Crosshair VII Hero Essential Info Thread

[gupsterg]

UEFI 2901 has become my fave so far .

I'm using 4x8GB with R5 3600+C7HWIFI, UEFI 2801 has been pretty sound, I had noted at times when I expected the board to power down and up on a UEFI setting change it did not for some, this seems to have been rectified in UEFI 2901.

I decided to take my usual 3800MHz C16 profile, loosen a few timings and aim to cut VDIMM from ~1.405V, meddling with the PMU menu I seemed to have gained it now with ~1.35V.

This ZIP has my base profile with 3800MHz C16 @ 1.35V settings and test data. If you are gonna set SOC via AMD OC menu do change SOC mode on Extreme Tweaker page to [Offset/+/Auto].

3x AIDA64



3x AIDA64 with +100MHz PBO



Gonna continue further testing and aim to get back to +150MHz PBO.

[Technos15]

I couldn't find SVM enable in Bios 2901, so switched back to bios 2801 to get VirtualBox working again.

[robbiec]

OT on the AGESA thing but has anyone had any luck in getting an Asus Hyper M.2 V2 card working on the Crosshair VII?
By working I mean showing all 4 M.2 devices and not just the first one.

[bluestang]

UEFI 2901 has become my fave so far .

I'm using 4x8GB with R5 3600+C7HWIFI, UEFI 2801 has been pretty sound, I had noted at times when I expected the board to power down and up on a UEFI setting change it did not for some, this seems to have been rectified in UEFI 2901.

I decided to take my usual 3800MHz C16 profile, loosen a few timings and aim to cut VDIMM from ~1.405V, meddling with the PMU menu I seemed to have gained it now with ~1.35V.

This ZIP has my base profile with 3800MHz C16 @ 1.35V settings and test data. If you are gonna set SOC via AMD OC menu do change SOC mode on Extreme Tweaker page to [Offset/+/Auto].

Will this profile work/load properly on the non-WiFi CH7? Just got a 3900X so was wondering if/how it would work or not with it and my 4x8GB G.Skill 3600 C15 sticks.

[gupsterg]

OT on the AGESA thing but has anyone had any luck in getting an Asus Hyper M.2 V2 card working on the Crosshair VII?
By working I mean showing all 4 M.2 devices and not just the first one.

Ref the block diagram.



Ryzen mainstream platform has 24 gen 3 lanes.

4x CPU/Chipset link

4x M.2_1

16x lanes can go to PCI-E x16 slot 1, if PCI-E x16 slot 2 comes into use then each get 8x. If M.2_2 ever comes to use, regardless if PCI-E x16 slot 2 is used or not, PCI-E x16 slot 1 will become 8x, PCI-E x16 slot 2 will become 4x, as 4x from that will be used for M.2_2.

The UEFI has an option within Advanced > Onboard Devices Configuration, which will disable M.2_2 and make PCI-E x16 slot 2, 4x/4x, so 2 of the M.2 slots will work on Asus Hyper M.2 V2 card. In the end whichever way you look at you gain only 1 extra PCI-E 4x gen 3 M.2.

[gupsterg]

Will this profile work/load properly on the non-WiFi CH7? Just got a 3900X so was wondering if/how it would work or not with it and my 4x8GB G.Skill 3600 C15 sticks.

The profile is shared to show what can be achieved. As there can be variance between silicon, IMC, etc I can not say the exact same settings will work.

[happystargazer]

Does anyone know when we can expect a UEFI version for 3950X cpus? Like Agesa 1.0.0.4b?

I switched from AsRock to Asus C7H because Asus was better in terms of UEFI, fast new version, lot of betas.

But since a time they switched positions. AsRock pushes one UEFI after another and i am still waiting.

No information, no ETA, nothing.

I really feel left alone as customer.

[gupsterg]

I couldn't find SVM enable in Bios 2901, so switched back to bios 2801 to get VirtualBox working again.

It's in the usual place.

Advanced > CPU Configuration

Will this profile work/load properly on the non-WiFi CH7? Just got a 3900X so was wondering if/how it would work or not with it and my 4x8GB G.Skill 3600 C15 sticks.

The profile is shared to show what can be achieved. As there can be variance between silicon, IMC, etc I can not say the exact same settings will work.

Another thing I did to improve 4 dimms stability was to check dimms for which worked best in which slots, an effort to improve signalling.

I tested in pairs at first, I found 2 dimms were slightly less stable at same setup than other 2.

The pair which was less stable I placed in slots B1 & B2, as they would be closer to CPU, thus perhaps shorter tracing = improved signalling to them. What I thought were the more stable dimms I placed in A1 & A2.

Secondly I compared how the system behaved for OC'ing in that configuration and then interchanged the dimms. I moved B1 to B2 and B2 to B1, A1 to A2 and A2 to A1.

After doing these test cases I noted a configuration had more "post'ability" at higher frequency with wider range of ProcODT.

For example originally at 3800MHz on 4x8GB I had used ProcODT of 34.3, if I used any other system would Q-Code: F9 and after few posts reset OC. Once I had sorted the dimms, I had improved "post'ability", stability, etc.

The profile in post 161 I reapplied to rig today, I tested it for ~7hrs uptime, mixed testing Kahru RAM Test, P95 non AVX 28GB, Y-Cruncher, RealBench and HCI v6.0, here is ZIP link of screenies, etc.

[robbiec]

Ref the block diagram.



Ryzen mainstream platform has 24 gen 3 lanes.

4x CPU/Chipset link

4x M.2_1

16x lanes can go to PCI-E x16 slot 1, if PCI-E x16 slot 2 comes into use then each get 8x. If M.2_2 ever comes to use, regardless if PCI-E x16 slot 2 is used or not, PCI-E x16 slot 1 will become 8x, PCI-E x16 slot 2 will become 4x, as 4x from that will be used for M.2_2.

The UEFI has an option within Advanced > Onboard Devices Configuration, which will disable M.2_2 and make PCI-E x16 slot 2, 4x/4x, so 2 of the M.2 slots will work on Asus Hyper M.2 V2 card. In the end whichever way you look at you gain only 1 extra PCI-E 4x gen 3 M.2.

Thanks for the really detailed response. Appreciated. I'm running a couple of Vega64s and a couple of Samsung Evo Pluses so yep, PCIe lane limited on AM4

[gupsterg]

Thanks for the really detailed response. Appreciated. I'm running a couple of Vega64s and a couple of Samsung Evo Pluses so yep, PCIe lane limited on AM4

Ahh, no problem.

Ok fellow ROG'ers . My launch purchased R5 3600 has been doing 3800MHz C16 on 4 dimms for fair while, only change I have noted is with UEFI 2901 I could lower VDIMM to 1.35V vs earlier used ~1.4V. I changed to a R7 3700X, produced week 43 of 2019, so pretty fresh silicon and it can not do FCLK of 1900MHz , it falls shy by about ~3.5MHz .

This album has the photos of CPUs and UEFI screenies of tweaking BCLK on FCLK 1867MHz to see where abouts the R7 3700X fell short on FCLK.

This ZIP contains some tinkering with the R7 3700X, best to organise files by time to see process.

Initially the stock PPT/TDC/EDC limits held back all cores boost under Kahru RAM test, then some more scaling of frequency could be had by adjusting scalar.

To me it would seem after 6x there was no point in increasing scalar. This mornings testing when rig was cool, I could see higher all cores boost initially in same testing and as temperature increased it lowered, I forgot to capture this at the time.

The R7 3700X initially was tested with same SOC/CLDO_VDDP/CLDO_VDDG as R5 3600, so far I have managed to get it down from 1.062/0.927/0.980 to 1.043/0.921/0.972, VDIMM/VTTDDR/ProcODT is still the same as used with R5 3600, 1.35/0.675/43.6. I don't know if I should bother lowering any further as SOC/CLDO_VDDP/CLDO_VDDG stock is 1.025/0.9/0.95.

I also yesterday did some testing by increasing CPU clock for PBO, +100MHz didn't see CPU boost past 4.425GHz in Kahru RAM test, again only noted on a core IIRC, filename being with P100 has this capture.

I have settled on keeping CPU at stock for clocks, but given it PBO limts as 105W CPU (142/95/140) and using 3733MHz 1:1:1 (FCLK:UCLK:MEMCLK). I may tweak some timings, but when recently I did this with R5 3600, the gains were diminishing and hurt stability for "repeat'n'rinse" testing of profile.

The R7 3700X is showing ~4.3GHz ACB for Kahru RAM test, best I could get on R5 3600 was ~4.35MHz with PBO+150MHz, so the change to R7 3700X is approximately loss of ~50MHz ACB, but gain of ~33% more cores/threads over R5 3600. The R7 3700X shows the same aspect as R5 3600, once ~>60W is pulled through CPU the boost algorithm will curb ACB down from what I see with Kahru RAM Test, which pulls ~60W on both CPUs.

I have used a beta UEFI with AGESA 1.0.0.4 (link), with this UEFI latest Ryzen Master is showing activity on PPT/TDC, etc.

I set SOC mode as Offset/+/Auto on Extreme Tweaker and set SOC in AMD OC menu, this way with OC RAM at POST, SOC voltage is as I set, otherwise it can bounce to ~1.1V. I also set CLDO_VDDP and CLDO_VDDG in AMD OC Menu. Precision Boost tweaks I apply via the Extreme Tweaker menu, but scalar I change in AMD OC menu, otherwise it does not seem to apply. Using these methods also Ryzen Master shows set SOC/CLDO_VDDP/CLDO_VDDG correctly.

When using a 3000 series CPU I would set:-

Global C-state Control: [Enabled] (link to UEFI screenie of option)
CPPC: [Enabled] (link to UEFI screenie of option)
CPPC Preferred Cores: [Enabled]

With Global C-state Control enabled you will see FCLK clock be dynamic in HWINFO, The Stilt's post on this.

My current R7 3700X settings in this ZIP.

*** edit ***

A compare of P95 of R5 3600 using PBO+150MHz vs R7 3700X PBO as 105W CPU, link to ZIP. Other than extra cores/threads each is averaging very similar for MHz, temps, etc.