BIOS 3703! Maximus VIII Hero SSD has been crapped out!

[Retry]

MotherBoard: Maximus VIII Hero
OS: Windows 10 Pro X64
SSD: Samsung 850Pro 512GB(Firmware: Up to date)
Windows 10 Security and updates: Fully patched and up to date
IRST: v15.9.0.1015 Up to date
Everything is up to date!


Random(IOPS) Read: about 92000/ Write: about 87000


Random(IOPS) Read: about 75000/ Write: about 53000


Random(IOPS) Read: about 51000/ Write: about 37000

Samsung Magician v5.2 with Bios 3703



CrystalDiskMark v6.0.0 with Bios 3703



Hey ASUS? Look at the performance.
Are you kidding me?
What did you make?

[Retry]

Mr.Wolf wrote:
Sorry, I'm not going to install magician, I don't like like and don't use it.
I don't have any SSD performance problem before or after the new bios and updates.

I have an old win7 as-ssd screenshot (think it's from 2016) in this same HW to compare (or I could read any ssd review for my model).

What's your cpu?

6700k and 32GB 8 x 4 memory.

[Crashmoto]

Bios 3703 the speed ssd fell

[Korth]

These performance scores are all taken over a sequential one-way time period. They could indicate the SATA SSD is declining, not the system running it. They could indicate some other system component (CPU, RAM, operating system, etc) is declining or has been changed over time, not necessarily the BIOS firmware.

You could rollback to older BIOS versions to measure performances again. Determine if it's the BIOS or the drive which has slowed things down. With multiple benchmark softwares. On the same hardware, on the same operating system. CrystalDiskMark is good but it's still a synthetic test, numerical discrepancies on one synthetic test could indicate the testing software itself has changed, not the hardware it's measuring.

[pokuly]

Korth wrote:
These performance scores are all taken over a sequential one-way time period. They could indicate the SATA SSD is declining, not the system running it. They could indicate some other system component (CPU, RAM, operating system, etc) is declining or has been changed over time, not necessarily the BIOS firmware.

You could rollback to older BIOS versions to measure performances again. Determine if it's the BIOS or the drive which has slowed things down. With multiple benchmark softwares. On the same hardware, on the same operating system. CrystalDiskMark is good but it's still a synthetic test, numerical discrepancies on one synthetic test could indicate the testing software itself has changed, not the hardware it's measuring.

Easiest comparison is to disable the Spectre patch in OS with the InSpectre tool or over registry and bench before/after reboot.

[Retry]

Korth wrote:
These performance scores are all taken over a sequential one-way time period. They could indicate the SATA SSD is declining, not the system running it. They could indicate some other system component (CPU, RAM, operating system, etc) is declining or has been changed over time, not necessarily the BIOS firmware.

You could rollback to older BIOS versions to measure performances again. Determine if it's the BIOS or the drive which has slowed things down. With multiple benchmark softwares. On the same hardware, on the same operating system. CrystalDiskMark is good but it's still a synthetic test, numerical discrepancies on one synthetic test could indicate the testing software itself has changed, not the hardware it's measuring.

I rolled back to older BIOS as I said. As a result of that, the speed has been restored. That's why I think I have an issue with BIOS. I've tested with AS Benchmark, Magician, Crystaldiskmark. I'll test my SSD with other tools later. And the users who from the posts that I linked have same issue with their own BIOS.

[Mr_Wolf]

It could be some incompatibility with your SSD from the new bios, from intel microcode or from spectre patch.
If you still have the 3703 bios flashed, you could try to disable spectre patch in windows like pokuly said.
You can also load the new microcode using windows catalog update and having the old bios, meaning if problem is the new bios, but not the new MC, you can be protected against spectre with the old bios.

It doesn't affect my SSD, but I have a different model.

[Kincaid]

The microcode update as part of BIOS 3703 or Windows KB4090007 definitely hurts IOPS, period. If you want to protect against Spectre then that's the price we're all going to pay, unfortunately. It's not ASUS' fault.

ASUS Maximus VIII Extreme [BIOS 3504] & Samsung EVO 850 1TB without KB4090007:

Sequential (MB/s) Read/Write:

543/ 510

Random (IOPS) Read/Write:

80,810 / 69,335

ASUS Maximus VIII Extreme [BIOS 3504] & Samsung EVO 850 1TB with KB4090007:

Sequential (MB/s) Read/Write:

542 / 509

Random (IOPS) Read/Write:

52,001 / 45,898

If I switch back and forth with InSpectre as I have done numerous times then the IOPS will predictably go up and down.

[Mr_Wolf]

I can replicate your results in magician, but maybe a big part of the problem is from samsung magician bench.

Please open task manager and show all logical cores while you run the magician test.
Here in my 6600K@4.5GHz magician makes 100% load in only 1 core in some parts of the test, probably when doing random R/W with high queue deep.
When I run same test in AS SSD it makes like 30% load in all my 4 cores.
I can't replicate the low IOPs in AS SSD.

It's not normal 100% load in only 1 core to bench SSD IOs, it doesn't reflect real life.
It does 100% load in only 1 core with or without spectre patch enable.
We knew spectre has some cpu hit, maybe adding that to already 1 core at 100% lowers the bench.

See if you can replicate this.

-EDIT-
My personal conclusions from all the tests in my system:
1. We already knew spectre/meltdown patchs reduces cpu speed. Intel says it gets 6% to 10% slower in recent cpus (conservative intel values) and more in older cpus.
2. Magician bench is a bad test because it runs single core and doesn't reflect real life. Here it does 100% cpu load in only 1 core, before and after spectre patch, while all other cores are idle.
3. There is no real world SSD performance loss after the patchs for my tested SSDs, AS SSD and other benchs show that, problem is magician bench is not multithread.
4. SATA protocol has too much CPU overhead, we need something better.

[aceisme]

Updated from BIOS3504 to 3703 and I see no major SSD and M.2 benchmark differences between the two BIOS versions. That includes Samsung Magician. Sadly, I can't say the same about high load multithread benchmarks. Almost 10 precent performance hit.