Here here! I have been loving this board on my big bench loop and on LN2. The platform has a few quirks under LN2 conditions, but the board has behaved flawlessly.



Mine has been run on a 3970X ES as well. Waiting for a Retail 3960X to become available and then I'll test that chip

 

wish i could afford this platform, but my 3950x is keeping me happy just fine. sorry for having nothing to contribute to this thread

 

Excellent write-up.

Now if only I could actually find a 3970X for sale anywhere on the planet...

 

Overclockers UK, among many others that sell Worldwide have stock.. Once VAT is removed(it is done automatically if shipping outside EU) , it's about 2070 USD + Shipping by today's conversion rate.

 

Ryzen Threadrippers are based on a different product (Matisse vs. Starship)

Are you sure about the Starship codename? As far that I know, latest ThreadRippers are Castle Peak: https://en.wikichip.org/wiki/amd/cores/castle_peak
Starship is supposed to be an EPYC Embedded line product, successor to the current 3000 series (Which I'm very interesed in, by the way. That is why I'm used to that codename).

 

Are you sure about the Starship codename? As far that I know, latest ThreadRippers are Castle Peak: https://en.wikichip.org/wiki/amd/cores/castle_peak
Starship is supposed to be an EPYC Embedded line product, successor to the current 3000 series (Which I'm very interesed in, by the way. That is why I'm used to that codename).

Starship and Matisse are code names of the dies, as referred by AMD docs.

Rome = SSP, Castle Peak = SSP, Matisse = MTS.

Different families, different docs, different AGESA, different fuses, etc.

 

Bought one just 10 mins ago!

[BTW thank you very much! You remebered me i'm able to buy from other EU countries'a amazon withouth any fee... i was going crazy seraching one for a reasonable price and being able to recive it in normal time in Italy.. and anyhow "Sold and Shipped by amazon" is a guarantee itself.. rather than have to deal with,scary, 3dy part sellers..]


Now, considering this is my first time ever with amd, i'd love if somene could fill in the blanks:

1) At the moment i have 3000Mhz memories: 4x 16gb Corsair Dominator platinum.
Should i change them? I mean, i keep reading Ryzens work better with high frequency memories.. obviously i can overclock theam a littel more but i honestly don't like to push memories to
the max .. i doubt i'd ever be confortable with 3000 named rams pushed to 4000 for a dailly,,,just saying..


2) Could you kindly suggest me a coule of M.2 to mount on this build [M.2 is another thing i'm noob with.. never had one. I'm getting old boyz, this one colud be my last hi end pc ever, so please... suggest me sothing stunning].

 

The ideal speed for ryzen 3rd Gen seems to be 3600, and, if achievable.. Cl14. For this you need either Samsung b-dies or Micron-E chips.

Corsair sells memory with Samsung b-dies (as do Gskill and many others) , but it's a guessing game on the lower speed kits. For guaranteed bdie you need something like 3200 cl14 kit, or 3600 c16. But do pay attention to sub timings, b-die kits usually have sub timings in par with the first one, like.. CL14-14-14-34... Or buy something 3800 and above, that is guaranteed bdie. You can also use this website for some help:
https://benzhaomin.github.io/bdiefinder/

The alternative would be Crucial Ballistix Elite, 3600 or 4000. Those are guaranteed Micron-E and have great compatibility with Ryzen.

You don't need to change your kit, but you can get a decent boost in performance with a better one with the improvements in timings mostly, some applications are very sensitive to lower timings.

M.2 SSDs.. Samsung 970 EVO Plus is the most universal agreed suggestion. Sabrent is the cheaper alternative.

If you want to try something pci-e gen4.. You can try:
Corsair MP600 Gen4
Gigabyte Aorus Gen4
Sabrent Rocket Gen4

 

3rd gen. Threadripper CPUs have a power management related quirk, which we should discuss about.

Unlike the 3rd gen. Ryzen AM4 CPUs, 3rd gen. TR CPUs do not use telemetry for their power management related decisions.
AM4 CPUs base their decisions (i.e. know their current power consumption) on the current & voltage telemetry (SVI2 TFN), sourced from the motherboard VRM controller.

Meanwhile on the 3rd gen. Threadripper CPUs, this is not the case.
Instead of relying on external telemetry, the power management decisions are based solely on internal calculations (voltage, frequency, utilisation, etc).
I'm not sure what are the reasons behind this configuration, but I'd expect it to have something to do with TRs close relation to Rome EPYC (which uses cLDOs for the main planes) or possibly to
the limitations of the AMD SVI2 standard itself.

In practice, this can affect the users in couple different ways. In my case, even at stock, the power consumption seen by the CPUs power management is slightly inflated, roughly by 6-11% (depending on the workload).
Because the CPU is pegged against the default 280W power limit in most properly multithreaded workloads, this means that some performance is lost.

Granted, the CPU I'm using is an engineering sample so the inflated power reporting at stock can be an ES related anomaly which sometimes do exist.

However, the next issue will affect anyone how has plans to undervolt the CPU. While undervolting 3rd gen. Ryzens generally isn't recommended, at least in my case there is a ~ -50mV undervolting margin to be had, without any performance penalty in any workload.
The issue with this internal power management mode with undervolting is, that unlike with telemetry the CPU cannot "see" it. Despite the undervolting the CPU still calculates the voltage based on its internal targets / models and doesn't acknowledge that the voltage has reduced.
Because of that, the power consumption seen by the CPU doesn't change either. There will usually be a small performance boost from undervolting, but thats due to the lowered temperature and nothing else.

Here is an example:

This is at stock, with the default 280W PPT, 215A TDC and 300A EDC limits. PPT and EDC are constantly pegged to their limits.
Despite the PPT is reading ~ 280W constantly during the workload, the actual measured power consumption for the whole package is 263.178W on average.
The average CPU voltage is 1.15516V.

Now, everything else remains identical, but a negative offset of 50mV has been applied for the CPU.
The average, actual power consumption has reduced to 243.809W, yet the CPUs power management still sees the same power draw as before.
You can also see that the CPU voltage has also reduced to 1.11357V and that the CPU temperature has reduced by 3.25°C.

Reducing the CPU voltage by -50mV improved CB20 NT performance by ~ 40, which is far less than expected improvement. In this case the improvement is solely
caused by the lowered temperature (which the CPU acknowledges).

So how to get the full advantage of undervolting then?
As I've said before, the most recent version of HWInfo has complete support for Zenith II Extreme.
Since it is able to report the output currents and powers for both of the domains accurately, you can make adjustments based on the observed delta.
Prior undervolting, write up the averages of "CPU Core Power (SVI2 TFN)" and "CPU Core Current (SVI2 TFN)" during a sustained and stable workload (e.g. 5 minutes of Blender).
Apply the undervolt and repeat the same procedure for these two values.

When you have both sets of these values, go to the "Precision Boost Override" menu in the bios and select "Manual".
The new PPT limit will be 280W + the difference between the stock and undervolted "CPU Core Power (SVI2 TFN)" power. TDC will be 215A + the difference between the stock and the undervolted "CPU Core Current (SVI2 TFN)".
EDC can be set to reflect the same percentual difference as TDC was increased by, added to the default 300A value. PBO Scalar should be manually set to 1x to avoid it changing from the stock, when manual PBO mode is used.

So in my case, the power consumption of the CPU cores themselves (CR Pout) reduced 197.329W to 178.302W (19.027W) and the current for the CPU cores (CR Iout) reduced from 170.813A to 160.329A (10.484A).
Hence, my new PPT limit became 299W (instead of 280W), TDC limit became 226A (instead of 215A) and EDC limit became 315A (300 * (226/215)).

Doing the whisky-tango calibration described above, Cinebench R20 NT score improved by ~300pts with the -50mV undervolt, while maintaining pretty much the same power consumption as stock.

 

Reducing the CPU voltage by -50mV improved CB20 NT performance by ~ 40, which is far less than expected improvement. In this case the improvement is solely
caused by the lowered temperature (which the CPU acknowledges).

So how to get the full advantage of undervolting then?
As I've said before, the most recent version of HWInfo has complete support for Zenith II Extreme.
Since it is able to report the output currents and powers for both of the domains accurately, you can make adjustments based on the observed delta.
Prior undervolting, write up the averages of "CPU Core Power (SVI2 TFN)" and "CPU Core Current (SVI2 TFN)" during a sustained and stable workload (e.g. 5 minutes of Blender).
Apply the undervolt and repeat the same procedure for these two values.

When you have both sets of these values, go to the "Precision Boost Override" menu in the bios and select "Manual".
The new PPT limit will be 280W + the difference between the stock and undervolted "CPU Core Power (SVI2 TFN)" power. TDC will be 215A + the difference between the stock and the undervolted "CPU Core Current (SVI2 TFN)".
EDC can be set to reflect the same percentual difference as TDC was increased by, added to the default 300A value. PBO Scalar should be manually set to 1x to avoid it changing from the stock, when manual PBO mode is used.

So in my case, the power consumption of the CPU cores themselves (CR Pout) reduced 197.329W to 178.302W (19.027W) and the current for the CPU cores (CR Iout) reduced from 170.813A to 160.329A (10.484A).
Hence, my new PPT limit became 299W (instead of 280W), TDC limit became 226A (instead of 215A) and EDC limit became 315A (300 * (226/215)).

Doing the whisky-tango calibration described above, Cinebench R20 NT score improved by ~300pts with the -50mV undervolt, while maintaining pretty much the same power consumption as stock.

Hi Stilt, do you have undervolting tips for AM4 3000 Ryzen?
Sorry if i missed out.

 

What do u guys think Zenith IIvs the Gigabyte Aorus Xtreme ?

I'm leaning towards the Gigabyte because it has an overall better feature set.

Intel dual 10g lan (not aquantia + 1gig)
True 16 phase controller
Better pcie slot layout I'll be using all slots.

For the Zenith
Dimm.2
Better nvme drive pcie layout ( I'll be using min 4 m.2 )

Better looks wise.
Asus bios vs giga.

 

Asus bios is 10x better than..well anything else..
Now, it could be just me, so let say IMHO.. still..it's better. Period

That being said, those cards are both beasts,you can't go wrong.
My suggestion is to consider which hw feature is more usefull for you now and in teh next future.

Persanoally i have gone for asus because the bios.It's good,it's intuitive, it makes me feel at home ,i'm using it from 10 years on my own main pc now... but that's me.

 

This board has a total of 5 M.2 slots. That's 2 or 3 more than other boards. How are the M.2 PCIE lanes fed?

From the Manual I can see:

Dimm.2 1: No Impact
Dimm.2 2: No Impact
M.2_1 (PCIE Slot Area Upper): No Impact
M.2_2 (PCIE Slot Area Lower): PCIE X16_4 will run at 4X mode
M.2_3 (Behind Motherboard): SATA6G_E1-E4 will be disabled

From the AMD block diagram, I can see we have provisions for two M.2 from the CPU, and potentially two feeding from the TRX40 Chipset. The M.2_3 is clearly fed from the TRX40 Chipset but it's not clear where the other M.2 slots are being fed from.

 

The two Dimm.2 slots are connected to the CPU, the rest are to the chipset. The M.2_3 is getting the 4 PCIe lanes that are powering 4 of the SATA ports.
The chipset itself supports up to 12 SATA ports but the Zenith II only has 8xSATA. This makes me think that the last 4 PCIe lanes that can be done to SATA ports have been re-purposed for one of the M.2 slots. Then you have another m.2 sharing 4 lanes with the PCIe slot that goes from 8x to 4x.
Having in mind the placement of the M.2 slots on the board and how hard it is to place PCIe4 devices far from the controller, it is fairly certain that only the Dimm.2 slot is connected to the CPU and the other three m.2 slots to the chipset. That should not present any issues unless you want to put 5x PCIe4 nvme SSDs and run all of them in some extreme RAID configuration.
....
Actually... after looking at the TRX40 diagram again, I'm beginning to suspect that M.2_1 might also be connected to the CPU...

 

_Actually I answered my own question by finding this. M.2_1 and M.2_2 are off the CPU. Dimm.2 and M.2_3 are off the chipset.

 

Are you sure this diagram is correct? It seems very unlikely for the DIMM.2 slots that are located next to the CPU to get their lanes from the chipset that is so far away... It is possible but strange...

 

its all application dependent with Threadripper as it is Ryzen

 

I have been enjoying my 3970x and the Zenith II Extreme quite a bit.
I have 8 sticks of 8GB G.Skill 3600mhz CL15 sticks. (four two pack kits)

I have been trying to dial in what I call 'stock' settings.
Ram about where the XMP profile says.
CPU with default behaviors allowed (some spread spectrum and power saving options off).
Fixed voltages for SoC, CPU, and Ram.

So with that said I am completely stable (Prime 95 w/ AVX used for testing, both high memory and small FFT tested separately):
1.105 volts on vCore + LLC 4
1.39 volts on dram AB and CD
1.15 volts on SoC + LLC 3

My Stock Bios Settings:
https://imgur.com/a/d0BRh78


One quirk I have noticed. My cpu is completely stable (compute and intense memory utilization) @ .9 volts. However! The ancillary components experience what I am calling hiccups. By ancillary I mean, the ethernet ports, the usb ports, and the sata ports. (NVME drives not affected)
With that low of a voltage the SATA ports are sometimes shown, sometimes not during POST. In windows they function fine, but sustained writes have blips of high latency every 2-4 seconds.
Also the usb ports while writing to a USB stick have the same symptoms. They function perfectly fine, but have hiccups in sustained reads/writes.
The ethernet ports also have the same behavior, but it is much more annoying. While playing games for example your ping may be steady at 25ms to the game server, but occasionally blips to 450ms!!!!

After doing a lot of testing I found that I need 1.1 volts on the cpu cores to get rid of those issues, making my idle temperatures 10C hotter at 40 degrees. Under load, not change in the high temps (59C).



The Soc was ok at 1.1 when I only had 6 dimms. When i added in the remaining 2 I needed 1.15 volts to keep my stock settings of 3600mhz CL15.

Other small notes: I am not using the molex connector at the bottom of the board, nor am I using the extra PCIe power connector near the cpu power plugs. I have to order new power cables to reach these locations for my build. So if this adds more stability at lower voltages I would not know at the moment.


--------------------------EDIT----------------------------------

Did some quick overclocking this evening:

 

is the hiccups you are experiencing related to the OC?
do you have the same problem when using default bios settings?

do you use PBO or manual OC?

 

Why does my performance drop when the only thing I set is xmp?
Im using this board and a 3960x, was on bios 403 now I'm on 602. Performance dropped alone from three bios change...I only set xmp on the newer bios by the way. My memory is trident z 3600 cas 16.

 

Make sure to manually set the infinity fabric to 1800 to match your 3600mhz ram. (1800*2=3600 (DDR = double date rate))

 

The hiccups are related to setting a manual vCore (no overclock). Take a look at my link to the bios settings I am using.
That is default behavior as far as I am aware except for setting a manual vCore and vSOC.

(The default behavior of asus made the vCore go up to 1.5 volts far too often and stayed there....)

 

why setting a manual vcore if no overclock?

 

[/quote]

Make sure to manually set the infinity fabric to 1800 to match your 3600mhz ram. (1800*2=3600 (DDR = double date rate))[/QUOTE]

I did and my performance was lower still. I believe I enabled pbo to obtain similar scores to stock. Granted temps rose tremendously in comparison to stock. Since im using a temp cooler (h100v2). I'm a bit hesitant to really push things. Currently doing 2 builds and is more concerned about stability over performance. Originally I went x570 and had my 3950x in hand but sold it on Ebay to fund this project and im glad I did.

Having 24 cores is amazing and that feeling I had when I first got my 1950x on launch is back. I think mainstream desktop is a no go for mde from now on. I can't fathom ruining a chip just by trying to remove its cooler(hasn't happened yet) but am4 socket design needs to be a bit more secure at well.. securing the chip in the socket.

Enough rambling, I definitely see that there's more to do on this iteration of threadripper and will test some more once my CPU block comes. I've got an epic system in the making but AMDs lack of at least a 2080 super competitor is killing me. I'll have to bypass my gpu for now.

 

[. . .]

Having 24 cores is amazing and that feeling I had when I first got my 1950x on launch is back. I think mainstream desktop is a no go for mde from now on. I can't fathom ruining a chip just by trying to remove its cooler(hasn't happened yet) but am4 socket design needs to be a bit more secure at well.. securing the chip in the socket.

Enough rambling, I definitely see that there's more to do on this iteration of threadripper and will test some more once my CPU block comes. I've got an epic system in the making but AMDs lack of at least a 2080 super competitor is killing me. I'll have to bypass my gpu for now.

Just go nvidia, they have the better gpu at the moment. And unless amd's gpu teams pull a ryzen, they aren't going to be top end competitive.
Unless you need the compute performance that amd is natively better at.


PBO was overvolting my 3970x to 1.5volts far too often for my liking, I turned that off. Even though I overclock for benchmark world records, I run it basically at stock on the daily.

 

guys i got both 3960x and 3950x and i still don't know how to convince myself towards new TR platform.

All motherboards seems to be lucklasting and very expensive... the only one worth even mentioning is Asus.

 

The future of storage is NVME. With the TR asus board you can use 5 NVME drives at full speed natively before having to add any add-in cards (which it can also support at full speed).

That and the tons of cores will be helpful when the next generation of console games get ported poorly to pc and are ridiculously hard to run.


I plan on keeping my 3970x and board for like 10 years.

 

i am afraid to invest again is such a "pro" platform is usually suffers from lack of updates .. ;-(

 

I'd say the initial release software is already mature. The agesa code has all the improvements that ryzen 3k has received (from what I have read).

 

what bout gigabyte extreme

 

Happy to see that there is no major issue with the platform.
Are there any sata problems? Usb disconnections or something similar?

 

i just snatched 3960x from amazon us there were 10 procs which sold out in matter of minutes... it was sold at 1399.
Newegg has it at 1499.

 

I'm waiting for the cpu to arrive meanwhile I plugged the board to see if it goes thru the post but it does not even boot up only rgb lights up without post code. Should it at least run thru post codes without cpu ?

 

no, no post codes.

but the bios flash feature does work

 

Is anyone running the latest 702 bios? I get no post if I set my memory to its xmp/docp settings. Being that this is a big update I can't revert. Bios #2(408) works perfectly fine with my docp settings though.

 

Can u post BIOS settings and spec ? I'll be moving to testbench with this for next week or so and if tested properly i will

 

I have trident z royal memory 16 16 16 36 3600mhz(4 8gb sticks), this board, 3960x, 1000w titanium seasonic, 970evo plus. The only setting in the bios changed was from auto to docp. I manually set memory to 3600 and fclk to 1800 as well and that was a no go as well. It worked on every bios previous to 702. My secondary bios is 408 and doing the same change actually yields a post.

 

works perfect on mine 702 with b-die 16 16 16 36

 

with that many CCXes, i find it quite time consuming to test each so thought perhaps an auto tune tool might come in useful.
https://www.dropbox.com/s/w0evca1tj7yflcj/CCX_AutoTuner_1226.rar?dl=0

load defaults,
set any additional bios params you wish to use as test base, for example loadline level 3, or manual over-ride voltage 1.25v, or offset -0.05v or DOCP, etc)
create a shortcut to CCX_Tuner.exe, place shortcut in shell:startup folder to launch at start.
it uses prime95, with avx disabled (just change local.txt CpuSupportsAVX=1 for avx).
enter duration, enter VID you're comfortable with (if you set manual over-ride voltage from bios then doesnt really matter),
enter starting all core freq or just 0 to estimate one from vid.

Note: if you stop it midway it will assume instability. if you wish to restart midway, just delete log.csv.

it will start pushing CCXes in order and save results into CCX_Results.txt when all done.
tested on a few procs on C8 and zenith 2

 

The SoC power consumption on third gen. Threadripper greatly depends on the memory configuration and the clocks. Typically, the worst-case (Ram Test, Prime Blend, Linpack) power consumption is less than 45W when running 1 DPC DR memory configuration
at 3600MHz with FCLK and UCLK in sync. 45W at the typical ~ 1.05V operating voltage is just < 43A, which is peanuts for the SoC VRM, considering its true capabilities. The SoC VRM temperature stabilises rather quickly and I've yet to see it exceeding 69°C at any point.

My chipset fan runs pegged out at 4500 rpm and the chipset temp reads 102F. I'm running 64GB of DDR4 3600 and two M.2 in raid 0. The fan is kind of annoying. Do I have a bad sensor or does raid 0 NVME really generate that much heat?

 

Are those m.2 drives PCI gen3 or gen4?
Gen4 drives are usually warmer and the chipset will also heat-up more because of it and the RAID management.

 

Can anyone confirm DIMM.M2 is at the CPU?

Also, RAM seems to be the issue. Getting a kit that's supported by the motherboard for 128GB is hard AF, let alone 256 GB (seems none exist).

 

It’s not. I verified using lstopo. The only m.2 slots that come from the CPU are M.2_1 and M.2_2, but M2_2 shares bandwidth with the last PCIE slot. I didn’t want to believe it, but the block diagram from guru3d seems to be correct. The owner’s manual had all 5 coming from the CPU.