-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
FRONTPAGE AMD Ryzen 9 3900X and Ryzen 7 3700X CPU Review
- Thread starter Overclockers.com
- Start date
- Joined
- Nov 21, 2018
Guy thought he was cool on hwbot with a 4684MHz oc...with SMT off. So i just tossed up a 4698.9 with SMT on lol. I can turn SMT off and get it a bit higher. But i still got a little bit left in the tank with it on before i get to where im not at least 4 hours prime stable.
I love this cpu. Last cpu that i had that clocked this good was my Q6600. My Athlon's were always in the in the typical range. My i7 4700k was ok. My 6700k was the worst clocking turd off all time. 1.45v and i couldn't even get past 4.4.
I love this cpu. Last cpu that i had that clocked this good was my Q6600. My Athlon's were always in the in the typical range. My i7 4700k was ok. My 6700k was the worst clocking turd off all time. 1.45v and i couldn't even get past 4.4.
I'm not sure if it's memory or IMC limit but the same max clock works on Gigabyte X570 I Pro and ASRock X570 Extreme4. 1st screenshot was on GB, 2nd one on ASRock.
Micron E @4800 C18. My Micron kits work better on X570 than Samsung. Better like easier to stabilize and OC higher. The 4800 CL18 is a bit better than the 3600 or 3733 CL16 (can't compare results on different processors) but of course will be harder to set.
On both boards stability test is passing 2h+. I forgot to take a screenshot on the 2nd one and I had no time to run it again.
One more thing which was actually confirmed by a mobo manufacturer (based on their lab tests on mulitple CPUs). Some Ryzens may stuck at default 3200 memory clock and I actually had that situation with one CPU. Simply, 3200 is 100% stable, 3333+ won't even boot. Once I switched CPU then I was able to make 4800. Don't be surprised to see problems like that but I guess it's not affecting really high % of CPUs.
Micron E @4800 C18. My Micron kits work better on X570 than Samsung. Better like easier to stabilize and OC higher. The 4800 CL18 is a bit better than the 3600 or 3733 CL16 (can't compare results on different processors) but of course will be harder to set.
On both boards stability test is passing 2h+. I forgot to take a screenshot on the 2nd one and I had no time to run it again.
One more thing which was actually confirmed by a mobo manufacturer (based on their lab tests on mulitple CPUs). Some Ryzens may stuck at default 3200 memory clock and I actually had that situation with one CPU. Simply, 3200 is 100% stable, 3333+ won't even boot. Once I switched CPU then I was able to make 4800. Don't be surprised to see problems like that but I guess it's not affecting really high % of CPUs.
- Joined
- Aug 20, 2003
One weird thing I just now noticed (shows how little I've been able to play with this desktop) is that even though I set CL to 17 it runs at 18:
Humm, if I set it to 16, it goes to 16:
Humm, if I set it to 16, it goes to 16:
Last edited:
Disable gear down in BIOS and it will run at CL17. It's a standard option for all Ryzens. On the other hand, you can set CL16 so there is no problem 
Btw. I was able to run HyperX 4000C19 at 4533. Only some quick runs but was working fine. I guess it can make some more as on Intel I was able to pass 5200.
Btw. I was able to run HyperX 4000C19 at 4533. Only some quick runs but was working fine. I guess it can make some more as on Intel I was able to pass 5200.
Last edited:
- Joined
- Aug 20, 2003
I'm trying to decide on speed vs keeping it at 1:1. My guess is unless I could hit ~4800, it's probably better to keep it at 1:1?
I'm having the same issue. I'll have to see if I can find this "gear down" option. Thanks Bart.
If you want to compare it looking at AIDA64 then check memory copy. It usually shows performance in applications while read/write is only synthetic bandwidth. There are tips after memory benchmark which are telling that (in the latest versions).
For me it usually looks like 3600/3733 CL16 = 4500/4700 CL18. Shawn mentioned that for him 3600 CL16 is about the same as 4400 CL17 but he was testing it on a different processor. There is a difference between 6/8 and 12 cores so the answer what is faster is not really clear. For you it can be lower memory clock like 4400-4500 at tighter timings, as memory performance is generally higher than on 6/8 cores because of higher internal bandwidth etc. (2 CCX and additional link). Supposed to be only write bandwidth but on the screenshots you can also see 3-5GB/s higher read and copy.
3600-3733 1:1 is optimal and much easier to set. 3800 would be better but I find it much harder to stabilize. Maybe just compare some settings in the software which you are using. I don't expect any significant difference.
Silicon Lottery has posted their results so far of Ryzen 3 3700x/3800x/3900x results.
https://siliconlottery.com/pages/statistics
These things are all absolute duds when overclocking.
It's funny, after all the talk about a golden chip, bigazn's is the golden sample of golden samples if 6% only reach 4.2 GHz, LOL!
The 3700X I have does 4.25 GHz at 1.33V and is topped out thermally with both a MEH 2x120mm (EVGA 240 CLC) and a good 3x120mm (Corsair H150i)...as others have mentioned, such little difference, even with good cooling.
https://siliconlottery.com/pages/statistics
These things are all absolute duds when overclocking.
It's funny, after all the talk about a golden chip, bigazn's is the golden sample of golden samples if 6% only reach 4.2 GHz, LOL!
The 3700X I have does 4.25 GHz at 1.33V and is topped out thermally with both a MEH 2x120mm (EVGA 240 CLC) and a good 3x120mm (Corsair H150i)...as others have mentioned, such little difference, even with good cooling.
- Joined
- Aug 16, 2009
I've never tried anything that low voltage ..... I will when I get a chance tonight or tomorrow. Though I think if they put more voltage to these chips we might see more MHZ .... then again maybe not. We have enough guys here with chips maybe a few will post their stable clocks at 1.25v for fun / curiosity.
- Joined
- Mar 7, 2008
Interesting, looking back at my 3700X notes, I settled 12 minute Prime 95 128k FFT stable at 4125 at 1.30v set (maybe 1.24v load). So that's kinda right in their range too.
I've still not been able to confirm or deny a rumour that on the 3900X one CCD is a good bin, the other is a lower bin. So if going for all core, you're going to be limited by the lower bin CCD if this is the reality. I still don't have any intent to get one in short term, but if anyone has one it would be interesting to see if the two CCD OC differently.
I've still not been able to confirm or deny a rumour that on the 3900X one CCD is a good bin, the other is a lower bin. So if going for all core, you're going to be limited by the lower bin CCD if this is the reality. I still don't have any intent to get one in short term, but if anyone has one it would be interesting to see if the two CCD OC differently.
- Joined
- Nov 21, 2018
I tried it and I gained nothing in terms over more clock. Some people have seen big gains but nothing here.
And those results are horrible.
1.25 I can pull 4.35. I know my CPU is good but there not even pushing it. I can hit 4.6 stable with ease but your not clocking a 3900x on an aio. I ran a Hi115 pro and just pbo with a .1v offset it was in the upper 80's and benchmarks would hit 90+c. You gotta have a high-end cooler. an AIO just won't cut it for a 3900x outside of stock clocks and even then that water gets really warm.
Now I seen multiple people hit 4.5 on them. And a few hit 4.7. I don't know how stable the 4.7's were but last night after I did my hwbot submission I forgot to go back to my daily driver clocks and I played wow for 3 hours why I had YouTube running in the background before I was like ohhh that's why it's warmer. It's clocked to the moon.
So is there highest binning able to hit the 4.5 golden chip clocks were starting to see? Or was it a use a cheap aio and that's all she's got even tho the. CPU has more left. If your paying that much for a 3900x binned your not using an aio.
Last edited:
- Joined
- Mar 7, 2008
I'm running a low end 240mm AIO on my 3700X stock, and that's hit a max of 58C over the last 285 hours running 120k FFT tasks one per core. It's doing a lot better than a Noctua D9L on my stock 3600, which (with thermal past restored, no more graphite pad) is hitting 85C under same work. Just how much cooling do these need?
I want to put a chiller on it some time but I'd need to get an AM4 mount for my block, or just get another block... BTW in my book, I have to separate out Prime95 FMA stable, and non-AVX stable as two separate categories. I'm only really interested in the first one, and tolerate the 2nd one for benching only unless it can be set separately with AVX offset, not an option on Ryzen.
I want to put a chiller on it some time but I'd need to get an AM4 mount for my block, or just get another block... BTW in my book, I have to separate out Prime95 FMA stable, and non-AVX stable as two separate categories. I'm only really interested in the first one, and tolerate the 2nd one for benching only unless it can be set separately with AVX offset, not an option on Ryzen.
It's tough to compare with you considering the testing isn't really default/preset P95.
My testing is done with AIDA64's default setup and with a 3x120mm AIO and 26.xC room, I hit about 85C after 30 mins. If you would like to compare, run that/something the majority use to see.
- Joined
- Mar 7, 2008
P95 small FFT with a recent version is close enough. I don't find Aida64 to be that stressful on the CPU, so only ever use it for ram. And I think I've found better anyway.
- Joined
- Dec 18, 2000
These things are all absolute duds when overclocking.
I thought we already knew this. They run nice and fast with the RAM set to XMP and PBO enabled but aren't much fun to play with.
I have no idea why my better cooling using a 'less stressful test' in AIDA64 yields 30C more. This is across multiple boards, mind you. I can't say I like the mount for AM4 on the H150i, but right now I'm using the Intel 115x mount on the ASRock X570 ITX board and the temperatures are remarkably similar... so its not the mount it seems. TIM applications are consistent.
P95 Small FFT yields temps to the mid 90s.
EDIT: It honestly feels like the cooling on these is, like many complained about Intel, just 'good enough', but not great... it's weird.
We do... but this confirms it with even more samples with more empirical testing. Just adding more information, Dave!
Last edited:
- Joined
- Jul 31, 2005
You have a nice chip.
I don't think mine will even do 4.35Ghz (all core/smt high stress) with 1.35v (+.10v) above yours.
I think the Silicon Lottery results are fairly representative of what is actually out there. I would think that they are stressing with the latest version of Prime95 small fft (using full avx) without an avx offset. It would be interesting to actually know what their test bed setup is for binning Ryzen 2? (Motherboard, memory speed/timings, dram kit and cooling?) And also what kind of LLC was being applied? We know the set BIOS vcore for the binning levels but we don't know what the actual idle and load vcore readings were for those various binning steps/levels?
They may be testing with a higher end AIO? And I agree with you that the Ryzen 3000 series can benefit from big water cooling maybe the 3900X even more so... I'm thinking most (or at least more) of the better quality chiplets are going towards the 3900X.
On a side note...
I noticed that AIDA recently up graded their test suite in the latest version to include more high stress AVX components... Probably still not as hard/sressful as the most recent Prime95 small fft (full avx) but AIDA cpu/fpu (only) can bring up the temps.
EDIT: Lately I've been running Rosetta@home all core/smt on my chip with the Silicon Lottery middle bin settings: [email protected] in BIOS with a lower LLC setting. Temps with custom water and a thick 360 rad are running from the mid 50's to lower 60's depending on what type of work units are being crunched.
Last edited:
Their QVL information is listed: https://siliconlottery.com/pages/amd-matisseYou have a nice chip.
I don't think mine will even do 4.35Ghz (all core/smt high stress) with 1.35v (+.10v) above yours.
I think the Silicon Lottery results are fairly representative of what is actually out there. I would think that they are stressing with the latest version of Prime95 small fft (using full avx) without an avx offset. It would be interesting to actually know what their test bed setup is for binning Ryzen 2? (Motherboard, memory speed/timings, dram kit and cooling?) And also what kind of LLC was being applied? We know the set BIOS vcore for the binning levels but we don't know what the actual idle and load vcore readings were for those various binning steps/levels?
They may be testing with a higher end AIO? And I agree with you that the Ryzen 3000 series can benefit from big water cooling maybe the 3900X even more so... I'm thinking most (or at least more) of the better quality chiplets are going towards the 3900X.
On a side note...
I noticed that AIDA recently up graded their test suite in the latest version to include more high stress AVX components... Probably still not as hard/sressful as the most recent Prime95 small fft (full avx) but AIDA cpu/fpu (only) can bring up the temps.
Not sure how they take Vcore readings though. I imagine them taking the load voltage regardless(?).
To add to your AIDA64 bit, I'm using the default which is CPU/FPU/Cache. I haven't tried CPU and FPU only on Ryzen. That said, it was 10C+ warmer last time I used it which puts it right up there with P95 Small FFT in the latest versions.
Similar threads
