省略されました)

July 20, 2024)
Windows: Version 0.8.5 Build 9542 (Released: July 20, 2024)
Linux      : Version 0.8.5 Build 9542 (Released: July 20, 2024)
Zen5 Optimizations: (July 20, 2024) - permalink
Optimizations for AMD's upcoming Zen5 processors has been released for y-cruncher v0.8.5. The latest build (9542) will a new binary "24-ZN5 ~ Komari".
At this time, I cannot say much other than the following:
The speedup of "24-ZN5" over "22-ZN4" is greater for single-threaded computations than multi-threaded computations.
The BBP program has not changed. Both "22-ZN4" and "24-ZN5" have the same BBP code.
None of the existing binaries have changed other than a minor bug fix. So existing benchmarks done with them will not be invalided by this update.
Version v0.8.5 Released: (July 1, 2024) - permalink
y-cruncher v0.8.5 has been released without Zen5 optimizations for now. Once I get final Zen5 hardware, I intend to follow up with an update to add it.
Originally, I intended to wait for the Zen5 binary to be ready before releasing v0.8.5. But it is now July and multiple media outlets are asking me about it since all signs are pointing at an end-of-July launch for Zen5.
y-cruncher v0.8.5 is a rather big update from v0.8.4 with new optimizations and a new benchmark category (the BBP). Thus hardware reviewers will need to rerun all benchmarks on older hardware. And this takes time.
Related: Version 0.8.5 Preview
Latest Releases: (July 20, 2024)
y-cruncher v0.8.5.9542.zip
42.9 MB
y-cruncher v0.8.5.9542-static.tar.xz
34.1 MB
(省略されました)

July 9, 2024)
Windows: Version 0.8.5 Build 9541 (Released: July 9, 2024)
Linux      : Version 0.8.5 Build 9541 (Released: July 9, 2024)
Latest Releases: (July 9, 2024)
y-cruncher v0.8.5.9541.zip
y-cruncher v0.8.5.9541-static.tar.xz
y-cruncher v0.8.5.9541-dynamic.tar.xz
The dynamic version has been tested on Ubuntu 24.04.
Comparison Chart: (Last updated: July 8, 2024)
Ryzen 9 7950X
AMD Zen 4
16/32
stock
28 GB - 5200 MT/s
v0.8.5 (22-ZN4)
x64 AVX512-GFNI
0.287
0.615
1.365
(省略されました)

July 1, 2024)
202 trillion digits - May 2024 (Jordan Ranous, Kevin O’Brien, and Brian Beeler - StorageReview)
105 trillion digits - February 2024 (Jordan Ranous, Kevin O’Brien, and Brian Beeler - StorageReview)
Windows: Version 0.8.5 Build 9539 (Released: July 1, 2024)
Linux      : Version 0.8.5 Build 9539 (Released: July 1, 2024)
Version v0.8.5 Released: (July 1, 2024) - permalink
y-cruncher v0.8.5 has been released without Zen5 optimizations for now. Once I get final Zen5 hardware, I intend to follow up with an update to add it.
Originally, I intended to wait for the Zen5 binary to be ready before releasing v0.8.5. But it is now July and multiple media outlets are asking me about it since all signs are pointing at an end-of-July launch for Zen5.
y-cruncher v0.8.5 is a rather big update from v0.8.4 with new optimizations and a new benchmark category (the BBP). Thus hardware reviewers will need to rerun all benchmarks on older hardware. And this takes time.
Related: Version 0.8.5 Preview
 
 
 
Note that the CPU Utilization of 3.33% is wrong. There is a yet-to-be-resolved issue that is causing long runs on many cores to bug out this measurement.
Pi Record Smashed at 202 Trillion Digits: (June 28, 2024) - permalink
StorageReview wasn't kidding when they said they were running a very large computation at the end of this video.
The previous record of 105 trillion digits set earlier this year was frought with problems. And with those lessions learned, this time it was done right.
Compared to last time, these changes were made:
Better storage configuration with most issues from last time resolved. So no more silly bottlenecks holding everything back.
(省略されました)

March 5, 2024)
Ramanujan's Constant.cfg
y-cruncher v0.8.4.9538a.zip

February 22, 2024)
Windows: Version 0.8.4 Build 9538 (Released: February 22, 2024)
Linux      : Version 0.8.4 Build 9538 (Released: February 22, 2024)
Latest Releases: (February 22, 2024)
y-cruncher v0.8.4.9538.zip
y-cruncher v0.8.4.9538-static.tar.xz
y-cruncher v0.8.4.9538-dynamic.tar.xz

February 21, 2024)
Windows: Version 0.8.4 Build 9537 (Released: February 21, 2024)
Linux      : Version 0.8.4 Build 9537 (Released: February 21, 2024)
Version 0.8.4 Released: (February 21, 2024) - permalink
y-cruncher v0.8.4 has been released with most of the improvements in non-Pi related math as well as continued improvements for large computations in swap mode.
 
 
New Math Improvements:
 
Jorge Zuniga has done it again! This time with new fastest formulas for Log(2), Log(3), and Log(5). You can grab the formula files here:
Log(2) - Zuniga (2024-i).cfg
Log(2) - Zuniga (2024-ii).cfg
Log(3) - Zuniga (2024-i).cfg
Log(5) - Zuniga (2024-i).cfg
More information on MathOverflow.
y-cruncher v0.8.4 will special-case for these values and will use these formulas instead of the auto-generated ArcCoth() Machin-like formulas.
And on a somewhat related note, the custom formula feature has gone through some rework and now has some new functionality:
Square(x)
Reciprocal(x)
(省略されました)

February 13, 2024)
Windows: Version 0.8.3 Build 9533 (Released: February 13, 2024)
Linux      : Version 0.8.3 Build 9533 (Released: February 13, 2024)
Version 0.8.3 Patched: (February 13, 2023) - permalink
A new patch for v0.8.3 has been released that fixes a serious bug in the N63 large multiply algorithm. The bug only affects specific versions and binaries.
The affected binaries are:
v0.8.2: 18-CNL ~ Shinoa
v0.8.2: 22-ZN4 ~ Kizuna
v0.8.3: 18-CNL ~ Shinoa
v0.8.3: 22-ZN4 ~ Kizuna
Both Windows and Linux are affected. And can only happen on computations above 29 trillion digits with the likelihood increasing for larger sizes. While this bug affects very few people, it is severe for those who are as it can block record attempts.
 
 
 
 
FLINT: The Rising Star of Bignum Libraries: (February 9, 2023) - permalink
 
So apparently there's a new crown for the fastest open-sourced bignum library. It is FLINT (Fast Library for Number Theory).
 
(省略されました)

Windows: Version 0.8.3 Build 9532 (Released: December 7, 2023)
Linux      : Version 0.8.3 Build 9532 (Released: December 7, 2023)
Latest Releases: (December 7, 2023)
y-cruncher v0.8.3.9532.zip
y-cruncher v0.8.3.9532-static.tar.xz
y-cruncher v0.8.3.9532-dynamic.tar.xz

December 6, 2023)
Windows: Version 0.8.3 Build 9531 (Released: December 5, 2023)
Linux      : Version 0.8.3 Build 9531 (Released: December 5, 2023)
Version 0.8.3 Released: (December 2, 2023) - permalink
Latest Releases: (December 5, 2023)
y-cruncher v0.8.3.9531.zip
33.0 MB
y-cruncher v0.8.3.9531-static.tar.xz
y-cruncher v0.8.3.9531-dynamic.tar.xz
16.6 MB

December 3, 2023)
Windows: Version 0.8.3 Build 9530 (Released: December 2, 2023)
Linux      : Version 0.8.3 Build 9530 (Released: December 2, 2023)
Version 0.8.3 Released: (December 12, 2023) - permalink
And it's out now with the big improvements to swap mode SSDs and the new formulas for several constants! Thus this release wraps up the rewrite of v0.8.x as well as the year itself.
Looking forward to next year, I have a ton of things that I want to do. Whether I'll get to them is another story of course...
Optimizations for Zen5. If the leaks about the 512-bit execution width are accurate, this could be a very juicy CPU.
Replace the "11-BD1" binary for AMD Bulldozer with "12-BD2" for AMD Piledriver. This replaces FMA4 with FMA3 which makes it much easier to test and maintain.
Fix some unparallelized code causing Amdahl's Law on modern large servers (specifically the 512-vcore Bergamos). More on this in a future post...
Add the ability to checkpoint inside the radix conversion. This is very much needed as it has been a common failure point in the past.
Increase y-cruncher's limit beyond 1015 digits. The current Pi record is 1014 digits, so it's getting close. This will require eliminating the "float-indexing limit".
Modern SSDs have fast SLC caches. Can this be exploited? y-cruncher has the ability to exploit locality in hierarchical storage, but it has never been enabled due to difficulty in testing. Is it time to take another look?
For now, I don't expect to do any further Intel optimizations until one of the following happens:
AVX512 returns to desktop. (unlikely)
APX and AVX10 become available to desktop. (ETA 2025?)
I manage to get a suitable Sapphire Rapids+ system for cheap.
The Windows on ARM devkits look promising (if they're ever in stock). Though I'm hesitant to pull the trigger on something I have no other use for.
December 2, 2023
November 27, 2023
(省略されました)