How about with PCH lanes?
Well then, only for you I’ll dust off a folder with AS SSD when I’ll be testing the next mainboard unit - don’t blame me if it says something like negative 1.21 sec as latency 
I’ll go look where I have the previous results of the chipset-supplied PCIe lanes.
Found the X570-Chipset slot results, Optane 905P 480 GB #1 was drive D: and got its lanes via CPU (PCIE 3):, Optane 905P 480 GB #2 was drive E: and was installed in PCIE 5 of the Taichi (chipset PCIe slot):
CrystalDiskMark 6.0.2 x64 © 2007-2018 hiyohiyo
Crystal Dew World : https://crystalmark.info/
MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
KB = 1000 bytes, KiB = 1024 bytes
Sequential Read (Q= 32,T= 1) : 2784.179 MB/s
Sequential Write (Q= 32,T= 1) : 2480.782 MB/s
Random Read 4KiB (Q= 8,T= 8) : 2435.072 MB/s [ 594500.0 IOPS]
Random Write 4KiB (Q= 8,T= 8) : 2468.074 MB/s [ 602557.1 IOPS]
Random Read 4KiB (Q= 32,T= 1) : 518.451 MB/s [ 126575.0 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 513.638 MB/s [ 125399.9 IOPS]
Random Read 4KiB (Q= 1,T= 1) : 227.881 MB/s [ 55635.0 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 206.931 MB/s [ 50520.3 IOPS]Test : 1024 MiB [D: 0.0% (0.1/447.1 GiB)] (x5) [Interval=5 sec]
Date : 2019/07/27 20:59:20
OS : Windows 10 [10.0 Build 18362] (x64)
CrystalDiskMark 6.0.2 x64 © 2007-2018 hiyohiyo
Crystal Dew World : https://crystalmark.info/
MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
KB = 1000 bytes, KiB = 1024 bytes
Sequential Read (Q= 32,T= 1) : 2776.345 MB/s
Sequential Write (Q= 32,T= 1) : 2450.234 MB/s
Random Read 4KiB (Q= 8,T= 8) : 2429.078 MB/s [ 593036.6 IOPS]
Random Write 4KiB (Q= 8,T= 8) : 2462.459 MB/s [ 601186.3 IOPS]
Random Read 4KiB (Q= 32,T= 1) : 508.509 MB/s [ 124147.7 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 480.098 MB/s [ 117211.4 IOPS]
Random Read 4KiB (Q= 1,T= 1) : 252.247 MB/s [ 61583.7 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 234.019 MB/s [ 57133.5 IOPS]Test : 1024 MiB [E: 0.0% (0.1/447.1 GiB)] (x5) [Interval=5 sec]
Date : 2019/07/27 21:04:33
OS : Windows 10 [10.0 Build 18362] (x64)
Edit: Corrected copy & paste error with multiple text files open, sorry.
@aBav.Normie-Pleb So the below result with 252.24 MB/s 4KQ1T1 random read throughput is achieved with PCH lanes? It’s even better than CPU lanes.
Yeah, keep in mind that that was with earlier AMD chipset drivers and BIOS versions (late July).
I’ve tried to keep the results “clean” (unplugged ethernet cable and made sure Windows wasn’t doing anything in the background), but I’m not Gamers Nexus and repeat each test 100 times and establish any margin of error etc.
Also note that I used PCIe-M.2 adapters combined with Intel’s OEM M.2-to-U.2 adapters so not exactly an “ideal” setup regarding signal integrity but it worked just fine and didn’t influence anything compared to the M.2-U2 adapter directly installed in an M.2 slot supplied by the CPU.
. Have you tried virtio-scsi on Optane also?Since I’m a normie-pleb, the most VMmie stuff I do is “click-‘n’-slay” ESXi, so unfortunately not.
Exactly one of the main reason why i don’t really understand,
why people limit their brand new Ryzen3000 cpu,
by buying an older chip set board.
I mean sure if you already have a previous gen board,
then upgrading the cpu for more cores makes sense.
But if you buy a new setup, then why bother with previous gen boards?
@ MisteryAngel It’s probably because while some people doesn’t even notice SATA SSD and PCIe SSD anyway, but the same socket with newer CPUs makes possible to upgrade from old computer with less budget.
I myself, of course, will buy 3950x with X570 board 
1 Like
Yup makes sense for those who already have like a X470 board with 2700X,
and want to upgrade to higher core count chip.
But for those who buy new setups, i see allot of ¨tech tubbers¨ recommend,
to get an cheaper B450 board with a 3700X /3800X etc.
And i just cannot really get at all.
But get back to your question, on X570 + 3000 series cpu,
the dmi between the cpu and the chipset is 4x gen4.
So there basically is more bandwith available to the chipset / pch.
1 Like
Poor AMD, even in a tech-savvy forum, Intel marketing terms are being used for describing the interfaces/components (DMI and PCH)
Somehow I got the feeling it was time for an early fall/autumn cleaning of the other system that currently uses the Optanes so I might report back a little sooner than in a few weeks.
Yup, I already know that but I don’t know more appropriate term for this. I can’t call this Southbridge either lol(also derived from Intel). The term “Chipset” has somewhat confusing meaning.
Nothing wrong with those terms if you ask me
The reason older AM4 boards are recommended as an alternative is there are an extra 4x pci-e 3.0 lanes from the cpu dedicated to an m.2 slot, so the chipset providing pci-e 2.0 doesn’t effect most people any more than Intel’s mainstream setup does.
Repeated the tests today.
Details:
- 3700X, PBO disabled since it is said to be buggy at this point
- 4 x 16 GB Samsung DDR4-2400 ECC UDIMM @ DDR4-3200 with loose Auto-Timings
- Infinity Fabric clock speed set manually to 1900 MHz (DDR-3800)
- ASRock X570 Taichi with UEFI 2.10/AGESA 1.0.0.3 ABBA
- Latest AMD chipset drivers from mid-August 2019
- Windows 10 1903 Patch level 2019-09
- Standard Microsoft NVMe driver
- No ethernet adapter connected
2 x Intel Optane 905P 480 GB (D:/E:)
- Drive D: is connected via X570 chipset
- Drive E: is connected directly to CPU
CrystalDiskMark 6.0.2 x64 © 2007-2018 hiyohiyo
Crystal Dew World : https://crystalmark.info/
MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
KB = 1000 bytes, KiB = 1024 bytes
Sequential Read (Q= 32,T= 1) : 2777.208 MB/s
Sequential Write (Q= 32,T= 1) : 2449.609 MB/s
Random Read 4KiB (Q= 8,T= 8) : 2429.958 MB/s [ 593251.5 IOPS]
Random Write 4KiB (Q= 8,T= 8) : 2460.176 MB/s [ 600628.9 IOPS]
Random Read 4KiB (Q= 32,T= 1) : 533.336 MB/s [ 130209.0 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 512.888 MB/s [ 125216.8 IOPS]
Random Read 4KiB (Q= 1,T= 1) : 264.621 MB/s [ 64604.7 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 244.962 MB/s [ 59805.2 IOPS]Test : 1024 MiB [D: 0.0% (0.1/447.1 GiB)] (x5) [Interval=5 sec]
Date : 2019/09/18 17:28:33
OS : Windows 10 Professional [10.0 Build 18362] (x64)
CrystalDiskMark 6.0.2 x64 © 2007-2018 hiyohiyo
Crystal Dew World : https://crystalmark.info/
MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
KB = 1000 bytes, KiB = 1024 bytes
Sequential Read (Q= 32,T= 1) : 2783.847 MB/s
Sequential Write (Q= 32,T= 1) : 2452.787 MB/s
Random Read 4KiB (Q= 8,T= 8) : 2434.585 MB/s [ 594381.1 IOPS]
Random Write 4KiB (Q= 8,T= 8) : 2465.334 MB/s [ 601888.2 IOPS]
Random Read 4KiB (Q= 32,T= 1) : 533.914 MB/s [ 130350.1 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 502.823 MB/s [ 122759.5 IOPS]
Random Read 4KiB (Q= 1,T= 1) : 265.095 MB/s [ 64720.5 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 246.057 MB/s [ 60072.5 IOPS]Test : 1024 MiB [E: 0.0% (0.1/447.1 GiB)] (x5) [Interval=5 sec]
Date : 2019/09/18 17:30:29
OS : Windows 10 Professional [10.0 Build 18362] (x64)
Edit: Forgot the AS SSD results:
Long story short: Doesn’t seem to matter, all hail PCIe 4.0 x4 as an interface between CPU and chipset
Nothing like slow storage on a X570 chipset
and you popped for a x570 Taichi even 
Unfortunatey, I didn’t have the choice to get any other motherboard model on the entire market due to wanting AMD with Thunderbolt 3 with 2 x 100 W Power Delivery via a Gigabyte GC-Titan Ridge AIC.
@aBav.Normie-Pleb thanks! It seems it’s almost same either on cpu or chipset lanes. BTW, how about X570 taichi? I’m thinking of using this for my VFIO system.