1usmus
patreon
HYDRA 2.0E PRO (build 2040) - RELEASED!
🕑 Added 2025-10-31 16:47:44 +0000 UTCHello super community!
This update has been in development for several months and introduces the long-awaited AMD Curve Optimizer auto-tuning powered by the SQ AGENT algorithm. A few important notes:
Expect different results. AMD CO outcomes will look radically different from what you’ve seen before. The optimizer targets overall performance and energy efficiency rather than “the most negative per-core CO.”
No time wasted on non-critical cores. SQ AGENT skips cores that don’t move the needle. If a core cannot improve performance or efficiency, the algorithm doesn’t search AMD CO for it.
Diagnostics may take longer on some CPUs. Despite 24/7 bench operation, we haven’t covered every processor. On certain models (e.g., 9950X3D and 9900X3D), diagnostics may take longer than ideal.
Build 2040 is not final. Several innovations aren’t included yet. Expect builds 2041 and 2042 over the next two weeks.
About “stability.” There will always be a stress test capable of declaring a system unstable—even at AMD CO +10. The current presets are tuned for real-world usage and should cover ~90% of typical workloads, which is a significant improvement over previous HYDRA releases.
Modernized search flows. The CORE HOC CO, CCD HOC CO, and PROFILE CREATION algorithms have been updated. Coverage is not exhaustive yet. For this build, please limit experiments to AMD CO search—some unblocked features haven’t been fully validated.
Logs welcome. Please share your logs; I’m actively reviewing them.
ZEN 5 EXAMPLES:
Here’s an example of what AMD CO can look like on a given processor. On CCD 0, the spread between some per-core CO values can approach ~20 steps. This is expected: the new search focuses on the aggregate performance of all cores within a CCD, balancing each core so no weak cores remain.
I call the resulting effect “intersection”—when per-core curves within a CCD converge as closely as possible. Why it matters:
Tighter per-core curves. In every CCD, each core’s curve is brought as close to its neighbors as possible—this convergence is the “intersection.”
Fewer boost overshoots, fewer crashes. Intersection reduces the risk of over-boosting caused by an overestimated CO on any single core.
Better thermal uniformity. With cores aligned, CCD thermals become more even.
Faster, smarter CO search. The raw AMD CO phase on a 9950X now takes ~60–90 minutes without sacrificing accuracy. It also avoids “silly” overestimates on cores that don’t improve real performance.
Maximum multi-threaded throughput. Most importantly, intersection prioritizes overall CCD health, translating into higher multi-thread performance.
It may sound ambitious, but HYDRA is moving fast—and intersection is a big step forward.
ONLINE-GUIDE (1.8H) :
CHANGELOG:
New AMD CO diagnostic (auto-tuning) based on SQ AGENT. New presets. Better stability and accuracy. Data set selector.
SEARCH CORE HOC CO, SEARCH CCD HOC CO, and PROFILE CREATION search algorithms have also been modernized.
Phoenix/Autoloading has received 3 additional triggers. No gaps in the auto-start process.
Autoloading is programmed only for the user who launched HYDRA.
Minimization/normalization of the HYDRA window by clicking on the icon in the taskbar.
Fixed distortion of the HYDRA window shape if the user forcibly turned off the monitor.
Fixed a rare bug where a closed previous instance of HYDRA might not allow a new instance to be launched.
Users of NVIDIA PRO graphics cards will be able to use most features without restriction (except for GPU/RTX-TUNER auto-overclocking).
Correction of the eCLK description; it is now called BCLK and is supported by any motherboard. Safe range up to 103 MHz.
Reduced CPU usage of HYDRA for cases when anti-cheat blocked some functions.
Windows Defender loyalty has been slightly improved.
Fixed some bugs related to Ycruncher's unpredictable behavior during CPU diagnostics.
Improved mathematics for processor statistics. It has become more accurate.
Performance comparison and single-threaded clock speed measurement functions using Cinebench have received a number of improvements.
The sharpness of a number of UI elements has been improved.
If Vulkan Memtest does not work on your system for some reason, this will not invalidate the GPU auto-tuning results.
KNOWN ISSUES:
* False positive of antivirus (like always).
DOWNLOAD:
LINK (only an early access plan)
Since piracy in recent months has become widespread early access will be in Discord. If you do not have access to Discord, make sure your Patreon profile is complete. The process of granting access is automated. Also note that pirate resources may add malicious software to the archive. Use only a proven resource.
VIRUSTOTAL:
Comments
Carlo Gosamo
I can't access the link on reddit (red pop up error, impossible to load message). I have the 1 month subscription. EDIT: i've discovered why, with 1 month subscription you don't have access to the early acces, i've send you a pvt msg asking for a refund, thanx.
Mr. Nobody
Cheers! Love the progress you're making! Can i get added to your discord server? will send a chat about it. Thanks!
Maximiliano Hernandez Cervantes
It won't let me join the Discord server
angelzx10r
Hi, where can I download the program?
TomCatT
Looks like another big update. Can’t wait to take it for a spin.
