I use PBO, Ram is 3600 at 10 Minute test was within 100pts on Multi and the same for SC
Check out my score and details and try them out. I am sure you will achieve better score than you already did.
Either that or the RAM does not make any difference after all. As I am using the cheapest one I could get. But I doubt the RAM doesn't make difference. You are losing points because of throttling. Try the offset voltage, for me -0.0875V turned out to be the sweetest sweet spot. Lower or higher reduce the score. One due to lowering the frequency, the other for increasing the temps. I am using H105 water cooling in push/pull configuration.
Check out my score and details and try them out. I am sure you will achieve better score than you already did.
Either that or the RAM does not make any difference after all. As I am using the cheapest one I could get. But I doubt the RAM doesn't make difference. You are losing points because of throttling. Try the offset voltage, for me -0.0875V turned out to be the sweetest sweet spot. Lower or higher reduce the score. One due to lowering the frequency, the other for increasing the temps. I am using H105 water cooling in push/pull configuration.
Ram isnt making any difference here and even if it did you are using 2x16gb dual rank 3000 which is almost fast as his 2x8gb single rank 3600.
Check out my score and details and try them out. I am sure you will achieve better score than you already did.
Either that or the RAM does not make any difference after all. As I am using the cheapest one I could get. But I doubt the RAM doesn't make difference. You are losing points because of throttling. Try the offset voltage, for me -0.0875V turned out to be the sweetest sweet spot. Lower or higher reduce the score. One due to lowering the frequency, the other for increasing the temps. I am using H105 water cooling in push/pull configuration.
Thanks, I will try that. I use a ARCTIC Liquid Freezer II 240
When I see 5800X single core and this claimed to be Intel score I completely distrust the new Cinebench iteration. I doubt a 28W is capable of nearly achieving the same single core performance as the new AMD architecture. I am sure the results "from Intel" are fake. Fortunately it's easy to remove them so I can only see the real scores
Looks good!
why is that, Rocket lake is based on new IceLake arch with double L2 cache, which is on par with new Zen3, so 512KB and extra L1 which has now higher L1 then Zen3 (64KB Zen, RL 80KB), plus new optimizations and it is what it is.
I think desktop version will surpass 1600 mark, just like Zen3 does.
why is that, Rocket lake is based on new IceLake arch with double L2 cache, which is on par with new Zen3, so 512KB and extra L1 which has now higher L1 then Zen3 (64KB Zen, RL 80KB), plus new optimizations and it is what it is.
It is interesting how a 28W 10nm MOBILE CPU Core i7 1165G7 is on par with 65W 7nm DESKTOP ZEN 2 in CB20, but on par with 65W 7nm DESKTOP ZEN 3 in CB23. They must be on something. The test takes 10 minutes, there's no chance on Earth a notebook can maintain turbo for that long. Additionally their turbo is 4.7GHz, which according to your fantasies means the magical Intel mobile 10nm CPU is on par with ZEN 3 IPC wise (or ZEN 2, depending on the moment in time when Intel decided to finance MAXON).
3900X
PBO set to enabled
Fmax Enhancer set to enabled
(ALL other PBO settings left at default)
XMP enabled (3600MHz CL16)
BIOS: Crosshair VIII Hero Beta BIOS AMD AM4 AGESA V2 PI 1.1.0.0 Patch C
Global C State enabled
Ryzen High Performance Power Plan used
Everything else in BIOS at default values.
With this AGESA I have noticed a lot higher boost clocks with more or less default settings, now boosting to 4.25-4.3GHz on its own during all core loads, whereas before it boosted to 4.0-4.1GHz. Single core boosts hitting around 4.4-4.5GHz and momentary boosts to 4.65GHz.
I suggest you turn on PBO and set offset voltage like I did. It makes great impact on long term boost and frequency maintanance. I did the whole test, almost the same results as he did. I guess he uses better ram with better timings, or maybe a negative offset, I don't know. He needs to share more info.
Yes is seems PBO was not fully on for me (I had turned it on in Ryzen Master but appears it didn't work). I turned it on in BIOS and now see OC Mode of "Auto Overclocking" in Ryzen Master.
Re-did my runs and got a small increase from 17115->17768 and 1252->1266
I didn't touch voltage though yet.
Note I am only using the stock cooler that came with the 3900X and the temp while running the multi-core was about 83C, so that is obviously affecting all boost. I have plan to get a proper AIO water cooler but can't make up my mind which one yet.
EDIT: I changed the VSOC from what is was (1.025V) to the offset mode with (-0.0875V) and score went slightly up to 17813 pts. I notice the temperature of CPU did not go down at all - still over 83C so obviously this voltage decrease was still not good enough to lower temps with stock cooler.
Here is the test run once, without 10 min duration set as the previous time I did the test. Slightly higher numbers as the frequencies were to be maintained for shorter period of time, thus less heat and higher avg frequency was achieved, obviously from the results.
Same settings as before:
3900X @ stock + PBO
auto vcore with -0.0875V offset
everything else is stock
32GB GSkill Aegis DDR4 3000 XMP
Just want to mention that the results from @CPC_RedDawn are great, but are with ASUS specific PBO feature(Fmax Enhancer ) that works similarly to CTR (not in the crucial methods, but in terms of the results and OC dependent on the silicon lottery )
Here is the test run once, without 10 min duration set as the previous time I did the test. Slightly higher numbers as the frequencies were to be maintained for shorter period of time, thus less heat and higher avg frequency was achieved, obviously from the results.
Same settings as before:
3900X @ stock + PBO
auto vcore with -0.0875V offset
everything else is stock
32GB GSkill Aegis DDR4 3000 XMP
Just want to mention that the results from @CPC_RedDawn are great, but are with ASUS specific PBO feature(Fmax Enhancer ) that works similarly to CTR (not in the crucial methods, but in terms of the results and OC dependent on the silicon lottery )
I ran the test with the default 10 minute duration and get the same score (20 points less tbh) as my 3900X is cooled by a 360mm AIO (arctic) so it stays cooler and boosts higher for longer. My 3900X is also a launch CPU too so silicon lottery probably wasn't the best for me.
And finally Fmax Enhancer only effects single core boost not multicore. As stated in the BIOS.
I ran the test with the default 10 minute duration and get the same score (20 points less tbh) as my 3900X is cooled by a 360mm AIO (arctic) so it stays cooler and boosts higher for longer. My 3900X is also a launch CPU too so silicon lottery probably wasn't the best for me.
And finally Fmax Enhancer only effects single core boost not multicore. As stated in the BIOS.
You are wrong. The Fmax Enhancer main goal is to boost the multicore performance, not single core. Just like CTR does. That's where it can optimize things, per CCX. The single core frenqencies are what they are and if they depend on the cooling then you'd see much better results than mine(standard 3900X scores). The only reason you achieved better multicore scores is the use of Fmax Enhancer and that's why I brought that up so other 3900X users pay attention to it before they start worrying about their CPU MC performance.
ASUS ROG Strix-E Gaming (BIOS 2606)
3800X @ stock + PBO 3
vcore @ -0.1000V offset
2x8=16GB GSkill DDR4 3600CL15@CL14 via DRAM calculator
10 mn run :
Single core : 1310 => 4425-4450MHz
Multi cores : 12535 => 4100MHz => 73°C with Corsair H105 AIO
I think you should try vcorce @ -0.0875 or -0.0750. The max boost of 4.5GHz is not achieved because the resulting vcore with that offset applied is too low. Maybe try and see.
You are wrong. The Fmax Enhancer main goal is to boost the multicore performance, not single core. Just like CTR does. That's where it can optimize things, per CCX. The single core frenqencies are what they are and if they depend on the cooling then you'd see much better results than mine(standard 3900X scores). The only reason you achieved better multicore scores is the use of Fmax Enhancer and that's why I brought that up so other 3900X users pay attention to it before they start worrying about their CPU MC performance.
If what you say is true then they need to reword their definition in the BIOS
Just want to mention that the results from @CPC_RedDawn are great, but are with ASUS specific PBO feature(Fmax Enhancer ) that works similarly to CTR (not in the crucial methods, but in terms of the results and OC dependent on the silicon lottery )
I don't think this "Fmax Enhancer" is on all ASUS boards either. I was looking for it on my Hero VIII and it was not found even using F9 search. I see screenshot above and it is missing on my same page of BIOS.
3900X PBO set to enabled Fmax Enhancer set to enabled (ALL other PBO settings left at default) XMP enabled (3600MHz CL16) BIOS: Crosshair VIII Hero Beta BIOS AMD AM4 AGESA V2 PI 1.1.0.0 Patch C Global C State enabled Ryzen High Performance Power Plan used Everything else in BIOS at default values. With this AGESA I have noticed a lot higher boost clocks with more or less default settings, now boosting to 4.25-4.3GHz on its own during all core loads, whereas before it boosted to 4.0-4.1GHz. Single core boosts hitting around 4.4-4.5GHz and momentary boosts to 4.65GHz.
Here is the test run once, without 10 min duration set as the previous time I did the test. Slightly higher numbers as the frequencies were to be maintained for shorter period of time, thus less heat and higher avg frequency was achieved, obviously from the results. Same settings as before: 3900X @ stock + PBO auto vcore with -0.0875V offset everything else is stock 32GB GSkill Aegis DDR4 3000 XMP Just want to mention that the results from @CPC_RedDawn are great, but are with ASUS specific PBO feature(Fmax Enhancer ) that works similarly to CTR (not in the crucial methods, but in terms of the results and OC dependent on the silicon lottery )