AMD Ryzen 5000 Builders Thread

[Keysplayr]

Ugh. Not in stock anywhere...... I wanted to go for the 5900X. Most impressed by AMD. Well done.

 

[amrnuke]

Ugh. Not in stock anywhere...... I wanted to go for the 5900X. Most impressed by AMD. Well done.

You could always go buy a 10900K for 10% less.

 

[UsandThem]

You could always go buy a 10900K for 10% less.

10% less in price or performance? Because if it's performance, the difference is more than 10%.

 

[Det0x]

After wasting a few days on bios 3103 for the crosshair viii hero motherboard i'm back to the 3003 bios and have finalized my everyday 24/7 settings.

32gigabyte @ 1900MT/S 4x8gigabyte b-die gskill memory
Prettymuch all timings and voltages are handtuned/maximized/minimized.

Zero errors/WHEA after 1 hour in TestMeM5 with the 1usmus preset. (9 cycles for 32gig)


Screenshot show all settings and voltages, but i can write them also:
dram = 1.56 volt (warmest memorystick reached 44 degrees in TestMeM, normal temp is 32 degrees)
vsoc = 1.1188 volt
cldo vddp = 0.8796 volt
vddg iod = 1.0477 volt
vddg ccd = 0.8796 volt
ProcODT = 43.6

If anyone see something that can be improved/mistakes in the timings, please let me know

 

[Keysplayr]

You could always go buy a 10900K for 10% less.

Something about the "10" series doesn't sit right with me. It's no major improvement over 9th gen. I guess I could wait for Rocket Lake or whatever it is that is next. But Intel usually rests on it's laurels every few gens and lets AMD take the lead for a gen or two. Got to keep your monopoly "proofness" alive and kicking. AMD has a great CEO now. Doing great things. I don't think we'll ever see another "Conroe" from Intel. They're really milking things.

 

[amrnuke]

10% less in price or performance? Because if it's performance, the difference is more than 10%.

Something about the "10" series doesn't sit right with me. It's no major improvement over 9th gen. I guess I could wait for Rocket Lake or whatever it is that is next. But Intel usually rests on it's laurels every few gens and lets AMD take the lead for a gen or two. Got to keep your monopoly "proofness" alive and kicking. AMD has a great CEO now. Doing great things. I don't think we'll ever see another "Conroe" from Intel. They're really milking things.

The 10900K is a poorer value than the 5900X and I would not actually recommend it blindly without asking about use case, unless I was recommending it for another reason, like... sarcasm.

 

[Noid]

sarcasm

You need to declare such usage here.
( or you get misunderstood and flamed )

/sarcasm

 

[MalVeauX]

I would like to do a new build, mostly to satisfy the geek new shiny side of me. I really would. But this is the last year of AM4, and the prices are crazy especially for graphics cards. So I am torn. What I have now works just fine, though a better graphics card would be nice. Or do I hold out for AM5 goodness?

AM5 is over a year away. This COVID shipping stuff isn't going to magically just stop, so I wouldn't expect an easy roll out of anything at this point, and lots of delays. If your system is doing well now, stick with it. A GPU can migrate in a year, a CPU changing sockets cannot migrate. So if you have the itch, scratch it with a GPU or new display or something that migrates platforms.

Very best,

 

[KentState]

I think I figured out why my temps are so high attempting static overclocks on my 5950x - turns out optimus forgot to install the internal jet plate oring, so presumably the water is escaping over the top of the fin channels instead of being forced through them.

They're sending me some orings so I can fix it. No idea how much temps are affected but my intuition is maybe 5-10C.

That may have made a huge difference. My 5950X is one of the coolest running CPU's I've had under water in recent years. It takes a heavy single core load to get it into the 60s and runs in the low 50s in multi core situations. At idle it's near ambient which is around 21-22C.

 

[Midwayman]

That may have made a huge difference. My 5950X is one of the coolest running CPU's I've had under water in recent years. It takes a heavy single core load to get it into the 60s and runs in the low 50s in multi core situations. At idle it's near ambient which is around 21-22C.

Wow you either have amazing cooling or got a real golden sample. Most people seem to be in the mid to low 30s at idle.

 

[XabanakFanatik]

That may have made a huge difference. My 5950X is one of the coolest running CPU's I've had under water in recent years. It takes a heavy single core load to get it into the 60s and runs in the low 50s in multi core situations. At idle it's near ambient which is around 21-22C.

I hope it makes a huge difference. After doing some looking around it seems like my load temps are in line with what's expected of AIO's, not a high end waterblock with microfins and a high pressure dual pump.

Wow you either have amazing cooling or got a real golden sample. Most people seem to be in the mid to low 30s at idle.

Even without the jet plate o-ring mine idles in the low 20's.

 

[KentState]

Wow you either have amazing cooling or got a real golden sample. Most people seem to be in the mid to low 30s at idle.

I missed typed that part. Was answering work emails and going back and forth. Meant that it's ambient is near 21-22C. It idles in the higher 20's.

 

[Elfear]

After wasting a few days on bios 3103 for the crosshair viii hero motherboard i'm back to the 3003 bios and have finalized my everyday 24/7 settings.

32gigabyte @ 1900MT/S 4x8gigabyte b-die gskill memory
Prettymuch all timings and voltages are handtuned/maximized/minimized.

Zero errors/WHEA after 1 hour in TestMeM5 with the 1usmus preset. (9 cycles for 32gig)


Screenshot show all settings and voltages, but i can write them also:
dram = 1.56 volt (warmest memorystick reached 44 degrees in TestMeM, normal temp is 32 degrees)
vsoc = 1.1188 volt
cldo vddp = 0.8796 volt
vddg iod = 1.0477 volt
vddg ccd = 0.8796 volt
ProcODT = 43.6

If anyone see something that can be improved/mistakes in the timings, please let me know

Looks great for 24/7 use! But that VDIMM! I thought I was pumping some voltage through my B-Die at 1.45V.

Did you mess around with your Curve Optimizer since the last update you posted?

 

[Kenmitch]

Looks great for 24/7 use! But that VDIMM! I thought I was pumping some voltage through my B-Die at 1.45V.

Wimp! /s

The G.Skill 3800 CL14 b-die kit's stock voltage is 1.5v's. If it wasn't safe I'd imagine they'd be smart enough to not even offer it.

G.SKILL Releases Optimized DDR4-3800 CL14 Memory Kit for AMD Ryzen 3000 & X570 Platform-G.SKILL International Enterprise Co., Ltd.

Not that I'd go in the crazy voltage mode, but them crazy benchmark kings pump up to 2v's into b-die.

 

[Det0x]

Looks great for 24/7 use! But that VDIMM! I thought I was pumping some voltage through my B-Die at 1.45V.

Did you mess around with your Curve Optimizer since the last update you posted?

I have been running my memory at or above 1.5volts for over 1 year now, no problems yet, just remember to keep them cooled.. I use a fan for that.

Have stabilized my curve optimizer settings at -30 on all cores with 100% stability, the trick for me was to turn off global c-stats (lose ~50-100mhz singlethread) and use the minimum required voltage to keep them from crashing the system into a reboot at low load/idle.
I use LLC2 (1=weakest on asus bios) and a voltage offset of +0.0125volt to do that, but i also think i have a better then average cpu, so would guesstimate LLC3 and + 0.025 volt to be more common settings.
I found Aida memory copy benchmark to be a good quick test to check CO stability as it pretty much always crashed my system when the cores didn't get enough volt.

In regards to the discussion about static OC vs curve optimizer PBO, i found the best settings for me to be CO PBO for 90% of the applications i use..
This is some data i have from my CO setup:



"Effective clocks" from hwinfo with cinebench r23 running.
(one thread per core forced with taskmanager affinity)

1 thread/core load = 5025mhz
2 thread/core load = 4975+4975
3 thread/core load = 4855+4855+4855
4 thread/core load = 4829+4829+4829+4829
5 thread/core load = 4815+4815+4815+4815+4815
6 thread/core load = 4790 on all cores
7 thread/core load = 4755 on all cores
8 thread/core load = 4760 on all cores
9 thread/core load = 4714 on all cores
10 thread/core load = 4708 on all cores
11 thread/core load = 4692 on all cores
12 thread/core load = 4686 on all cores
13 thread/core load = 4678 on all cores
14 thread/core load = 4656 on all cores
15 thread/core load = 4652 on all cores
16 thread/core load = 4635 on all cores

16cores/32threads = 4600 on all cores -> 30655 points in r32 (warm waterloop after running cinebench for a long time)
(my highscore in r23 with CO netted me 31900 points)

My max static OC so far is 4700/4600 with 100% stability even in Prime95 and IBT.
Ofcorse PBO will clock lower in prime95 or IBT then static OC, but PBO will clock higher then Cinebench in games, and that is something which i actually use this computer for..
So for my cpu in my setup, curve optimizer PBO is the logical choice

 

[Elfear]

Wimp! /s

The G.Skill 3800 CL14 b-die kit's stock voltage is 1.5v's. If it wasn't safe I'd imagine they'd be smart enough to not even offer it.

G.SKILL Releases Optimized DDR4-3800 CL14 Memory Kit for AMD Ryzen 3000 & X570 Platform-G.SKILL International Enterprise Co., Ltd.

Not that I'd go in the crazy voltage mode, but them crazy benchmark kings pump up to 2v's into b-die.

Good info. I have pretty good cooling on my ram so I'll have to experiment around a bit.

I have been running my memory at or above 1.5volts for over 1 year now, no problems yet, just remember to keep them cooled.. I use a fan for that.

Have stabilized my curve optimizer settings at -30 on all cores with 100% stability, the trick for me was to turn off global c-stats (lose ~50-100mhz singlethread) and use the minimum required voltage to keep them from crashing the system into a reboot at low load/idle.
I use LLC2 (1=weakest on asus bios) and a voltage offset of +0.0125volt to do that, but i also think i have a better then average cpu, so would guesstimate LLC3 and + 0.025 volt to be more common settings.
I found Aida memory copy benchmark to be a good quick test to check CO stability as it pretty much always crashed my system when the cores didn't get enough volt.

In regards to the discussion about static OC vs curve optimizer PBO, i found the best settings for me to be CO PBO for 90% of the applications i use..
This is some data i have from my CO setup:



"Effective clocks" from hwinfo with cinebench r23 running.
(one thread per core forced with taskmanager affinity)

1 thread/core load = 5025mhz
2 thread/core load = 4975+4975
3 thread/core load = 4855+4855+4855
4 thread/core load = 4829+4829+4829+4829
5 thread/core load = 4815+4815+4815+4815+4815
6 thread/core load = 4790 on all cores
7 thread/core load = 4755 on all cores
8 thread/core load = 4760 on all cores
9 thread/core load = 4714 on all cores
10 thread/core load = 4708 on all cores
11 thread/core load = 4692 on all cores
12 thread/core load = 4686 on all cores
13 thread/core load = 4678 on all cores
14 thread/core load = 4656 on all cores
15 thread/core load = 4652 on all cores
16 thread/core load = 4635 on all cores

16cores/32threads = 4600 on all cores -> 30655 points in r32 (warm waterloop after running cinebench for a long time)
(my highscore in r23 with CO netted me 31900 points)

My max static OC so far is 4700/4600 with 100% stability even in Prime95 and IBT.
Ofcorse PBO will clock lower in prime95 or IBT then static OC, but PBO will clock higher then Cinebench in games, and that is something which i actually use this computer for..
So for my cpu in my setup, curve optimizer PBO is the logical choice

I tried the same process from your last post and wasn't having much luck until I updated to the latest Asus bios (3201) and then cleared the CMOS by pulling the battery. After that I had a lot more luck with PBO.

What's your R23 ST score?

Seeing your results gives me new incentive to keep tweaking.

 

[Det0x]

This is my 24/7 scores:


And these are my alltime high Cinebench scores:



Good luck tweaking and happy hunting for those new highscores

 

[IEC]

I just started dabbling with PBO+CO on my 5800X. PPT limit 95W (vs stock 142W) since I like efficiency

CO settings -10 best cores -20 everything else. No core voltage offset yet, haven't played with that. In lightly threaded workloads like below it'll briefly shoot cores above 5GHz:


Sustained clocks aren't much different than stock but temperature is very nice and power usage is roughly 2/3 of stock.


Being able to adjust the VF curve with CO is a step in the right direction for those of us who like to OC.

I definitely have more headroom to go up on PPT as temps are mid-60s C on full AVX load with high ambient temp.

 

[Elfear]

This is my 24/7 scores:


And these are my alltime high Cinebench scores:



Good luck tweaking and happy hunting for those new highscores

That is impressive. My highest so far is 1,650/29,858.

Thanks again for posting up your experience.

 

[Kenmitch]

Happy New Year!

New Year
New uEFI version
New adventure!

MSI dropped the latest, but beta uEFI for my MB so I decided to give it a whirl and see how the FCLK responds.

I pulled 2 memory sticks out of the rig for some testing. Still using 1 of the G.Skill 3600_CL16 16GB kit.

I've been toying around with it for a few hours so far. I've booted beyond the 2000 FCLK to see if it was possible, but stopped at 4166MHz just to see if it was doable.

So far I've made it to the following.



I've only done initial stability testing as the 1st go around usually spits out errors if timings aren't up to snuff. Of course this is after you successfully boot, load windows, and eliminate the bsods. In the past I usually only have to up the dram voltage to get stability during more extensive memory testing.

EDIT: I could test the above and it wouldn't spit out any memory errors during extensive testing the WHEA error 19 struck again. I could find no way to eliminate them. 3800/1900 there are no WHEA errors at all. Maybe next non-beta will keep the 1900 FCLK and eliminate the WHEA error 19?

I might try the other 2 sets of b-die I have to see if the WHEA errors go away, but I'm leaning towards nothing will change in the end.

 

[Det0x]

Tighter memory settings and TestMem5 1usmus 1 hour stabile

 

[amrnuke]

Was on CoolerMaster Hyper 212 RGB, upgraded to Arctic Liquid Freezer II 240mm
Was on 3200CL16 RAM, upgraded to 3600CL14 (14-14-14-30)
Changed PBO to +100 MHz (my mobo doesn't have the new AGESA yet) - took boost from 4.65 GHz to 4.75 GHz (2.15%)

GB5 ST: 1598 -> 1643 (2.8%)
GB5 MT: 8572 -> 9200 (7.3%)
Speedometer 2.0: 174.4 -> 178.2 (2.1%)
CB20 ST -- 600 -> 611 (1.8%)
CB20 MT -- 4609 -> 4660 (1.1%)
CB23 ST -- 1538 -> 1575 (2.4%)
CB23 MT -- 11929 -> 12011 (0.7%)

The ST (and Speedometer) increases fall within a range consistent with the increased single-core boost (which it hits in all those workloads).
Cinebench doesn't hit memory much, and the added cooling really didn't help much, perhaps suggesting that this is possibly a thermal migration issue? But cracking 12000 in CB23 MT was nice!
Geekbench 5 MT really loved the faster RAM.

CPU-Z ST 653.1
CPU-Z MT 5151.4

I'm going to hold off on further PBO increases and instead optimize memory timings, probably running 3800 CL 16 first to see if 1900 FCLK is possible, then optimizing timings from there. The kit is a B-die 4000 CL 18 set, so I'm hoping for some luck!

Overall happy with how easy it has been to "tune" this with my entry-level B450M BIOS.

 

[repoman0]

So I got my fancy new 5900x today at MC and have gotten it to boot exactly once. That was the first boot, after which it promptly blue screened five minutes later, which at first I just figured was a Windows problem since I changed motherboard at the same time (ASUS X570 -> another ASUS X570). I have been unable to get it to boot again on either board .. just stuck at the DRAM diag LED. Tried one stick in multiple slots with no dice. My existing board that I've been using with my 3700x is a Prime X570 Pro updated to the latest non-beta 3001 BIOS, so it should be ready. RAM is Crucial Ballistix 2x16GB DDR4-3200, rock stable on my 3700x.

I'm currently in Windows typing this on my 3700x .. no apparent issues, booted right up immediately once I swapped back. Is my new 5900x a paperweight?

 

[Dave3000]

Have you tried a different brand of DDR4-3200 memory? What about if you use the default (non-XMP) setting for the memory in the BIOS, is it stable then on the 5900x?

 

[amrnuke]

So I got my fancy new 5900x today at MC and have gotten it to boot exactly once. That was the first boot, after which it promptly blue screened five minutes later, which at first I just figured was a Windows problem since I changed motherboard at the same time (ASUS X570 -> another ASUS X570). I have been unable to get it to boot again on either board .. just stuck at the DRAM diag LED. Tried one stick in multiple slots with no dice. My existing board that I've been using with my 3700x is a Prime X570 Pro updated to the latest non-beta 3001 BIOS, so it should be ready. RAM is Crucial Ballistix 2x16GB DDR4-3200, rock stable on my 3700x.

I'm currently in Windows typing this on my 3700x .. no apparent issues, booted right up immediately once I swapped back. Is my new 5900x a paperweight?

Have you tried to reset CMOS?
Is the RAM on your mobo compatibility list (even if it worked before, if it isn't on the QVL list, then it's possible -- though admittedly not probable -- that it will work with some hardware configurations and not others)?