Why so few PCI-E M.2 NVMe SSDs? Where are the budget options? Where's the sweet spot?

[Billy Tallis]

My 4K randoms doubled when converting from MBR to GPT.

You must have had a mis-aligned partition when you were using MBR. It's possible to properly align partitions even with a MBR setup. GPT vs MBR doesn't matter after the OS is loaded.

 

[XavierMace]

Nope, alignment was checked before and after. In principal, I agree, it shouldn't matter. But I re-tested the results multiple times.

 

[Billy Tallis]

Nope, alignment was checked before and after. In principal, I agree, it shouldn't matter. But I re-tested the results multiple times.

How were you checking alignment?

Did you go back to an MBR configuration and reproduce the lower performance, or just re-run the benchmark in the new GPT configuration and observe similar results? If the latter, then the performance boost could have been a lingering effect of wiping the drive, while the former would strongly suggest that there was an alignment problem.

 

[XavierMace]

Alignment was checked via msinfo32 and drive wasn't wiped, it was converted in place via gptgen and diskpart.

 

[Topweasel]

The problem is human perception is not terribly reliable and too easily influenced by expectations. I'd swear my car makes odd noises but every car guy I know who's listened to it and rode in it says they hear nothing odd. Ask anybody working an IT Help Desk how many calls they get for "my computer is slow" but they see nothing abnormal with it. As Benjamin Franklin wrote, a watched pot never boils.

The problem is that "feels" is what made the SSD the go to OS drive for the last 6 years. Raptors and 3-4TB drives where nearly as fast. Faster if you raided them and while you easily if you were slightly technical point to the random access time to point out while it feels faster, but that isn't what made SSD's explode. Feeling did.

Fast forward to NVME drives. Now you have a drive that can be upwards of 10x faster than some people's previous SSD's. But the feeling that made SSD's an obvious system upgrade is gone. Now you are running into on the home user level is a programing issue. Programs are installing faster because of how the installation process works. Games don't load faster because of the loading process works, even windows might actually load slightly quicker (but 10 seconds vs. 12 is almost nothing) but sometimes at the cost of an actually slower boot because the process to boot to NVME might take an extra moment or two before loading windows. For home users the transfer speed is always going to be limited by the devices you are copying to and from which unless you have a server with NVME array, a TR or X99+ system you are unlikely to have more than one NVME drive to copy to.

Now when Zen2 TR comes out I will build a system for it and I will CPU raid 0 an NVME 2 drive array so I can have fun knowing my OS drive is almost silly fast. But I am not hoping for tangible results. Which frankly is important. Whether you are talking about FPS in games, Time to finish in a professional task, or load time for any program, you typically get new hardware for tangible results. Not for slight statistical advantages or numbers in a benchmark that don't actually get replicated in any general users use case. Feelings are anecdotal, but in the end it's the only measurement that matters.

 

[whm1974]

My 4K randoms doubled when converting from MBR to GPT.

A perfectly good reason to switch over to GPT from MBR.

 

[Micrornd]

Nope, alignment was checked before and after. In principal, I agree, it shouldn't matter. But I re-tested the results multiple times.

I agree, doesn't reall make sense according to accepted lore.
I'd like to try and duplicate this, to see if I can improve my 4ks.
What testing software did you use, what were the final results, was this Nvme or just an M2 disk, drivers, etc.

 

[XavierMace]

Samsung 960 Pro (512Gb), Windows 10 Pro, both Samsung and Microsoft NVMe driver. Asrock X370 Taichi with current BIOS. Drive manually trimmed prior to testing. Latest version of CrystalDiskMark. Win10 was a fresh install (new drive) but using MBR. Drive was manually converted to GPT without reformatting. My results are mostly normal now, so personally not really looking to spend any more time on it.

Let me be clear, the reason I went down this road is I was getting extremely poor 4K performance, far below anyone else's. So this was done as a fix not a performance boost per say.

 

[Micrornd]

I see, it would appear it was an isolated problem with your system and not an endemic problem in Windows.
My own testing with MBR, 960 pro (512gb) and the Samsung 2.2 driver in all tests meets or exceeds the results posted by Anandtech and Ars, so I don't see a need to pursue this further either