sqwerty: If you look at the image, you can see few things. First, the cinebench test was still ongoing at the time of the snapshot. You can tell by the unfilled squares. I took it during near the end of the test to show CPU temps. Also, you don't get as score before the test has finished. Second, if you look at the Ryzen Master window, you can see on the CPU graph that only one or two cores have been used the majority of the test. Each core is color coded.
Also, thank you for you input for the 3700x. There have been test on the 3600x, 3700x, and 3800x showing them hit their max boost. But, from what I've read it depends on the motherboard. That could be different now. I personally believe that AGESA 1.0.0.3 ABB reduced MAX Boost on at least the 3900x if not all 3000x due to high voltage trying to hit the Max Boost.
Furthermore, this is a test/result thread. If you want to comment please follow the guidelines above and post your results. Now, it's for 3900x but if you wish to contribute for a different core, that's fine. You are welcome to add theories and speculations in your post. But, you post needs to have a test and result. Thank you.
I wan to thank ajlueke and shinkojiro for those reports. That is amazing work.
Tried running some games to see if that changed anything. This is from running Monster Hunter World for an hour or so. My results are about the same.
Cannot reach 4.6Mhz on Max Boost. Test with Auto Over Clock set with Ryzen Master.
MB: Asus Prime x570 Pro BIOS: ver 1005 (AGESA 1.0.0.3 ABB), RAM: G.SKill 3200c14. Highest MAX Boost: 4.253
Notice here it "boosted" 3 cores. With Cinebench, CPU-Z, and SuperPI it only boosted 2. It also seems to be a consistent boost. None reaching 4.6 but above 3.8. I'm curious why 3 cores and not 2 or 4. Also, my CPU temp stays around 67C which is around the average for all of my single core test. I would guess that temp is not an issue. I've been told that Farcry 5 bench is a good game test for Max Boost. I'll try it out later.
*Addition:
Tried FarCry: New Dawn Bench 3 times. Best numbers I got are below. Same issues as above. Cannot reach 4.6 Mhz Max Boost. I can't even get 4.4 it seems.
Just to expand on my initial testing here. I already indicated that a Ryzen 9 3900X allowed to boost on it's own, will add almost 1.5V to the CPU, but somehow behaves as though 1.3V were being applied. To illustrate that, I manually set my core to 4.5 (max boost achieved) and 1.3V and obtained the identical Cinebench score.
Now, what happens on the other end of that? What if I set my core to 4.6 GHz manually, and also manually apply 1.48V like precision boost does? Can I hold 4.6 GHz in that scenario?
The answer, is yes. Manually boosting only core 02, and giving it 1.48V, 4.6GHz is held rock solid during the Cinebench single core test. I also see an expected performance uptick in the final result.
So, from these data, we know:
1. My fastest core on my 3900X is perfectly capable of 4.6GHz is high voltage low current situations.
2. Precision boost appears to appropriately apply the voltage needed to maintain a 4.6 GHz boost.
3. Somehow, when the voltage is applied by PB, the maximum boost isn't reached, instead the processor behaves more like 1.3V is applied.
4. 1.3V is what the processor sets for itself under a high current situation (all cores loaded).
So it would seem, based on these data, that when PB is used, the high current voltage limit (1.3V) is still being applied in the low current situation, even though Ryzen Master shows a higher voltage. Or, the voltage is actually being raised, but somehow isn't being used to do any additional work.
You can probably do it with less voltage than 1.487, unless you've tried already. I get 4.6 with 1.4625. AMD overvolt FTL. And stock configuration is definitely screwing up the applied voltage, not only by overvolting in the first place. As I said before, AMD's algorithms are being too strict when it comes to boosting. It demands 0 load being on any other core, and with all that core bouncing its almost never 0 load on any other core. If you ask me these are perfectly capable of 3-core boosting at a minimum to the advertized clocks. 3950x should be capable of 4-core boosting. And, this is primarily because 1 CCX at the same speed is a perfect condition. 3x cores for 3900x and 4x cores for 3950x. I'd say peoples 3600/3600x and 3700x/3800x should be capable of 3/4 core boosting to the advertized as well. That is ofc as long as its limited to 1 CCX, tho I have yet to test with 1 CCD and 4.6 in all cores on 1 CCD.
Cinebench limited to 3 threads, loading up all 3 cores on CCX0 to 4600mhz @ ~50c
I've discovered something of grave importance. Ryzen Master doesnt change the Vcore, it only changes the VID. This means if you start at say 1.4v and 4.4ghz, but decide to set it to 1.05v and 3.8ghz in Ryzen master, the actual Vcore remains unchanged at 1.4v. This results in higher power usage and higher temperatures than you should get had it changed the actual Vcore.
For me, since I have Asus AI suite, I am able to use that to change the actual Vcore voltage in windows. It, funnily enough, cannot change the clock speed of Zen 2 processors. So In order to properly test different speeds and their power draws/temperatures without having to go into bios, I have to use both Ryzen master and Asus AI suite.
From having a Vcore setting actually to 1.4625v, I see my single core temperatures has risen from ~42c to ~46c, similar to the PBO temperature I get, and All-Core temperature of my manual has risen to ~82c.
Now, I have to test one more thing, for believe STOCK operation of the processor, including PBO and auto-overclocking, is NOT changing the Vcore dynamically, but only the VID. This in turn results in higher than should-be temperatures for the operating speeds we are getting. The VID which i believe is set to 1.5v.
Ok tested and it appears the Vcore is indeed being dynamically changed as well. So its all about how we set manual overclocks. Without a tool to set Vcore in windows, Ryzen master usage imo is pretty limited as we will end up with incorrect power draw and temperatures.
4.65ghz for single core for this geekbench 5 run.
Hi
Cannot reach 4.6Mhz on Max Boost. No changes in bios except fan curve and enabled XMP profile on RAM
My config is following
Ryzen 9 3900X MSI x570 GAMING PRO CARBON WIFI (AGESA 1.0.0.3) G.Skill 3600 MHz CL16 Noctua DH15 SE-AM4
IDLE
temp around 40°C and CPU voltage 1,1-1,2V
LOAD single thread
Prime95
Max clock: 4333Mhz
Temp: 80°C
CineBenchR20
Max clock: 4440Mhz (Average 4410 Mhz)
Temp: 65°C
LOAD all core
Prime95
Max clock: 4174Mhz
Temp: 60°C
CineBenchR20
Max clock: 3940Mhz
Temp: 71°C
Operating System | Microsoft Windows 10 Pro (64-bit) |
Processor | 3900x w/ macho rev b cooler, CM h500/mesh case |
Motherboard | Gigabyte Technology Co., Ltd. X570 AORUS PRO WIFI |
Memory | 32768 MB DDR4 SDRAM 1700MHz (3400 c16) |
BIOS | American Megatrends Inc. F4 |
Test - CB20
On power plan
- Ryzen High Performance I can't hit over 4309 MHZ, CB20 SC score = 489
- Balanced I can't hit higher than about 4288 in single core but it also doesn't seem to swap between 2 cores as much
I included the balanced run to show the temps while running (my high perf run had ended when I took the screen shot). Generally bounces between 56 and 59c
All settings are basically bios with load optimized defaults, load XMP for ram, set volts to 1.35, up ram ratio to 34 as it's stable there for a couple hours on memtest and other tools.
Geekbench:
Single-Core Score | Multi-Core Score |
---|---|
5569 | 45534 |
Geekbench 4.4.2 Tryout for Windows x86 (64-bit) |
Cannot reach 4.6Mhz on Max Boost. Test with Auto Over Clock set with Ryzen Master.
MB: Asus Prime x570 Pro BIOS: ver 1005 (AGESA 1.0.0.3 ABB), RAM: G.SKill 3200c14. Highest MAX Boost: 4.345
Geekbench 5.0 Beta 1
Got a little higher on single core boost than most tests. Can't seem to break 4.4 though. Full results here. System manufacturer System Product Name - Geekbench Browser
Saw a good video by level1techs today. He's findings seem to be in line with ours. Ryzen 3000 Update! Boost Clocks, Voltages, and More! - YouTube
I recommend watching it.
Updates over on the gaming blog. Should have new UEFI to test in 2/3 weeks.