Crosshair VII Hero Essential Info Thread

[gupsterg]

So I now have a R9 3900X as well and this is a FCLK Clock Edition .

Initially I set 3800MHz 1:1:1 (FCLK:UCLK:MEMCLK) on the R9 3900X as I would on the R5 3600.

ProcODT 43.6,
VDIMM 1.35V
VTTDDR 0.675V
SOC 1.062V
CLDO_VDDP 0.927V
CLDO_VDDG 0.980V

This failed Kahru RAM Test 8% . I bumped ProcODT to 48 and failed Kahru RAM Test 80% . Again I bumped ProcODT, went 53 and Kahru RAM Test upto 460% and failed at 607% .

As the R9 3900X was having a greater load line effect than than the single die R5 3600 & R7 3700X I decided to bump SOC/CLDO_VDDP/CLDO_VDDG a step each and now at ~1313% and counting .

When I targeted 3800MHz on my R5 3600 I found out which slots on my mobo favored which dimms. This held true also for the R7 3700X, ie same dimms favored same slots, it also seems the R9 3900X is the same also.

R9 3900X Kahru RAM Test screenie album, link. I'm using the beta 0002 UEFI with Combo-AM4 1.0.04B.

[bluestang]

Great info...thanks!

What LLC level are you using? What ProcODT value did you end up using for the 1313% run?

Still haven't ot my 3900X setup yet...life keeps happening

[gupsterg]

Great info...thanks!

What LLC level are you using? What ProcODT value did you end up using for the 1313% run?

Still haven't ot my 3900X setup yet...life keeps happening

NP, glad you found it useful .

R9 3900X (Batch code: BF 1944SUT)
ASUS ROG Crosshair VII Hero Wifi using UEFI 0002 (beta) AGESA 1.0.0.4B
F4-3200C14Q-32GVK

UEFI is all defaults except :-

ProcODT 53.3
VDIMM 1.35V
VTTDDR 0.675V
SOC 1.068V
CLDO_VDDP 0.930V
CLDO_VDDG 0.983V

I always use LLC [Auto] and prefer to bump voltage as needed.

That ~1313% run I stopped at 4075%, link to youtube screen capture video.

Now "repeat'n'rinse" testing under way, once I feel the profile has been tested more with HW will post a UEFI settings txt.

[gupsterg]

Gotta find the time to update OP, just been so snowed.

On another forum some one had wondered about the new effective clocks in HWINFO, see this post by author.

Also there seems to be a new block on the market which may interest Matisse owners, link.

*** edit ***

Really am liking this CPU sample. Some nice boost clocks at stock, really just thinking of testing the FCLK/MEMCLK OC and calling it a day.

This album has:-

i) HWINFO/Ryzen Master screenie to see which cores are being highlighted as best.

ii) 3x AIDA64 on 3800MHz C16.

iii) temps/boost of CPU at stock when CB20 multi run, room ambient 18C at the time.

iv) CB20 single/multi, first of the day run, room ambient 16C, CPU stock RAM/FCLK 3800MHz.

v) Another CB20 run later in the day after having done some stability testing, so some what warm rig, room ambient 18C.

vi) Running a SFC scan, not that having an issue, just wanted to see boosting and it's nice .

Still doing more testing, profile has shaped up nicely IMO and will soon share settings txt .

[gupsterg]

The triple showdown has begun ...

All CPUs are batch BF 1944SUT, all have UG stamped on IHS ...

As I felt the first had been tested enough for 3800MHz 1:1:1, I moved on to the second, at stock the 2nd is ~100 points higher in CB20 (~7323 points multi), let's see if it's also a FCLK edition ...

[gupsterg]

My 2nd/3rd samples of R9 3900X I have dubbed MHz Editions .

These responds to PBO [Enabled] better than FCLK Edition not yet tried +xxxMHz . It also seems the cores are more similar so better for all cores OC .

This ZIP contains stock CPU Statuscore loading each thread with Ryzen Master compare.

2nd/3rd R9 3900X seems to have better multicore boost, not only when checked for clocks but benches. For example for multicore in Realbench stress mode 32GB 30min run 1st is ~4.0GHz all cores average effective MHz, 2nd is ~4.1GHz. When loading with P95 8K 4096K 27GB, 1st is ~4.025GHz all cores average effective MHz, 2nd is ~4.075GHz.

They all seem a little differing on ProcODT for RAM MHz.

        R9_1    R9_2    R9_3

CCD0

C01    4551    4546    4545
C02    4515    4517    4524
C03    4555    4484    4452
C04    4543    4542    4547
C05    4515    4479    4517
C06    4549    4463    4470

CCD1

C07    4369    4421    4407
C08    4313    4373    4406
C09    4307    4365    4370
C10    4323    4346    4389
C11    4346    4395    4354
C12    4371    4341    4348

Only 1 of the R9 3900X does FCLK 1900MHz, MAX FCLK it can do is ~1903.5MHz.

One of the R9 3900X is shy of FCLK 1900MHz by ~8MHz, this one can do 4x8GB at 3733MHz.

One of the R9 3900X is shy of FCLK 1900MHz by ~4MHz, this one though the IMC seems pants, as I've yet to gain even 3733MHz MEMCLK on 4x8GB.

[bigtop1967]

My 2nd/3rd samples of R9 3900X I have dubbed MHz Editions .

These responds to PBO [Enabled] better than FCLK Edition not yet tried +xxxMHz . It also seems the cores are more similar so better for all cores OC .

This ZIP contains stock CPU Statuscore loading each thread with Ryzen Master compare.

2nd/3rd R9 3900X seems to have better multicore boost, not only when checked for clocks but benches. For example for multicore in Realbench stress mode 32GB 30min run 1st is ~4.0GHz all cores average effective MHz, 2nd is ~4.1GHz. When loading with P95 8K 4096K 27GB, 1st is ~4.025GHz all cores average effective MHz, 2nd is ~4.075GHz.

They all seem a little differing on ProcODT for RAM MHz.

        R9_1    R9_2    R9_3

CCD0

C01    4551    4546    4545
C02    4515    4517    4524
C03    4555    4484    4452
C04    4543    4542    4547
C05    4515    4479    4517
C06    4549    4463    4470

CCD1

C07    4369    4421    4407
C08    4313    4373    4406
C09    4307    4365    4370
C10    4323    4346    4389
C11    4346    4395    4354
C12    4371    4341    4348

Only 1 of the R9 3900X does FCLK 1900MHz, MAX FCLK it can do is ~1903.5MHz.

One of the R9 3900X is shy of FCLK 1900MHz by ~8MHz, this one can do 4x8GB at 3733MHz.

One of the R9 3900X is shy of FCLK 1900MHz by ~4MHz, this one though the IMC seems pants, as I've yet to gain even 3733MHz MEMCLK on 4x8GB.

Very interesting - thanks for posting. My 3900x seems to have pants FCLK though haven't tested as thoroughly as you. I did do some BCLK tests which got me nearer to 1900 FCLK, but not done much more since. Been waiting for the 1004 AGESA but they really are taking a long time to release it for the CH7.

I've been wondering if I should try RMA - but even if they accepted it, it looks like i will prob get a CPU of very similar performance.

Have you tested if these 2nd and 3rd gen CPUs are using less power like the 3950x? If that were true, I'd definitely RMA!

[gupsterg]

NP .

I wouldn't RMA just as you're CPU doesn't do FCLK 1900MHz, you could end up with slightly worse CPU on aspects I have noted between mine.

All the R9 3900X I have, are same production batch, BF 1944SUT with UG stamp on IHS. 2 of the CPUs are even close serial, but differ from each other in aspects.

First and foremost core MHz is king. Sample 2 under PBO edges past sample 3, ~25 to 50MHz averaged all cores boost. Sample 1 which does FCLK:UCLK/MEMCLK 1900MHz is behind them both by ~50-100MHz on average all cores MHz.

Secondly I've noted the delta between CCDs temperature can be larger on sample 3 compared with 1 & 2. Averaged each CCD temps difference is ~7C on sample 3 vs 2C on sample 1 & 2.



I shared before just each thread MHz for Statuscore per CPU, this ZIP has test loading each thread in sequence. I pitted sample 1 at stock vs sample 3 with PBOE, as sample 1 does not respond to PBOE I felt I wanted to compare best of each. You'll see From 1-12 sample 1 as having the better CCD0 is higher MHz, sample 3 then starts besting it as cores 13-24 are loaded. As both CCDs come into play I believe then the weaker CCD in away affects boosting that stronger CCD can't perform as good as it may.

After you digest that last test, next metric I believe to determine how good a CPU is, is how the cores for performance are ordered. Below left to right is sample 1, 2 & 3.



In HWINFO perf # first number is how SMU is categorizing core performance order. Sample 3 is best on that fact if an OS is not CPPC aware IMO, sample 2 is next best and lastly sample 1 (ie one which does FCLK 1900MHz).

The binning is tight regardless of these aspects. The best CB20 result I have so far is on sample 2, as I don't use all cores OC metric is either CPU stock or PBO if it responds to it.



Sample 2 & 3 are more consistent for run to run variance in a bench, as sample 1 has bigger core to core MHz difference it seems to differ more run to run, context multicore runs, single core all are the same ~529-531.

[bigtop1967]

NP .

I wouldn't RMA just as you're CPU doesn't do FCLK 1900MHz, you could end up with slightly worse CPU on aspects I have noted between mine.

All the R9 3900X I have, are same production batch, BF 1944SUT with UG stamp on IHS. 2 of the CPUs are even close serial, but differ from each other in aspects.

First and foremost core MHz is king. Sample 2 under PBO edges past sample 3, ~25 to 50MHz averaged all cores boost. Sample 1 which does FCLK:UCLK/MEMCLK 1900MHz is behind them both by ~50-100MHz on average all cores MHz.

Secondly I've noted the delta between CCDs temperature can be larger on sample 3 compared with 1 & 2. Averaged each CCD temps difference is ~7C on sample 3 vs 2C on sample 1 & 2.



I shared before just each thread MHz for Statuscore per CPU, this ZIP has test loading each thread in sequence. I pitted sample 1 at stock vs sample 3 with PBOE, as sample 1 does not respond to PBOE I felt I wanted to compare best of each. You'll see From 1-12 sample 1 as having the better CCD0 is higher MHz, sample 3 then starts besting it as cores 13-24 are loaded. As both CCDs come into play I believe then the weaker CCD in away affects boosting that stronger CCD can't perform as good as it may.

After you digest that last test, next metric I believe to determine how good a CPU is, is how the cores for performance are ordered. Below left to right is sample 1, 2 & 3.



In HWINFO perf # first number is how SMU is categorizing core performance order. Sample 3 is best on that fact if an OS is not CPPC aware IMO, sample 2 is next best and lastly sample 1 (ie one which does FCLK 1900MHz).

The binning is tight regardless of these aspects. The best CB20 result I have so far is on sample 2, as I don't use all cores OC metric is either CPU stock or PBO if it responds to it.



Sample 2 & 3 are more consistent for run to run variance in a bench, as sample 1 has bigger core to core MHz difference it seems to differ more run to run, context multicore runs, single core all are the same ~529-531.

Exactly my thoughts - thanks for the extra data!

I have found while comparing my performance with others that I often beat those with 1900 FCLK anyway, indicating that some are pushing further than is beneficial..... Though of course we all want 1900 FCLK

[Riverdog64]

I have the Crosshair Vii X-470 w/ Ryzen 7, Radeon 7. Used mostly for gaming, I was curious about the overclock presets in the bios. Most of you all are well beyond using such basic stuff, but for a newbie are they a safe way to start overclocking my system?