Beginner in Overclocking. Some Questions
Beginner in Overclocking. Some Questions
Hi all, my name is John. I just built my first Rig and I am totally into learning about overclocking. I'm 65 and almost retired and I can see this becoming a labor of love well into my retirement years (along with photography and gaming!!!). First, let me start with my build:
CPU - AMD Ryzen 9 5900x
Mobo - Asus Tuf Gaming x-570 Pro wifi
CPU Cooling - Corsair 115i -RGB platinum (280 mm) - mounted on front as a push intake
RAM - Corsair Vengeance RGB rt 32gb (2 16gb) ddr4 3600mhz C16
Memory - 1) Samsung 980 PRO PCIe 4.0 NVMe® SSD 1TB
2) Samsung 860 EVO 1TB SATA 3 SSD
3) Crutial MX300 2.5 in 500 GB SATA 3 SSD
4) Toshiba external SSD 2 TB USB 3.0
Case - Lian Li Lancool ii Mesh RGB
Graphics - EVGA GeForce GTX 1060 (awaiting prices to drop and availability of the Radeon 6800 xt)
PSU - Enermax Revolution D.F 850W
Monitor - BenQ ex3202r 144 Hz 2k
Case Cooling - 4 Corsair 120mm LL RGB PWM Dual Light Loop fans (2 intakes on bottom, 1 top back, 1 upper back of case) plus the two front 140 mm aio cooler Corsair ML RGB fans
Yes, I made it look pretty but that was important to me!!!
So, I have been reading through the CPU overclocking guide and tutorial for beginners and had some questions. First off I ran Memtest86 for 4 full passes with no errors (at default settings (3200 MHz after clearing CMOS on Asus Mobo)). Now I am ready to start changing clock frequency and voltage.
Question 1 - It says to start at base frequency and increase in small increments. Then it goes on and says to start at base frequency (which is 3.7 Ghz for the 5900x) and increase to the Max-Turbo frequency for a single core (4.8 Ghz for the 5900x) at stock voltage (I couldn't find what that was but I can follow the guide and determine what to set it at). Also, I could not find a setting in the BIOS for "per core" for the CPU Core Ratio. Anyhow, isn't a jump from 3.7 to 4.8 a big jump and not a small incremetal one as initially recommended???
Question 2 - First off, my temperatures jump all over the place which I understand is pretty normal for the AMD CPU's but what I am a little concerned about is my average CCD1 temp is almost 5.5 degrees hotter than the CCD2 after 1 hour of idle use. The cores 6-10 are between 31 and 32 avg (degees C) but Core 11 is up near 39. Cores 0-5 are at 37-40 avg. again at idle. I still see a difference when the system is being stressed although the difference tightens up to about 3 degrees between the CPU CCD1 vs 2 (Tdie) average with 1 again being higher. I also notice the power levels are about 1.5 W greater in Cores 0-5 (CDD1) than 6-11 (CDD2) suring the stability test. I used the stock paste which was preapplied on the Corsair AIO. Is this normal or do I have a cooling problem here??
Thanks for any help and being patient with my ignorance!!
I don't know about Memtest86, from what I hear it is not a very good stability tester for RAM. I think I've run that program once, 15 years ago. Here is more up to date things on RAM stability tests and overclocking it.
https://github.com/integralfx/MemTestHel...guide/DDR4 OC Guide.md
All silicon isn't made equal, there is always variation. I expect that is what you are seeing with regards to CCD1 vs CCD2. Ryzen also has 'Preferred core(s)'. Ryzen Master should list which cores those are. Preferred core(s) should be able to clock a little bit higher than the rest.
I don't really care about it. Why? Let's say you are in a race, Ferrari vs Porsche (your cores). Now, the Ferrari(s) might get to the finish line before the...
I don't know about Memtest86, from what I hear it is not a very good stability tester for RAM. I think I've run that program once, 15 years ago. Here is more up to date things on RAM stability tests and overclocking it.
https://github.com/integralfx/MemTestHel...guide/DDR4 OC Guide.md
All silicon isn't made equal, there is always variation. I expect that is what you are seeing with regards to CCD1 vs CCD2. Ryzen also has 'Preferred core(s)'. Ryzen Master should list which cores those are. Preferred core(s) should be able to clock a little bit higher than the rest.
I don't really care about it. Why? Let's say you are in a race, Ferrari vs Porsche (your cores). Now, the Ferrari(s) might get to the finish line before the Porsches BUT the Ferrari(s) would STILL have to wait on the Porsches to finish for the race to be over (your workload finishes). So what did I gain? 1 core boosting higher than the rest, in a multicore load. I gained nothing. Because that 1 core wont boost higher under a multicore load, that gets turned off. It only works in Singlecore loads, to my knowledge (never trust anything you read on the internet).
Back to OC. Most people tune their PBO2 and CO, Curve Optimizer I think. You have PPT, TDC and EDC. If you run a workload like Cinebench R23 and have Hwinfo64 open, you can see which of those is limiting your CPU from boosting. Usually it is EDC. The limiting factor will sit at 100% in Hwinfo64 out of those three, PPT, TDC, EDC.
Read the first posts Theory-part here to understand more.
https://www.overclock.net/threads/unders...c.1732088/
IIRC, PPT is measured in Watts, the rest in Amps. So a PPT with a value 100 will most likely throttle your CPU. 150-200 shouldn't. Look at Hwinfo64 again, what is CPU PPT sitting at under a heavy workload with PBO at stock?
With my 5600X and 240mm Corsair AIO, I can reach 140 watts before my cooling just gives up and CPU overheats. So I've settled for 120-130 Watts. Mind you, I don't use PBO, at all. I have a static overclock. But that amount of wattage needs dissipating/cooling, no matter what.
TDC I could leave at stock, didn't limit me. But EDC, that was my limiter. From what I've read countless times, EDC seems to work best if it reaches 90-98% or so. You get the best performance. Never 100% but just below it. If you set EDC too high, you might loose performance. Say it sits at 60-70% for example.
When it comes to Curve Optimizer, you can apply a negative offset for all-cores or per core. You will have cores that can do higher negative offsets than the rest. Some might do -5, others -25. For per-core CO, this is a tedious part. You would have to test each core individually, in something like OCCT. I haven't tried this. But the procedure is something like only allowing 1 core, in Windows, to run the OCCT test. You set the Affinity to 1 specific core, the one you are testing the negative offset on. Run the test for like an hour. But do also note, with negative core offset, the cores will usually crash when idle. Something to remember. In other words, you might be fine gaming, running benchmarks etc but as you are idling, system crashes.
With the all-core offset, you would have to settle with the value that the poorest core can run. Start with -5. Then go higher in steps of 3 or 5. Like -8 or -10 etc.
I haven't talked about CPU voltage at all. Well, for this, find out your FIT voltage. Deemed as the safe voltage. For this, reset BIOS to defaults, only enabling XMP and PBO. Run Prime95 smallFFTs and have Hwinfo64 open. Look at CPU core voltage (SVI2 TFN). The lowest value it reaches, that is your FIT voltage. 1-2 minutes of Prime95 should suffice for this.
Now, you have to go into BIOS, set a voltage around that FIT voltage for your CPU. It is not going to be exact. So you have to play with voltage and LLC for CPU.
Let's take an example: My FIT voltage is 1.275v. If I set CPU voltage to 1.275v in BIOS, it's not going to stay there in Prime95. It will go lower. It will 'droop' lower. In order to get it to not droop as much, I have to set LLC to 3. My Asus board has 5 levels, 1-5. 1 being the weakest and 5 being for LN2 ONLY. So I choose to use the middle-of-the-road LLC. Even with that, my voltage drops too low. I might sit at 1.26 volts in Prime95 smallFFT. With 1.275v and LLC 3 set in BIOS. So what I had to do, via trial and error was to increase the voltage, 1 step at a time, til it would droop to 1.275v. Turns out, for me, that was 1.3v and LLC 3, in BIOS. Note:These values, you can't copy them right off. As I said, every silicon is different, different characteristics. And higher core-count CPUs will have to run at lower voltages. You might find your FIT voltage is 1.20v or 1.225v. Perfectly normal.
In terms of vDroop (voltage droop), you want it. Voltage should drop under load. The heavier the load, the lower it will drop. Now, if your voltage goes the other way, increasing with load, that is bad. Avoid it at all costs. Your LLC is too aggressive and might lead to a dead CPU.
When it comes to your initial CPU OC question about frequency. 3700 Mhz vs 4800 Mhz. You know it can run the base clock, no problem. It can boost to 4.8 Ghz. You don't want to waste time and test 3.75 Ghz all-core, then 3.8, 3.85 Ghz and so on. You 'jump ahead'. To something like 4.4-4.5 Ghz. That is how I would do it. Mine only does 4.6 Ghz boost so I started at 4.3 Ghz. And ended up at 4.45 Ghz as final OC. Again, by running Prime95 smallFFT. If a worker-thread fails in Prime95, CPU is unstable. Prime95 wont give you an error box or anything. So keep an eye on those worker threads. Should be a quick test, if you survive for 2 mins, it should be good. Then, increase frequency by 50 Mhz, run Prime95 again. If Prime95 crashes, reduce frequency by 50 Mhz and that is your CPU overclocking done. If you set up your FIT voltage (CPU voltage and LLC) correctly.
One thing to bear in mind. If you are going to do RAM AND CPU overclocking, it is recommended to do the RAM first. Because if you do it the other way around, you might find your stable CPU OC isn't so stable anymore. OC'ed RAM puts an extra strain on the CPU.
You might see terms like vDIMM, vSOC etc. The 'v' stands for voltage. DIMM is your RAM. Some motherboards list DRAM voltage as vDIMM.
Another common thing is stuff like tCL, tRP etc on RAM. 't' stands for time, i think, measured in clock cycles when it comes to RAM. Samsung B-die runs tCL at around 14. But in terms of nanoseconds, that is around 7.7 ns.
Something else that confuses people is the 3200 Mhz, Megatransfers, Megabits. Some programs, I think CPU-Z is one, will list your actual RAM Mhz. And it will be half of that, in this case 1600 Mhz. This is because RAM is DDR, Double-Data Rate. Sends a signal on the High AND Low. So throughput is doubled. The most correct term would probably be Megabits. But no one uses it, especially not RAM manufacturers and their PR firms.
Thank you for your kind words about mamasan2000! I'll review it over the next few days and go through more material. If I miss anything, I'll replay it.
Also, concerning Cinebench R23, I'm unable to run it due to an application error each time. Instead, I'm using Cinebench R20.
R15, 20, 23 doesn't matter; I just chose one. Keep in mind, it's a benchmark, not a test for CPU stability as some mistakenly believe.
I have put about 4 hours of research in and have determined that there are as many ways to approach OCing as there are people doing it!!! I did find some great Youtube's by The Graying Tech on PPT, TDC, and EDC and have a much better understanding of what is going on with PBO and autoOC with my processor. I can also clearly see in Ryzen Master which of my cores are the preferred ones plus it is also evident that CCD0 is utilized far more than CCD1 so it makes sense that the temperatures are higher. I find it interesting though that CoreTemp App doesn't show the differences nearly as much as HwINFO64. Also, other AMD users see these differences in CCD0 vs CCD1 temps, and "knowledgeable people" say this is normal.
"what is CPU PPT sitting at under a heavy workload with PBO at stock?" It reads from 142 to 142.9 (100%) so I do understand that this needs to be increased so it is not throttling the CPU. Both TDC and EDC are also at 100% under load. Before I start playing with these variables, I have a few basic questions.
My PBO at stock was what I have after clearing my CMOS so I'm at default mobo settings correct? I do see that at the default settings, I am getting individual core boost's up close to 4900 MHz without load but then they increase to constant levels at 4200 running Cinebench R20 or 4400 with AIDA64. Isn't this PBO already?? Or is this what is meant when they say this processor is unlocked and PBO, autoOC, and Manual OC is going beyond the stock OC that I am getting right out of the box.
I want to work with PBO first before going to Manual OC. Getting ahead of myself a little, PPT, TDC, EDC, and curve optimizer are only adjusted with PBO and not manual OC correct? When and if I mess with a manual all-core OC, I would be changing the CPU Core ratio and voltages as I understand it.
I had to laugh when you stated to OC the memory first because that one is obviously controversial as I've read it both ways from "more knowledgable people". Although I am a newbie, I think I like the OC memory first camp. My memory did pass the OCCT memory and Large AVX2 CPU test. Unfortunately, I did get errors a few weeks ago on a CPU stability test (OCCT or AIDA64 extreme) by just changing my IA Overclock Tuner in BIOS to DOCP (using the full 3600 MHz of my memory instead of 3200 stock) and setting the FCLK to 1800. So these two changes in BIOS with everything else at stock did give me errors when running stability or stress tests. I won't swear to it but I seem to remember on one of the stress or stability tests, the problem was in Core 7. I am going to repeat this part (since I did it a while back) and see what is happening.
Update, at stock settings but changing to DOCP and FCLK to 1800, I did get 1 memory error when testing the memory on OCCT and the CPU crashed at the 9 min mark during the CPU OCCT test. Looks like I need to do some work with the memory. I will start reading about that. My memory is not on the tested Ryzen list for the 9 5900x so I guess there is no guarantee I will get 3600 from the memory.
CCD0 offers superior cores, while AMD's CPPC always prioritizes the best available cores initially.
Check if PBO is active in BIOS.
Yes, you can enable a full-core overclock manually without activating PBO or disabling Core Performance Boost, which prevents issues.
The CPU core ratio refers to frequency relative to base bus speed (old frontsidebus at 100 MHz). For a 4.4 GHz processor, the core ratio would be 44. Multiplying by 100 gives 4400 MHz. Alternatively, if you adjust the other settings, PBO can be enabled within its limits.
Do NOT interfere with CO settings. You can adjust them as needed.
Importantly, enable Core Performance Boost. Otherwise, the CPU may not exceed base frequencies and could stall at around 3.7 GHz.
I previously performed a CPU overclock out of habit—since I haven’t used RAM overclocking in years, I reduced it by 50 MHz after adjusting RAM speeds. This adjustment might be necessary.
My RAM isn’t supported by the Mobos QVL, but that doesn’t affect me.
My Asus X470 Prime Pro only supports up to 3433 MHz for RAM overclocking. I’ve reached a maximum of 3600 MHz and boot at 4400 MHz. However, this results in poor Aida64 performance due to the inability to reach 2200 Fclock. I run it at 1900 MHz with 3800 MHz on RAM using tighter timings than XMP. As a pro, I use 1.55 volts and a 120mm fan for cooling. The recommended daily voltage is 1.5V; without a fan it drops to 1.5V. I wouldn’t change timings unless necessary.
Technically, XMP is an overclock and may succeed with slight adjustments—possibly requiring a bit more voltage (around 1.38V). It applies to slots 2 and 4 on the motherboard, along with A2/B2.
I haven’t used XMP in years and still perform manual RAM overclocks.
Here’s a revised version of your text:
I’ve been experimenting with PBO, auto-overclocking, and curve tuning, and I feel like I’ve fine-tuned my system well. I’m consistently hitting over 5000 MHz across all 12 cores, with a Cinebench R23 multicore score of 22312—something I finally managed to install thanks to Maxon’s help. I also managed to squeeze out 3600 MHz from my memory by raising the DRAM voltage to 1.36 (up from 1.35). All this while keeping thermal cores at a max of 80°C on CCD1 and 76.5°C on CCD2.
My best results came when the EDC was set to 119, which was 103% of what’s available. The PPT and TDC varied between 99% and 100% under stress. If I reduced the EDC to 99%, my scores fell into the mid-21K range on Cinebench R23 and my top frequencies dropped to around 4800 MHz. A few others on YouTube reported similar gains using the Ryzen 5900x. I noticed improved performance when I set PPT, TDC, and EDC to the motherboard defaults instead of manually adjusting them. My Cinebench R23 score increased by about 200 points, and my temperatures dropped by roughly 4 degrees. I think running at slightly lower temps gives better results. Some people also shared using motherboard settings on YouTube, so I decided to give it a try.
For now, I’ll stick with PBO, auto-overclocking (200 MHz), and curve optimization. I ran the CO per core, which wasn’t too bad—only three cores needed tweaking to achieve the highest undervolt. Perhaps in the future I’ll explore manual overclocking. I’ve saved all my settings in a BIOS profile. Two weeks ago I was panicking about having to enter my BIOS, but now I feel confident enough to make adjustments. Lol. Thanks for your support!