3900x First impressions. Precision Boost Overdrive vs. Manual OC on X370

Gill.. · Jul 11, 2019

Bravo, AMD engineers!

X370 Asrock Killer SLI/ac
Gskill TridentZ 3600 AMD SKU - for now at 3200 CL16/18/18/18/36
Thermaltake 280mm Rad AIO with lots of bling
IC Diamond 24K (my personal favorite) thermal paste

So with a manual overclock, my R7 1700 topped out rock solid stable at 3.9 anywhere around the 1.38v mark. With the new Gskill I bought for Ryzen 3000, I got the same memory OC on the X370 with the newest BIOS's (5.3 and 5.4). It was a great chip, I loved it from day one when I got it in March of 2017. As much as I had loved my BE 550 with it's 2 free unlocked cores (because who doesn't love free). I knew the 1700 would win my heart. It did! From March of 2017 I knew one thing. Wait 2 gens...get a 12 core. And that's what I did.

Don't get me wrong, I'd love a 16 core 3950X...but for a $140 MSRP X370 board..I knew that would be asking a lot..12 cores...bring it.

As I'm sure most of you guys reading this waited with anticipation the last year or so...I was NOT waiting for reviews of this chip. It was going to be mine ASAP. It was a struggle getting it, as I'm sure you'll all have your own stories..but here it is day 3..in my hands...and I got home at 9pm tonight and knew it would be a long night.

Cleaned my rig, took out the 1700 (going to my son to bring his B350 to it's full potential, I'll use the 2200G as a backup chip)..and in it went.

I'm sure I don't have to mention to you guys on OCN...but just so it's said....UPDATE your BIOS FIRST!! I also put it back to stock settings before shutting it down. The boot process was long...way long...the suspense killed me (like heart in throat)....and into BIOS we went.

I started with stock settings. Ran Cinebench, Geekbench...and WOW..impressive. Almost 3100 on R15, Geekbench impressive too.

Then I went to Precision Boost Overdrive. From all I've read and watched, it seemed the way to go. The BIOS unlocked a TON of new options with the 3900 that weren't there with the 1700. I kind of let it do it's thing...and back into Windows. Cinebench and Geekbench a bit better, but obviously I'm still not done. I wanted to put first impressions here, so we can all learn together.

What's amazing is that a.)It's totally insane out of the box b.)I'm totally hitting 4.6 plus on multiple cores on Boost*see attached pic* c.)I'm averaging 150% to close to double the computational power of the 1700.....AT 65 watts in most cases.

Let me give you a frame of reference. My 1700 at 3.9 would hit 140 watts at 100%..temps would be mid 60's with the AIO. 1680s to 1730 Cinebench (stupid Windows updates!). 600 hash/sec at best on XMR Stak (I'm a dirty miner!). XMR uses 8 of the 16 threads..anything above that yielded no addition hashes.

R9 3900X - 4.1-4.2 when most or all cores in use. 5 or so cores regularly hit that 4.6 plus. I'm at 78% usage as I type..I have 18 of the 24 cores mining XMR at 1250 Hash/sec - at 65 watts...at 55C...same AIO.

Double the computation in some instances, 1.5X at minimum, half the energy in normal scenarios..at less temps...on PBO!

I tried manually OC'ing to 4.3 all cores. Got into windows, tried Geekbench. Dead..back to PBO. Obviously I'll keep trying. But I don't know if I want to. If multiple cores can hit 4.6 in say a gaming scenario, while the rest stay super efficient...and half the energy my 1700 was using..why would I bother. I'd rather spend the time on the memory clocks and timings (Help me Buildzoid, you're my only hope!).

To be fair to my 1700, the 3900X will get to 120 Watts at 100% usage. But really, there's NO comparison between these two chips...It's such a leap forward, I'm truly amazed (I knew I would be, but wow).

One more nail in the coffin. My Vega 56 hashes at 1850 hash/sec...pulls 157 watts. The 3900X...1250 hash/sec at 65 watts. Enough said. The future is here gentlemen/ladies. PBO is beating what we can do..but OCN will still exist, because we'll need to find the fine balance with the memory and the boards.

But if I could leave you with one thing. This is on a mid range X370. VRM's are decent (Buildzoid would still not approve)...but don't let any of these naysayers out there scare you into an X570 that some brands are pumping WAY too much energy into the chips/VRMS and RAM just to try to overclock something that has been superseded by superior engineering (PBO).

I wouldn't be a member here if I didn't keep trying. But the performance I'm seeing on my 3900X is making me a true believer in PBO. Now to tweak that (there are a bunch of settings in that alone as well)...so let the discussion commence!

To anyone waiting for backorders...STICK IT OUT. I'm entirely happy with mine! I could have flipped it on Ebay today for $150 profit. But this is a chip I've been waiting for, for two years. And I'd officially have to quit OCN if I were to pull that sort of blasphemy.

rdr09 · Jul 11, 2019

gonX · Jul 11, 2019

Have you looked into whether undervolting would allow PBO to boost higher?

Really been eyeballing the 3900x myself so these findings are very helpful. I did also get the impression that OC's would not be worth it because PB/PBO seemed to do the job very well.

Gill.. · Jul 11, 2019

Undervolting Vcore? I have that set to Auto now - let PBO do it. So one setting I have is PBO Scalar at 10X. From what I can tell, it will boost the voltage where needed for max frequency for longer durations.

There was another setting that enabled the Max clocks to go beyond where AMD supposedly allows. Definitely seems like 5 of the cores are stronger than the rest (4.6 plus), with the rest maxing out around 4.3ish. Another notable is that the main clocks seem to be 4.2 at all times with an idle wattage of 3.8 watts....yes, you read that right.

Part of me says, there's the potential gain - let's get them all to 4.3. But if I do that, I lose PBO - and those spikes above 4.6...so is it really worth it for the extra 100 Mhz for all of them - and fight for stability - or be happy with the silicon that I have. 5 cores (so 10 threads?? I need to find something that will show me ALL threads and their clocks...CPUID's HW Monitor is just showing me the cores as a whole) that are strong seem to be a good pull on the silicon lottery.

Now the RAM...that's being a punk when I try to bring it above 3200. That will take some work. I really thought the memory controllers would work like magic...

Too funny though..I ran Division 2, SOTTR, Timespy and Firestrike and GTA V for some quick game benches...I don't think any of them topped 30% usage on the CPU...I'm TOTALLY GPU bound now with my Vega 56 (which oddly was averaging 1715 or so core clock at 50% power limit on GTA V)...only game that was holding it above the 1700 mark consistently.

AIB 5700XT's can't come fast enough...

So, Go X370/X470 B350 and B450 if they're decent. No Chipset fans, no ridiculous pricing, no PCIE 4 marketing hype. Way less heat and wattage from not only myself, but from what I'm seeing with others.

How the cores and clocks behave is wicked interesting to me...especially with the power consumption. Hopefully we'll get more info as people start getting them on the ground.

Nighthog · Jul 11, 2019

The Memory support on the first launch B350/X370 was really crappy on some motherboards, had nothing much to do with the IMC on the CPU:s had as everyone raved from my own observations.

It was that the boards that could barely manage 3200Mhz speeds. I think the cheaper ASRock boards where some with major issues and another being the even more expensive ASUS X370 Prime Pro. (X470 version got fixed though)

MSI and Gigabyte had better support from the get go if I recall. Though this was mostly a AGESA issue as well with enabling the better memory speeds came quite a bit late in the game.

Not all X470 boards were good either.

thebufenator · Jul 11, 2019

Consider adding a negative voltage offset.

Typically the CPU starts downclocking due to thermals. Less voltage == less watts == longer time at boost.

At least that was how my 2700x performed. I suspect you'll have a higher all core boost with a negative voltage offset

mark_thaddeus · Jul 11, 2019

i thought this was a great video by Jayz2Cents on OCing Ryzen... It appears that auto settings for some mobos, the voltage for CPU, I.F., memory, etc. are all way too high! Jay is going to do a deep dive and see what this really turns out, but it does show some promise!

EDIT: I know you're board is not an X570 so I'm not sure if the same settings apply to your mobo, but if you could match you infinity fabric settings (1800) and push your memory to 3600 that would be nice to see. Please note I haven't used AMD in ages (since Phenom) so I'm not familiar with 1st and 2nd Gen Ryzen bios settings specially with Infinity Fabric.

mouacyk · Jul 11, 2019

I enjoyed your enthusiasm for the performance uplift from the 1700X to the 3900X (I built a 2700 rig for 4K video editing myself), but not for PBO or any automatic overclocking mechanisms without equal options to do them manually. Of course, AMD is only following the likes of NVidia's Boost on GPU's and most motherboard's BIOS overclocking features (which employ egregious voltages btw.)

Unoid · Jul 11, 2019

asrock boards are still waiting for AGESA 1.0.0.3 BIOS's for x370 It should provide more stable 3200+ speeds on memory.

roco_smith · Jul 11, 2019

Oh boy , this is the review I was waiting for to pull the trigger on My CrossHair VI Extreme and put that 3900X cpu monster . is stupid how good marketing can convince people to buy a expensive buggy X570 board with a PCIE.4.0 that you probably never use and see working .

AlphaC · Jul 11, 2019

I don't know people are ragging on him about the comments on PBO. I bet higher gains will be seen moving to 3600 memory divider.

The AIO cooler is probably going to limit 3900X for AVX loads if PBO actually starts working , whereas memory tweaking is going to yield less of a power difference.

AMD has stated PBO will allow + 200MHz but thus far I have not seen this.

Gill.. · Jul 12, 2019

Ok, Day two starting. Wendell (Level One Techs) is still experiencing low all cores an not seeing 4.6ish. But, he's on X570's to my knowledge. To the guys suggesting Negative Offset...I'm At Level One Offset right now (which I never ever did with my 1700 after I saw the voltages it was pulling. So how would I do negative...instead of using Auto, see where it sits manually and hit minus I'm assuming? I don't think I really need to do that with my X370....see below and check the pic...voltage and temps AND watts are all super fantastic.

Stress test is at 4074-4124 on CPU-Z...at 95W. CPU and mainboard temps are all low...Attached pic WHILE CPU-Z stress is going...pics or it didn't happen.

Also note that 6 cores are poppin right around that 4.6 mark! Granted, I don't know for how long....but it's possible.

Also tried 4.35 all core...booted into windows...died on Cinebench. Went through quite a bit of settings and had Vcore at 1.4...

What would be sweet would be to be able to set a per-core clock manually. IE, get those first five or six cores up to 4.5-4.6. I'd find the sweet spot after numerous tests for each core. Then...set the remaining cores at 4.2....we'll call it GBO - Gill Boost Overdrive.

Back to hashing. So ran it at 90%CPU from 4am (thanks 3900X, I didn't need sleep)..got up late for work, didn't have time to shower...left it going. Turned AC off. Perfect real life stress test (Vega 56 crunching away too). Of course my house lost power a couple of times today. We'll try it again tomorrow. Got like 22 of the threads mining at 1450 hash/sec...still have 10% CPU for web browsing etc. Up from 600 hash/sec on the 1700 and getting oh so much closer to the Vega 56 at 1830 hash/sec.

Another interesting thing about the mining algo....since I leave 2 threads open. On CPU-Z stress it's 4074 mostly...95 watts. Hashing is 4124 constant..4149 best...at 71 watts...that's insane. Double the computation of the 1700 at HALF the power.

Messed with RAM again...still not having it. Best I've done is lower tFRC and it's sub-timings a bit. Thanks to poster above, I hope new AGESA will fix it.

DON'T FORGET TO CHECK THE PIC BELOW OF HW Monitor

Gill.. · Jul 14, 2019

Unoid said:
asrock boards are still waiting for AGESA 1.0.0.3 BIOS's for x370 It should provide more stable 3200+ speeds on memory.

Here's hoping..I tried and tried..anything I touch, and it goes dead and back to stock BIOS.

Since last post, I've messed around with the PBO settings..:

a.)Enable Only
b.)Disable (letting it run stock - WITH LLC Level 1's)
c)Manual - enabling over clock, 10X Scalar and Motherboard for the limit

Seems like C gets me the best...LLC 2 though.

DDR Voltage doesn't help the mem - gotten it to boot into Windows with 1.395 - Crash on Cinebench. Going to mess with timings (not CL - leaving that 16 - touching that has been dreadful).

And to keep all you manual guys happy.. going to try for 4.3 all core (not 4.35 - I was being a tough guy and it failed miserably). I'll re-run benches to see if there's any improvement. Will report CPU temp and VRM temps.

Speaking of VRM temps...yeah they're getting into the 70's easily when I'm jamming it full tilt (mining due to constant load of 89%). So bought two 140mm fans...one will be soley focused on the VRM's.

Plan for manual:

4.3
3200 CL16
LLC 2 for CPU and SOC
1.35 to start...but will go up a bit if need be.

Regardless of if I can get it stable. If the performance (ie Cinebench, Geekbench and hash rates) don't justify the additional voltage and temps (I'm scared of VRM temps) and wattage...I'll go back...but I'll report back to you guys either way.

Gill.. · Jul 25, 2019

So thanks to other members' threads...I figured out the RAM. The Infinity Fabric was the issue with the Asrock boards. I have stability with 3600 now. BUT...with lower scores all around.

3600 with CL18 and an IF of 1500 yields a latency of about 90. 3200 with a CL16 and I can get latency to 79 (1600 IF - but subject to WHEA BSOD..boo)

ALSO to all Ryzen 3000 people, this IF lowering has also fixed the WHEA BSOD's. Don't know if this is solely on the Asrock boards. But the threads I was reading were showing the same issues on all three gen boards X370, X470 and X570 (mine was the inclusion of the X370)..AGESA is the oldest...so maybe it will calm down with a newer AGESA (that AMD DOESN'T have to pull from mobo vendors).

So, what say you guys....keep the 3600 with the higher latency just to be a tough guy...or go back to 3200.

Playing the new PUBG Erengal map tonight, I had more lag and frame drops than ever before...BUT..could be the new map is not optimized fully...or...Vega could be messing with me (I saw GPU clocks drop randomly by 1000...not kidding).

Update = staying 3200 CL16 and I've been tightening the timing. LLC 3 seemed to let the clocks drop too low, but didn't drop the idle temp below that 28C mark. So quickly went back to LLC2

AND...got a sweet 7406 on Cinebench R20....so I'll take it...got my best Geekbench too 43114....I've gotten higher Single...but Multi is where I'd prefer to put my efforts. Stupid latency is still 90....I can only imagine what I could score if I could increase IF to 1600 or higher...

Gill.. · Aug 9, 2019

New Agesa (1.0.0.3 ABB) was released today for my Killer SLI/AC X370.

Clocks and voltages all seem the same. PBO options changed a tiny bit.

BUT....3600 RAM with FCLK of 1800 is now stable!!!!!! Tonight I'll make sure audio and stability stay the same.

Best part?? Cinebench and Geekbench are better! Geekbench just STOMPED. Check it out below! Both Single and Multi absolutely destroyed all previous attempts.

CPU temps are great, only downside is that I could only get the RAM stable at 1.45V...so much for 1.35...everything I've read assures me that should be fine. 1.43 wouldn't even work..

I'll try to lower sub-timings a bit...but I got a latency of 73-74..which was substantially better than what I was getting before. CL is 18...but the numbers don't lie...Buildzoid did a 3700X on a X370 the other night RAM OC video...helpful...I tweaked what Ryzen Mem Calculator was saying at 3200 and it definitely helped...so we'll see.

VPII · Aug 9, 2019

Gill.. said:
New Agesa (1.0.0.3 ABB) was released today for my Killer SLI/AC X370.

Clocks and voltages all seem the same. PBO options changed a tiny bit.

BUT....3600 RAM with FCLK of 1800 is now stable!!!!!! Tonight I'll make sure audio and stability stay the same.

Best part?? Cinebench and Geekbench are better! Geekbench just STOMPED. Check it out below! Both Single and Multi absolutely destroyed all previous attempts.

CPU temps are great, only downside is that I could only get the RAM stable at 1.45V...so much for 1.35...everything I've read assures me that should be fine. 1.43 wouldn't even work..

I'll try to lower sub-timings a bit...but I got a latency of 73-74..which was substantially better than what I was getting before. CL is 18...but the numbers don't lie...Buildzoid did a 3700X on a X370 the other night RAM OC video...helpful...I tweaked what Ryzen Mem Calculator was saying at 3200 and it definitely helped...so we'll see.

Hi Gill, Interesting story you telling. Well it took a while for me being in South Africa to actually get a Ryzen 9 3900X, but I got it. I did use it on my Asus Crosshair VII Hero (X470) board for a while, but it appears I may have killed the board doing a Dry Ice run, not that there was even Dry Ice on it, I think the board got scared. Well right now I'm using a MSI Meg X570 Ace motherboard, just wanted one with the Qpost LCD to identify issues if failed to start. Well at present I'm pretty happy, have not yet reached the CB15 and CB20 scores I got with the C7H mobo, but it was with Per CCX clocking which worked great on the C7H but does not work with this new board as it raises vcore ridiculously. So here was with the C7H X470 board.

Now with the X570 board, the results is a little low for my liking, but partly because they vary from one run to the next. Cannot seem to split the screen grabs but you'll see from the screen grabs which was which, bottom X570. Now I do not believe in PBO or running anything stock which is why my CPU sits at 4.36ghz using 1.33vcore set in bios

Gill.. · Aug 12, 2019

VPII said:
Hi Gill, Interesting story you telling. Well it took a while for me being in South Africa to actually get a Ryzen 9 3900X, but I got it. I did use it on my Asus Crosshair VII Hero (X470) board for a while, but it appears I may have killed the board doing a Dry Ice run, not that there was even Dry Ice on it, I think the board got scared. Well right now I'm using a MSI Meg X570 Ace motherboard, just wanted one with the Qpost LCD to identify issues if failed to start. Well at present I'm pretty happy, have not yet reached the CB15 and CB20 scores I got with the C7H mobo, but it was with Per CCX clocking which worked great on the C7H but does not work with this new board as it raises vcore ridiculously. So here was with the C7H X470 board.

Now with the X570 board, the results is a little low for my liking, but partly because they vary from one run to the next. Cannot seem to split the screen grabs but you'll see from the screen grabs which was which, bottom X570. Now I do not believe in PBO or running anything stock which is why my CPU sits at 4.36ghz using 1.33vcore set in bios

Wow that's pretty insane...so if you weren't using Dry Ice, what was your cooling?? I just got 5911 Single, 46423 Multi....and I gave Linux Version a go...that was insane..I don't have my Snip it tool over there though!

I got my latency down and stable to 72ish..lowered the RAM v to 1.39.

But interesting with the new AGESA.. I haven't cracked 4.6 at all...not once...4.55..but no 4.6.....boo.

Tried manual to 4.3...I gave up at 1.39....seeing what you're getting, maybe I'll try again tomorrow. Thanks for your post!

VPII · Aug 12, 2019

Gill.. said:
Wow that's pretty insane...so if you weren't using Dry Ice, what was your cooling?? I just got 5911 Single, 46423 Multi....and I gave Linux Version a go...that was insane..I don't have my Snip it tool over there though!

I got my latency down and stable to 72ish..lowered the RAM v to 1.39.

But interesting with the new AGESA.. I haven't cracked 4.6 at all...not once...4.55..but no 4.6.....boo.

Tried manual to 4.3...I gave up at 1.39....seeing what you're getting, maybe I'll try again tomorrow. Thanks for your post!

I all honesty when I was about to do the dice run the pot was fitted and all good to go. But the moment I added dice it would struggle to start and I got qcode errors. Ultimately the board my c7h lost a memory channel. Which is why I got the Msi Meg x570 Ace. This board at least has a qcode lcd so fault finding is not an issue.

My new everyday clocks I've toned down to 4.29ghz using only 1.25vcore. Ill try 4.3ghz later but Ill probsbly need a small vcore bump to get it stable. But Ill check later.

Sent from my SM-G960F using Tapatalk

Gill.. · Aug 12, 2019

VPII said:
I all honesty when I was about to do the dice run the pot was fitted and all good to go. But the moment I added dice it would struggle to start and I got qcode errors. Ultimately the board my c7h lost a memory channel. Which is why I got the Msi Meg x570 Ace. This board at least has a qcode lcd so fault finding is not an issue.

My new everyday clocks I've toned down to 4.29ghz using only 1.25vcore. Ill try 4.3ghz later but Ill probsbly need a small vcore bump to get it stable. But Ill check later.

Sent from my SM-G960F using Tapatalk

Do you think I'm trying too much voltage?? Maybe get real and start at 4099 (where it all cores to when I'm mining constantly - above 55Cish)...It just pains me not to go for the gold!

VPII · Aug 13, 2019

Gill.. said:
Do you think I'm trying too much voltage?? Maybe get real and start at 4099 (where it all cores to when I'm mining constantly - above 55Cish)...It just pains me not to go for the gold!

It may be better to start low, like 4.15ghz or there about using only 1.25vcore... see if it is stable, then if stable bump it up to 4.175 and if still stable go for 4.2ghz... it is worth a shot.

3900x First impressions. Precision Boost Overdrive vs. Manual OC on X370

Gill..

Attachments

rdr09

gonX

Gill..

Nighthog

thebufenator

mark_thaddeus

mouacyk

Unoid

roco_smith

AlphaC

Gill..

Attachments

Gill..

Gill..

Attachments

Gill..

Attachments

VPII

Attachments

Gill..

VPII

Gill..

VPII

Top Contributors this Month

Recommended Communities