Improving your 5900X's speed and efficiency
Improving your 5900X's speed and efficiency
UPDATE 07/18/21 Began experiencing random instability with my old PSU. Switched it out for a Corsair RM850x (for other reasons, not solely for this), and those issues disappeared most of the time. I realized I still haven’t fully determined the correct CPU VDD settings. The assumption that full-scale current matches EDC was incorrect, as crashes kept occurring at 170A (EDC level) even after the new PSU. 180A proved stable, though higher currents caused crashes, and adjusting full-scale current also affected telemetry offsets. I reduced the current from 5mA to 20mA to restore power reporting accuracy after increasing full-scale current by 10A. Since then, I’ve overclocked my RAM. Currently running at 3800MHz CL16 versus 3600MHz CL16 with a 1900MHz FCLK—only a modest improvement, as Ryzen’s calculator suggests pushing beyond 3800MHz isn’t practical. I’m not inclined to tweak every timing parameter manually. What I can say is I’m a lazy overclocker: I want the best performance with minimal effort. All stress tests passed without issues. Hopefully, this will continue to improve significantly. ORIGINAL POST I had already undervolted my 5900X using the curve optimizer (30 steps, all cores), but after adding an Artic Liquid Freezer II 280, I decided to explore further. Here are some observations and results. They might help others or spark ideas for improvement. My specs are listed below if you’re interested:
- Top line: 4.6GHz all-core, 5.1GHz single core @38c idle (1,230 RPM), 84c under full load (1,481 RPM)
- Cinebench R20: 636 649 single core, 8770 9002 multicore
- Cinebench R23: 1635 1659 single core, 22562 23198 multicore
- CPU-Z: 683 (unchanged), single core, 10025 10116 multicore
BIOS settings: -30 steps all cores in curve optimizer, +150Mhz Offset PPT: 220W, TDC: 200A, EDC: 170A
CPU VDD Full Scale Current: 170A / 180A
Telemetry offset: 25mA at 20mA, 30mA set to reduce deviation
I found that PPT and TDC have little impact if they stay high enough. Performance drops sharply below certain thresholds. Running at 110A pushed max temp to 57°C—27°C lower than before. The performance loss wasn’t extreme, but it’s noticeable. Losing 200Mhz across all cores and around 300 points on R20 was significant.
Interestingly, 110A caused a 27°C temperature spike at full load. That’s a big improvement from my previous max temp. While the gain wasn’t massive, it showed that pushing lower voltages helped. I’m focusing on max performance with minimal tweaking.
I’ve also noticed that EDC plays a huge role. Testing values from 110A to 200A gave inconsistent results. The sweet spot seemed between 170–180A, where stability and speed improved noticeably. If cooling or case quality is an issue, adjusting VDD settings could pay off.
Some online users reported varied outcomes depending on EDC values. I suspect each rig needs its own fine-tuning. BIOS adjustments matter a lot—values like PPT, TDC, and EDC should be set carefully. I looked into OC guides but found little concrete advice on VDD telemetry offset.
I originally saw these numbers in my BIOS while trying to fix HWINFO’s “Power Reporting Deviation” error. Under load, HWINFO falsely indicated the CPU was drawing more power than it did. Since I’m using PBO for performance tuning, this might be a sign of misconfiguration. I searched online and found some OC tips, but nothing concrete on VDD offset. I tried a 30mA setting, then adjusted down to 25mA, landing at 99–101% efficiency. The final tweak was 1mA, which gave me near-perfect results (deviation at 76%).
Overall, my single-core scores stayed close to stock, even though I’m pushing 5.1GHz on one core—far beyond typical stock performance. Multicore scores are now the best across the board, which is great for gaming or heavy workloads. If you have any other interesting settings to share, feel free to let me know.
Your motherboard and configurations vary across brands and manufacturers, especially within the MSI lineup.
It was about the Artic Freezer I received from Amazon. The first unit I got was an older version, clearly not fully closed since the AIO was partially hanging out. There was even a half peeled sticker still visible, indicating "RMA" was on it. I was reassured they would send another one, which turned out to be fine. I’m curious about what the warehouse staff thought with that particular one.
I use 200 ppt 130 130 -30 CO for +200MHz. I don't enable LLC, or extra current, no offsets other than the one. I still have a little room to fine tune, I'm saving it as a surprise for later. It runs everything just fine. Single thread stuff its good for 5150, heavy duty like Linpack and it runs at 4500, WCG/f@h runs at 4600-4650, OCCT runs at 4750, memory testing like TM5, memtest, OCCT run at 4850, and 3d is usually 4950-5150. Manual tops out at about 4750, I can push x48 for certain things only.. the socket wont give me the power I need.. Max CPU PPT I have seen is roughly 215-220w I do let my fans ramp up, so its not quiet at the top end.. but for the most part she is pretty well behaved. She can get pretty rowdy though..
During testing with my 5900X, a specific range of EDC values showed optimal performance for the benchmark (compiling bzImage in the Linux kernel). Raising EDC further initially boosted scores but eventually caused declines. To isolate EDC effects, I set PPT and TDC to high values. The ideal setting turned out to be PPT=500 and TDC=245. I adjusted EDC between 140 and 220 in increments of 10 to map average compilation time versus EDC. A graph will be created for sharing later. Results may vary.