BTW: If the reinstall runs fast, it probably didn't relink. You might have to first remove old kernel and then reinstall.
Looks like it's worth it. The PVSCSI works great for emulation to get up and running, but performance is terrible compared to running in vmware...
With virtio:
BS=512 ReadWriteRatio=0 IODEPTH=4 RW= write: IOPS=90.8k, BW=44.4MiB/s (46.5MB/s)(26.0GiB/600002msec); 0 zone resets
BS=512 ReadWriteRatio=100 IODEPTH=4 RW= read: IOPS=106k, BW=51.9MiB/s (54.4MB/s)(30.4GiB/600002msec)
BS=16384 ReadWriteRatio=0 IODEPTH=4 RW= write: IOPS=66.3k, BW=1037MiB/s (1087MB/s)(607GiB/600001msec); 0 zone resets
BS=16384 ReadWriteRatio=100 IODEPTH=4 RW= read: IOPS=79.8k, BW=1246MiB/s (1307MB/s)(730GiB/600001msec)
BS=65536 ReadWriteRatio=0 IODEPTH=4 RW= write: IOPS=23.5k, BW=1471MiB/s (1543MB/s)(862GiB/600004msec); 0 zone resets
BS=65536 ReadWriteRatio=100 IODEPTH=4 RW= read: IOPS=35.4k, BW=2214MiB/s (2321MB/s)(1297GiB/600002msec)
BS=512 ReadWriteRatio=0 IODEPTH=32 RW= write: IOPS=94.6k, BW=46.2MiB/s (48.4MB/s)(27.1GiB/600004msec); 0 zone resets
BS=512 ReadWriteRatio=100 IODEPTH=32 RW= read: IOPS=107k, BW=52.3MiB/s (54.8MB/s)(30.6GiB/600004msec)
BS=16384 ReadWriteRatio=0 IODEPTH=32 RW= write: IOPS=67.2k, BW=1050MiB/s (1101MB/s)(615GiB/600004msec); 0 zone resets
BS=16384 ReadWriteRatio=100 IODEPTH=32 RW= read: IOPS=80.8k, BW=1262MiB/s (1323MB/s)(739GiB/600004msec)
BS=65536 ReadWriteRatio=0 IODEPTH=32 RW= write: IOPS=23.2k, BW=1447MiB/s (1517MB/s)(848GiB/600014msec); 0 zone resets
BS=65536 ReadWriteRatio=100 IODEPTH=32 RW= read: IOPS=36.9k, BW=2304MiB/s (2415MB/s)(1350GiB/600009msec)
With vmware driver:
BS=512 ReadWriteRatio=0 IODEPTH=4 RW= write: IOPS=19.0k, BW=9493KiB/s (9721kB/s)(5563MiB/600001msec); 0 zone resets
BS=512 ReadWriteRatio=100 IODEPTH=4 RW= read: IOPS=22.4k, BW=11.0MiB/s (11.5MB/s)(6573MiB/600001msec)
BS=16384 ReadWriteRatio=0 IODEPTH=4 RW= write: IOPS=17.5k, BW=274MiB/s (287MB/s)(160GiB/600003msec); 0 zone resets
BS=16384 ReadWriteRatio=100 IODEPTH=4 RW= read: IOPS=20.8k, BW=325MiB/s (341MB/s)(190GiB/600001msec)
BS=65536 ReadWriteRatio=0 IODEPTH=4 RW= write: IOPS=15.2k, BW=951MiB/s (997MB/s)(557GiB/600002msec); 0 zone resets
BS=65536 ReadWriteRatio=100 IODEPTH=4 RW= read: IOPS=17.8k, BW=1111MiB/s (1165MB/s)(651GiB/600001msec)
BS=512 ReadWriteRatio=0 IODEPTH=32 RW= write: IOPS=18.9k, BW=9452KiB/s (9679kB/s)(5538MiB/600002msec); 0 zone resets
BS=512 ReadWriteRatio=100 IODEPTH=32 RW= read: IOPS=22.6k, BW=11.1MiB/s (11.6MB/s)(6635MiB/600002msec)
BS=16384 ReadWriteRatio=0 IODEPTH=32 RW= write: IOPS=17.6k, BW=275MiB/s (289MB/s)(161GiB/600001msec); 0 zone resets
BS=16384 ReadWriteRatio=100 IODEPTH=32 RW= read: IOPS=20.9k, BW=327MiB/s (343MB/s)(192GiB/600003msec)
BS=65536 ReadWriteRatio=0 IODEPTH=32 RW= write: IOPS=15.8k, BW=989MiB/s (1037MB/s)(579GiB/600003msec); 0 zone resets
BS=65536 ReadWriteRatio=100 IODEPTH=32 RW= read: IOPS=19.2k, BW=1203MiB/s (1262MB/s)(705GiB/600003msec)
I think it's worth it if there is at least moderate level of disk activity...
The above tests were on an R730 recently obtained for home lab from ebay and a old Intel NVME pcie card running in a LVM-thin.
I didn't bother monitoring cpu utilization during test as the IOPs are poor enough that it would likely be moot in most cases... (ie: want higher IOPs even if more cpu/iop).
I do get much better performance with vmware and the driver under vmware... so it's more an emulation of that hardware than the driver.