Just got a new 5800X CPU/MB and I have questions...

[DanielNS84]

So I haven't been as lucky as I wanted getting parts (even watching NowInStock and their Telegram/Discord groups) so I ended up getting a 3080 FTW3 instead of a 3090 FE last year and this month I got a 5800X instead of a 5950X so I ordered a MSI X570 Unify (My Asus WS Z390 Pro ran great and I loved it but Asus not pushing out support for Re-Bar with everyone else left a bad taste in my mouth so I went with MSI this gen), a 980 Pro 2TB, and a Dark Rock Pro 4 and put it all together about a week ago. I've since gone back to my H115i for cooling as the DRP4 didn't work that great and the only reason I didn't before was that I was missing the AMD adapter for my water cooler but I found one on Amazon. Now, here are my questions for you guys...

 

1: My Scores Seem Low on Cinebench Compared to Reviews.

For instance, Cinebench R20 gives me something in the 5900 range but reviews tend to put this chip in the 6200 range...is this to be expected or can I do something about it? I played with PBO but my best results have been from leaving everything at auto except for the voltage curve and putting that at -15 for all cores. Other Cinebench versions show about the same percentage below the review scores as R20 so I'm thinking it could be me doing something wrong.

 

2: My Temps Seem High (Even With the H115i).

My idle with the DRP4 was 45 or so and my max temps under synthetic load was like 89c with AVX so I swapped back to the H115i...now I idle at 35 or so and my max temps are 71-72 but that's still higher than my 9900k with the same cooler which doesn't seem right as those were notoriously space heaters. I've repasted a couple times, tried Kryonaut Extreme, Kryonaut, and even when I used a graphite pad during testing it still showed these temps. Is this normal or are my temps to blame for the low scores?

 

3: My Memory is Potentially at the Wrong Speed/Timings.

I have a kit of 4 x 8GB Corsair Vengeance RGB DDR4 3200 CL14 memory and since it's B-Die I've tried a few speeds but am unsure where to keep it. I used the "Memory Try It!" function in BIOS to set it to a few different speeds/timings and here's what worked/didn't using Memtest86's default 4 hour test run (Never exceeded 1.45v...kept the voltage at the auto/XMP value):

3200MT 14-14-14-34 - Worked

3600MT 14-14-14-34 - Didn't Work

3600MT 14-16-16-36 - Worked

3800MT 18-18-18-38 - Worked

3866+ - Haven't tried it yet...

I even tried the Ryzen Memory Calculator out but the 3600MT CL14 settings it recommended didn't post so I bailed. Also, no WHEA errors at all yet if anyone is curious. I haven't really seen any benchmark differences between these speeds but I haven't spent much time on them besides the 3600MT CL14 one which I'm running now. Should I keep playing with them or forget about it? I'd really like to get some additional performance out of this PC since I didn't get the parts I wanted so any help would be amazing!

 

Thanks for any help you guys can give!

[Skiiwee29]

1. 100% normal score. Youtube benchmarks are done on 100% sterile machines so its basically the very best you can get. 

 

2. Those temps are 100% normal. 

 

3. There is a lot of different timings that need to be adjusted with the Ryzen Memory Calculator. There is a 2nd tab you should look at as well. Ive had great success using it personally. You likely won't get much further than 3800 because of infinity fabric generally being on the verge of unstable on like 99% of all Ryzen 5000 CPUs once its at 1900mhz, or 3800 on your memory unless you uncouple them, which introduces more problem than it solves. 

[jwwagner25]

just adding 2nd opinion that those temps are absolutely normal for 5800x, it is the hottest running of all the zen3 cpu's. Also, my r20 score was the same as you, about 6k at best, and with PBO all set to auto except for -25 all core curve offset. Did you try going lower than -15?

 

I am by no means an expert, but i think i've read that your best chance of hitting high memory speeds are by using two sticks. When there is four, you have worse chance of being stable at higher speeds. Someone with more knowledge can confirm or deny though. 

 

anyway congrats on the new system. 

[DanielNS84]

6 minutes ago, Skiiwee29 said:

1. 100% normal score. Youtube benchmarks are done on 100% sterile machines so its basically the very best you can get. 

 

2. Those temps are 100% normal. 

 

3. There is a lot of different timings that need to be adjusted with the Ryzen Memory Calculator. There is a 2nd tab you should look at as well. Ive had great success using it personally. You likely won't get much further than 3800 because of infinity fabric generally being on the verge of unstable on like 99% of all Ryzen 5000 CPUs once its at 1900mhz, or 3800 on your memory unless you uncouple them, which introduces more problem than it solves. 

Ah okay, that makes me feel a bit better...I am curious about a few things though. 

 

1: My stock voltages looked crazy high which is why I used the curve in the first place and are still around 1.435v Idle in HWInfo...they drop a bit under load but this seems kinda nuts. Is this normal behaviour for a Ryzen 5XXX?

 

2: Ouch, also, I have 3 temps listed in HWInfo and they vary a LOT...is that normal too? Currently my idle while typing this is Tctl/Tdie: 43c, Average: 35c, and Tdie: 38c respectively. I feel like I only had 1 or 2 temps to look at on my 9900k besides the core temps lol. Also, AVX load temps tend to be almost identical for all 3 at Tctl/Tdie: 72.1c, Average: 72.1c, and Tdie: 72.5c respectively which seems more normal but the idle temps can randomly creep up to like 65c on Tctl/Tdie with nothing going on which seems very odd.

 

3: My memory is a relic from before I even had my 9900k, I had an 8086k before that which that memory seemed perfect for at the time but if there's a better option I'm open to replacing it with something else. Would risking a brand name I've never heard of to get PC4000 CL14 (Who's OLOY? lol) be worth it or would going higher on the speed but lower on the CAS and try to downclock it and tighten timings be better? Alternately, should I not even try past 3800 and go for a 3800 CL14 kit? I'm a bit lost and some places say 3800 is the sweet spot and some say 4000 is but I've not seen anywhere that shows conclusively whether timings or speed are most important and/or what speeds/timings the priorities shift or when you get diminishing returns. I can do 3800 CL18 with my current kit of 4 x 8GB but if I can get higher performance with a new kit (and potentially find a good 2 x 16GB kit to let me upgrade to 64GB later) that's not insanely priced (My cap is $400-500 but I'd prefer closer to $200 unless there's a huge performance jump) then I'll go for it for sure. Thanks for your help!

 

[Skiiwee29]

Quote

1: My stock voltages looked crazy high which is why I used the curve in the first place and are still around 1.435v Idle in HWInfo...they drop a bit under load but this seems kinda nuts. Is this normal behaviour for a Ryzen 5XXX?

100% normal. Ryzen 3000 and 5000 series are tuned this way so there is voltage there for the ramp up/burst when a load is put on. Once it sees this and realizes it doesn't need all that voltage, it dials it back like you are seeing. 

 

Quote

2: Ouch, also, I have 3 temps listed in HWInfo and they vary a LOT...is that normal too? Currently my idle while typing this is Tctl/Tdie: 43c, Average: 35c, and Tdie: 38c respectively. I feel like I only had 1 or 2 temps to look at on my 9900k besides the core temps lol. Also, AVX load temps tend to be almost identical for all 3 at Tctl/Tdie: 72.1c, Average: 72.1c, and Tdie: 72.5c respectively which seems more normal but the idle temps can randomly creep up to like 65c on Tctl/Tdie with nothing going on which seems very odd.

Tctl/Tdie is the most accurate in my experience with my 3700x. It matches what shows in Ryzen Master. 

 

Quote

3: My memory is a relic from before I even had my 9900k, I had an 8086k before that which that memory seemed perfect for at the time but if there's a better option I'm open to replacing it with something else. Would risking a brand name I've never heard of to get PC4000 CL14 (Who's OLOY? lol) be worth it or would going higher on the speed but lower on the CAS and try to downclock it and tighten timings be better? Alternately, should I not even try past 3800 and go for a 3800 CL14 kit? I'm a bit lost and some places say 3800 is the sweet spot and some say 4000 is but I've not seen anywhere that shows conclusively whether timings or speed are most important and/or what speeds/timings the priorities shift or when you get diminishing returns. I can do 3800 CL18 with my current kit of 4 x 8GB but if I can get higher performance with a new kit (and potentially find a good 2 x 16GB kit to let me upgrade to 64GB later) that's not insanely priced (My cap is $400-500 but I'd prefer closer to $200 unless there's a huge performance jump) then I'll go for it for sure. Thanks for your help!

Stick with the B-Dies you have now. They're great and the best at overclocking. Since you were able to get 3600mhz at 16-18-18-38, I would use that and call it good. You will gain nothing more really from trying to tune further. They say 3800 because of the Fclock I mentioned previously. 

[DanielNS84]

18 minutes ago, jwwagner25 said:

just adding 2nd opinion that those temps are absolutely normal for 5800x, it is the hottest running of all the zen3 cpu's. Also, my r20 score was the same as you, about 6k at best, and with PBO all set to auto except for -25 all core curve offset. Did you try going lower than -15?

 

I am by no means an expert, but i think i've read that your best chance of hitting high memory speeds are by using two sticks. When there is four, you have worse chance of being stable at higher speeds. Someone with more knowledge can confirm or deny though. 

 

anyway congrats on the new system. 

Thanks! Also, that's good to hear...also, both of you should know I've been running this same install of Windows 10 since it was released...I installed it onto my i5-2500 system with 2 x 4GB DDR3 1066 and a 250GB HDD using an iGPU for work. I repaired PC's for a living at the time so reinstalling Windows was essentially "giving up" so transitioning this install to new hardware configurations was something I felt I had to at least try lol. Since then I've swapped every component several times and cloned this installation from the 250GB HDD to a 1TB HDD, a 120GB SSD, 250GB SSD, 512GB SSD, 1TB SSD, 1TB NVMe SSD, and finally to the 2TB 980 Pro it's on now so it's seen a lot since I installed it. In any case...I'm attached now so I can't bail on it lol!

[jwwagner25]

1 minute ago, DanielNS84 said:

Thanks! Also, that's good to hear...also, both of you should know I've been running this same install of Windows 10 since it was released...I installed it onto my i5-2500 system with 2 x 4GB DDR3 1066 and a 250GB HDD using an iGPU for work. I repaired PC's for a living at the time so reinstalling Windows was essentially "giving up" so transitioning this install to new hardware configurations was something I felt I had to at least try lol. Since then I've swapped every component several times and cloned this installation from the 250GB HDD to a 1TB HDD, a 120GB SSD, 250GB SSD, 512GB SSD, 1TB SSD, 1TB NVMe SSD, and finally to the 2TB 980 Pro it's on now so it's seen a lot since I installed it. In any case...I'm attached now so I can't bail on it lol!

that is some hardcore dedication, quite possibly world record for longest-running windows installation?!

[DanielNS84]

4 minutes ago, Skiiwee29 said:

100% normal. Ryzen 3000 and 5000 series are tuned this way so there is voltage there for the ramp up/burst when a load is put on. Once it sees this and realizes it doesn't need all that voltage, it dials it back like you are seeing. 

 

Tctl/Tdie is the most accurate in my experience with my 3700x. It matches what shows in Ryzen Master. 

 

Stick with the B-Dies you have now. They're great and the best at overclocking. Since you were able to get 3600mhz at 16-18-18-38, I would use that and call it good. You will gain nothing more really from trying to tune further. They say 3800 because of the Fclock I mentioned previously. 

Ah okay, thanks! So I shouldn't stick with 3600 14-16-16-38 or 3800 18-18-18-38? Both of those were stable and I haven't tried 3600 16-18-18-38 yet so I'm not sure but I'd assume so.

[Skiiwee29]

1 minute ago, DanielNS84 said:

Ah okay, thanks! So I shouldn't stick with 3600 14-16-16-38 or 3800 18-18-18-38? Both of those were stable and I haven't tried 3600 16-18-18-38 yet so I'm not sure but I'd assume so.

Go with what you know is stable and then work from there and see what you can come up with. Thats the beauty of overclocking! Never know where you will end up at. Just make sure to keep your Fclock 1 to 1 with your RAM speed. IE: 3600mhz on RAM is 1800 Fclock, 3800/1900 etc.. 

[DanielNS84]

1 minute ago, jwwagner25 said:

that is some hardcore dedication, quite possibly world record for longest-running windows installation?!

Technically it's made it longer if you count the Windows 7 install I upgraded from since I didn't do a clean install and kept all my programs. Also, I have another PC that's made it longer but it's got ugly hardware in it lol...it's made it from a Pentium 3 until now but with less frequent hardware changes overall. It had (I think) the first version of Windows 10 when it came out of beta and it's made it until now and it has a 8th gen CPU in it...I believe it started it's life as a Windows 2000 system before I upgraded to the Core 2 Duo (I Skipped the pentium 4). I still remember upgrading it to have the "brand new" USB 3 ports in it lol. It may have even had an AGP port in it before I swapped it to the LGA 775 board with PCIE (I think it was a Asus P5Q Deluxe?) so it was running for a loooooong time before it even got Windows 10. I stuck with Windows 2000 for a long time and hated Vista enough to skip it and hop onto Windows 7 as an upgrade when it came out. Also...now that I think about it...I'm old lol.

[DanielNS84]

15 minutes ago, Skiiwee29 said:

Go with what you know is stable and then work from there and see what you can come up with. Thats the beauty of overclocking! Never know where you will end up at. Just make sure to keep your Fclock 1 to 1 with your RAM speed. IE: 3600mhz on RAM is 1800 Fclock, 3800/1900 etc.. 

Got it, thanks! I'll play with different clocks and timings a bit, do some gaming and stress tests and keep an eye open for WHEA issues and I'll post here again with any updates. Thanks again!

[DanielNS84]

I gave up and drove to micro center and bought a 5950x and PC 4000 ram (2 x 16GB). Will update my info and get everything put in shortly!

[DanielNS84]

Okay, all good news I guess...

 

1: My 5800x is in a box on the road to Best Buy...good riddance! Virtually all the little issues I was seeing that I thought were unrelated have gone away...maybe I should have tried to hang onto my old memory and OC it with the new CPU to see how it did...now I already put it in my daughter's 9900k machine.

 

2: The scores for this 5950x without any tweaking or messing around are 10,924 in R20, 656/12524 in CPU-Z's bench, and 4774 in R15. These scores are around where I'd hope they would be. The temps, oddly enough, seem to be about the same as with the 5800x...granted I did swap from Kryonaut Extreme to ProlimaTech PK-3 Nano because Kryonaut Extreme was way to sticky for my liking but otherwise it's identical cooling. The system hits a max of 72c unless I'm doing AVX loads in which case it hits close to 80c (78c last runs) and stays there and the idle temps are just a couple degrees higher (although there's a bigger idle difference between BIOS and Windows which is odd) at 30-32c depending on the time of day.

 

3: The new memory (Just some G.Skill 4000 CL18 16GB sticks x 2 to test the max RAM speed on this thing) boots at 4000MT/s using XMP with the IF manually at 2000Mhz and all seemed well until the first cinebench when my WHEA alarm went off and listed several of them. The system seems completely fine otherwise but just in case I dropped it to 3800MT/s and 1900Mhz for now until I can test it further. Maybe I can find some tips to fix those and run it at that frequency. The benchmarks I ran were even higher at 2000 IF so I'd really like to figure that out. I could also potentially sneak my daughter's 3200 CL16 4 x 8GB LPX kit back in and grab my old memory again for testing if needed lol.

 

Overall I'm MUCH happier with this CPU and it was worth the $350 upgrade for sure...the single thread performance is even better which I wasn't expecting at all. Maybe I'll OC this CPU at some point but maybe not (which is the first time I've been happy enough with speed initially since I got my 8086K (which I ended up OC'ing like 6 months later lol)). In any case, very open to any help or tips if you guys have them. What else should I do (besides go to a 3090 which I may do since Micro Center had plenty of those too...or maybe a 3080 ti when those hit) to keep this party going? Thanks guys!

[freeagent]

Congrats!

 

I don't have your CPU but I needed to raise VSOC voltage to ~1.15v to nail down 2k and keep the WHEA's at bay. You may also have to play with CLDO VDDP, VDDG CCD, and VDDG IOD a little.

 

Outside of that I am out of my league as I have not touched one of those..

[DanielNS84]

1 hour ago, freeagent said:

Congrats!

 

I don't have your CPU but I needed to raise VSOC voltage to ~1.15v to nail down 2k and keep the WHEA's at bay. You may also have to play with CLDO VDDP, VDDG CCD, and VDDG IOD a little.

 

Outside of that I am out of my league as I have not touched one of those..

Gotcha, I'll give it a shot...thanks!!!

[DanielNS84]

Also here's my AIDA results for the current config...look normal?

 

 

Edit: Added My Cinebench R23 as Well (28117)...

[DanielNS84]

2nd Run (28905)...creeping up on that 2990WX!

 

Edit: 29052!

 

Edit 2: 29123! May just let it sit here lol...only done -15 Curve optimizer and PBO on at defaults so this is pretty good to me I think.

 

 

[DanielNS84]

So this is intriguing...I found a guide and was able to get about 28k R23 without going over 70c under load...kind of worth the 1k points to run that cool I think. Gonna try it for a few days to see how it feels. I get 670/12756 in CPU-z and my single core seems the same so I may not be missing much...single core speeds (on 2 cores) hit 5.05Ghz and 2 more hit 5.025Ghz so it seems pretty cool. I hit 26.5K with temps hitting 65c max but that felt like too much of a loss so I boosted it a bit. Thoughts?

[DanielNS84]

Yo WTF! On a whim I put my 3200 Vengeance RGB kit back in (4 x 8GB) to see how it does on the 5950x and it’s running faster and with tighter timings than my new 4000 kit (just finished my first memtest cycle)...is B-Die that good? To OC 20%+ despite the 2 vs. 4 sticks? That’s strange...guess I’m returning the new memory lol.

[Guest]

1 minute ago, DanielNS84 said:

Yo WTF! On a whim I put my 3200 Vengeance RGB kit back in (4 x 8GB) to see how it does on the 5950x and it’s running faster and with tighter timings than my new 4000 kit (just finished my first memtest cycle)...is B-Die that good? To OC 20%+ despite the 2 vs. 4 sticks? That’s strange...guess I’m returning the new memory lol.

Yes, yes B-Die is that good. Scales well with voltage too. 

 

Most of the better end kits will OC 16-16-16-36 timing sets very well up to and past 4000mt/s. It just requires a little extra Dram voltage.

1.45v on stock kits, so OC 1.55v is no issue. Best to keep them cool at load. I recommend 40-45c but no more or you sacrifice possible stability. 

 

2x8gb will OC the best however. So get what you can from the IMC by giving that additional voltage as well.

XMP enabled, the SOC should auto over-volt to about 1.08v, you can safely run 1.15 - 1.20v on ambient if needed for 3800-4000mt/s.

 

If you wanted to go past 4000mt/s, unlink that IF 2:1 and go for it. 17-19-19-39 or 19-19-19-39 should net you 4200+ without going crazy on the Dram voltage.

 

Didn't want to say anything about your new kit until I saw conclusions. Yep, I'd return it too. 

[DanielNS84]

8 minutes ago, ShrimpBrime said:

Yes, yes B-Die is that good. Scales well with voltage too. 

 

Most of the better end kits will OC 16-16-16-36 timing sets very well up to and past 4000mt/s. It just requires a little extra Dram voltage.

1.45v on stock kits, so OC 1.55v is no issue. Best to keep them cool at load. I recommend 40-45c but no more or you sacrifice possible stability. 

 

2x8gb will OC the best however. So get what you can from the IMC by giving that additional voltage as well.

XMP enabled, the SOC should auto over-volt to about 1.08v, you can safely run 1.15 - 1.20v on ambient if needed for 3800-4000mt/s.

 

If you wanted to go past 4000mt/s, unlink that IF 2:1 and go for it. 17-19-19-39 or 19-19-19-39 should net you 4200+ without going crazy on the Dram voltage.

 

Didn't want to say anything about your new kit until I saw conclusions. Yep, I'd return it too. 

Glad I checked it within the return period...I literally just hit “memory try it 3800 18-18-18-38” in the BIOS thinking it would kick it back...I haven’t even bumped voltage yet. It’s my old 3200 CL14 kit from my 8086k system that I reused for my 9900k system...that’s nuts that I’ll be using it again for a 3rd time...this may be a stupid question but why don’t they sell this memory at a higher speed (and maybe add higher XMP profiles)? It seems like they could make more money...is it because they only test it at low voltages?

 

Edit: Also, since I’m going to Micro Center anyway I guess I could get something better...I already checked and nothing over 3800 is in stock that’s on my QVL there BUT there was some memory that did score higher than 3800 memory even when decoupled and that’s the 5100MT memory which I think is in stock (but very expensive)...is that throwing money away? I know it does great on memory benchmarks but I haven’t seen any real world performance numbers in games and such.

[Guest]

30 minutes ago, DanielNS84 said:

Glad I checked it within the return period...I literally just hit “memory try it 3800 18-18-18-38” in the BIOS thinking it would kick it back...I haven’t even bumped voltage yet. It’s my old 3200 CL14 kit from my 8086k system that I reused for my 9900k system...that’s nuts that I’ll be using it again for a 3rd time...this may be a stupid question but why don’t they sell this memory at a higher speed (and maybe add higher XMP profiles)? It seems like they could make more money...is it because they only test it at low voltages?

 

Edit: Also, since I’m going to Micro Center anyway I guess I could get something better...I already checked and nothing over 3800 is in stock that’s on my QVL there BUT there was some memory that did score higher than 3800 memory even when decoupled and that’s the 5100MT memory which I think is in stock (but very expensive)...is that throwing money away? I know it does great on memory benchmarks but I haven’t seen any real world performance numbers in games and such.

They do make high speed kits.

Mine are 4266mt/s 1.45v CL19 2x8gb. 

Running with 8700K actually.

 

5000mt/s kits have crazy loose timings. 

But you're banking on the memory controller though. 

Don't think it's worth it. But I have seen 4600mt/s on Ryzen chips.

Just depends on how much effort you're willing to put in past XMP speeds.

 

For to just set it up with good performance CL14-14-14-34 @3600mt/s.

This I think is very doable. Maybe 1.45v is plenty actually. Touch more.

Disable power down mode.

Disable Gear down mode for odd CL timings.

[DanielNS84]

22 minutes ago, ShrimpBrime said:

They do make high speed kits.

Mine are 4266mt/s 1.45v CL19 2x8gb. 

Running with 8700K actually.

 

5000mt/s kits have crazy loose timings. 

But you're banking on the memory controller though. 

Don't think it's worth it. But I have seen 4600mt/s on Ryzen chips.

Just depends on how much effort you're willing to put in past XMP speeds.

 

For to just set it up with good performance CL14-14-14-34 @3600mt/s.

This I think is very doable. Maybe 1.45v is plenty actually. Touch more.

Disable power down mode.

Disable Gear down mode for odd CL timings.

I’ll give it a shot, it’s on pass 2 @ 3866 CL 18 which is further than my 4000 kit got even though it’s rated for that cas latency at a higher frequency. I’d like to get 3800 at a low cas but I know that’s asking a lot. I’m getting my frequency cap at 1:1 memory:IF then playing with timings. I’ve seen a lot of benchmarks where performance nosedives after 3800 but is that just decoupling causing that? Asking because if I can maintain 1:1 I’d like to know if there’s a limit on how far I should go even if the memory can take it (or should I just target the lowest AIDA latency with a speed/cas balance)?

[DanielNS84]

Okay, so I took a break from bumping up and memtesting to try 3600 14-14-14-34 and it booted fine at the default voltage and is testing now. If this runs fine then it will be better performance than 3800 CL-18 or 16?

[Guest]

1 minute ago, DanielNS84 said:

Okay, so I took a break from bumping up and memtesting to try 3600 14-14-14-34 and it booted fine at the default voltage and is testing now. If this runs fine then it will be better performance than 3800 CL-18 or 16?

They'll do more than 3600mt/s. You start here. Increase frequency incrementally until it's not stable, add some Dram voltage. 1.5 will be fine even for daily.