A tale of two SSD's and one is not like the other

[A Computer Guy]

So I had these two 1TB Crucial NVMe drives I replaced and was going to put them to use in Asus Hyper M.2 card (positions #3 & #4) for some experiments but after setting them up in RAID-0 I noticed unexpected poor performance and uneven usage in Task Manager (sorry I don't have a screen shot of that) I made sure the drives were empty and trimmed before testing them.

So after that I opted to test them individually. I even swapped the slots and tested them again just to be sure it wasn't some bifurcation issue with my Ryzen 4750G in x8x4x4 mode.
In order to help identify the correct NVMe I put some files on the poor performing one so I could see it in Crucial Software and via the Asus M.2 cards LED indicator.
Turns out the better NVMe had an onboard activity indicator which was also helpful in identifying the correct drive along with the labels that indicated the firmware version.

( edit 9/14/2023: both drives are working fine but the M.2_4 slot is not running at full speed. see subsequent posts for more info )

Here is a picture with the stickers peeled back to show two different controllers.

Additional information found on TPU

• https://www.techpowerup.com/review/crucial-p1-nvme-m-2-ssd-1-tb/
• https://www.techpowerup.com/ssd-specs/crucial-p1-1-tb.d118

 

[AsRock]

PCBs been change maybe that's what is making the differences, but that different seems like one might be RMA time.

You could all so switch them around on the M.2 slots and make sure every thing is setup right.

 

[A Computer Guy]

PCBs been change maybe that's what is making the differences, but that different seems like one might be RMA time.

You could all so switch them around on the M.2 slots and make sure every thing is setup right.

So it seems I must have made a mistake in my testing. I switched the worse one to M_1 from M_4 and now it's working fine. So overall this isn't quite working as I expected where originally in M_1 I had a seagate drive expecting to work as x4 in the x8 (and it was working at full speed) and M_3 and M_4 expected to work as x4 each in the remaining x4x4 lane configuration. It's like the M_4 isn't working at full bandwidth.

I'm getting much better results from the raid-0 configuration too.

 

[DarkDreams]

So it seems I must have made a mistake in my testing. I switched the worse one to M_1 from M_4 and now it's working fine. So overall this isn't quite working as I expected where originally in M_1 I had a seagate drive expecting to work as x4 in the x8 (and it was working at full speed) and M_3 and M_4 expected to work as x4 each in the remaining x4x4 lane configuration. It's like the M_4 isn't working at full bandwidth.

In which pcie slot do you have the card? Because x8x4x4 could only work in pcie_1, in the picture it's hard to tell but looks like pcie_2. Also which motherboard are you on?

 

[A Computer Guy]

In which pcie slot do you have the card? Because x8x4x4 could only work in pcie_1, in the picture it's hard to tell but looks like pcie_2. Also which motherboard are you on?

Ryzen 4750G (Renoir APU) in an Asrock Fatal1ty X470 Gaming-ITX/ac (mITX). Since it has the iGPU I can sacrifice the only x16 slot and bifurcate to x8x4x4 using M.2_1, M.2_3, and M.2_4 to get 3 of the 4 slots useful however M.2_4 isn't getting full performance as I thought it would.

[Motherboard] Compatibility of PCIE bifurcation between Hyper M.2 series Cards and Add-On Graphic Cards | Official Support | ASUS Global

 

[DarkDreams]

Ryzen 4750G (Renoir APU) in an Asrock Fatal1ty X470 Gaming-ITX/ac (mITX). Since it has the iGPU I can sacrifice the only x16 slot and bifurcate to x8x4x4 using M.2_1, M.2_3, and M.2_4 to get 3 of the 4 slots useful however M.2_4 isn't getting full performance as I thought it would.

[Motherboard] Compatibility of PCIE bifurcation between Hyper M.2 series Cards and Add-On Graphic Cards | Official Support | ASUS Global

Yeah, then it really should work the way you set it up initially. Very strange. Only other thing I can think of is maybe bad contact somewhere (either cpu or the pcie slot) that makes you drop lanes. From the speed it got on the first test it looks like only a single lane, right?

 

[A Computer Guy]

Yeah, then it really should work the way you set it up initially. Very strange. Only other thing I can think of is maybe bad contact somewhere (either cpu or the pcie slot) that makes you drop lanes. From the speed it got on the first test it looks like only a single lane, right?

Well the speed is about 1/2 (a bit less actually) so if I had to guess I'd say at least two lanes must be working. Is there a way I can tell how many lanes an NVMe is utilizing?
I put my seagate Gen4 into M.2_4 and it's also capped at 800MB/s so something is definitely up with that last available NVMe slot.
My expectation was the x16 slot is getting all 16 lanes but perhaps the iGPU is taking some lanes?

(edit)

So I was browsing HWiNFO64 and was able to get some info. The link speed on M.2_4 is much lower despite having 4 lanes.

M.2_1 & M.2_3

M.2_4

Speed Reference Chart

PCI Express - Wikipedia

en.wikipedia.org

 

[unwind-protect]

I knew this would be about something Micron related before opening this thread...

 

[DarkDreams]

Well the speed is about 1/2 (a bit less actually) so if I had to guess I'd say at least two lanes must be working. Is there a way I can tell how many lanes an NVMe is utilizing?
I put my seagate Gen4 into M.2_4 and it's also capped at 800MB/s so something is definitely up with that last available NVMe slot.
My expectation was the x16 slot is getting all 16 lanes but perhaps the iGPU is taking some lanes?

(edit)

So I was browsing HWiNFO64 and was able to get some info. The link speed on M.2_4 is much lower despite having 4 lanes.

So it is running on 4 lanes, but at pcie 1.0. Maybe a signal integrity problem since the latency of that one is also significantly higher than the other two. Perhaps check if all the SMDs are still there compared to the other m.2 slots. Especially since there appear to be some under the m.2, might be easy to knock one off when installing the SSD.

 

[Selaya]

isnt the p1 some random qlc garbage

 

[A Computer Guy]

isnt the p1 some random qlc garbage

I didn't realize it was QLC until a few days ago but works just fine for low intensity workloads and it's a DRAM based drive too. I ran them as OS and gaming drives for a while until I needed something better.

 

[kapone32]

Best example I can think of was the ADATA 8100 series NVME drives.

 

[AsRock]

Well the speed is about 1/2 (a bit less actually) so if I had to guess I'd say at least two lanes must be working. Is there a way I can tell how many lanes an NVMe is utilizing?
I put my seagate Gen4 into M.2_4 and it's also capped at 800MB/s so something is definitely up with that last available NVMe slot.
My expectation was the x16 slot is getting all 16 lanes but perhaps the iGPU is taking some lanes?

(edit)

So I was browsing HWiNFO64 and was able to get some info. The link speed on M.2_4 is much lower despite having 4 lanes.

M.2_1 & M.2_3

View attachment 313595View attachment 313596

M.2_4

View attachment 313597

Speed Reference Chart

PCI Express - Wikipedia

en.wikipedia.org

View attachment 313599

I know Hard Disk Sentinel does, i believe it will show you even in the free version too.

 

[Selaya]

I didn't realize it was QLC until a few days ago but works just fine for low intensity workloads and it's a DRAM based drive too. I ran them as OS and gaming drives for a while until I needed something better.

are you sure it's got dram?
looks like dramless to me

 

[A Computer Guy]

are you sure it's got dram?
looks like dramless to me

I linked the specs at the bottom of the OP from TPU database.

 

[Selaya]

spec db might be wrong.
look at those photographs. all im seeing is the controller and 2x qlc nand ...

 

[A Computer Guy]

spec db might be wrong.
look at those photographs. all im seeing is the controller and 2x qlc nand ...

I did look. You can clearly see the DRAM chip.

 

[mechtech]

Not yet……….

 

[iameatingjam]

I haven't personally had that happen with an m.2. But I mean I know it does happen, companies love to shift around ssd components based on market conditions which I kind of understand, as long as performance isn't significantly impacted, but often it is.

I do however have two mx500 sata ssds, and the 3 year old one consistently does better in tests than the 6 month old one, but not to a huge degree, I think it was around 5 - 10% difference. They still work pretty well in raid 0 together though so w/e. Also it was ridiculously cheap so I don't really care.

Also I didn't even know the p1 existed. You'd think the p3 would be cheap enough already. Actually, looks like where I live, the p5 plus is cheaper and the p1 on amazon. Is it out of production or something?

 

[A Computer Guy]

I haven't personally had that happen with an m.2. But I mean I know it does happens, companies love to shift around ssd components based on market conditions which I kind of understand, as long as performance isn't significantly impacted, but often it is.

I do however have two mx500 sata ssds, and the 3 year old one consistently does better in tests than the 6 month old one, but not to a huge degree, I think it was around 5 - 10% difference. They still work pretty well in raid 0 together though so w/e. Also it was ridiculously cheap so I don't really care.

Also I didn't even know the p1 existed. You'd think the p3 would be cheap enough already. Actually, looks like where I live, the p5 plus is actually cheaper and the p1 on amazon. Is it out of production or something?

Yea I think the P1 is out of production. I picked up a couple P1's and P2's on fire sales long before the NAND prices started dropping where you could get a Samsung 980 Pro 2TB for $99. If I remember correctly they happened to be the cheaper NVMe drives for awhile.

Update. So I popped the Asus M.2 card into my Asrock B450M Pro (PCIe 3.0) w/ 3950x and x4x4x4x4 bifurcation and the card is working fine.

The Seagate 520 is in M.2_4 and working at full 8.0GT/s speed now.

I imported the Windows RAID-0 array and it's fine too. Here you can see disk 3 & 4 working at equal percentages during the test.

Interestingly the performance is down quite a bit in some tests for the Windows raid-0 array.
(4750G in x8x4x4) ---> (3950x in x4x4x4x4)

So the issue is not with the drives or the card as they are all working fine. The 4750G and/or x470-ITX motherboard seem to be the limiting factors.

 

[nomdeplume]

Luck of the draw with box they pull from on production line (NVMe) chaining them to luck of the draw system matching mobo/PCIe card.

At least it mostly works as intended now.

 

[SPDIF]

The P1 Crucial has DRAM, 1MB per 1GB. But it's cons are;

Poor direct to QLC write speed
Lower than average application performance

The Crucial P1 comes equipped with the four-channel Silicon Motion SM2263EN controller that limits it to a maximum of 2GB/s of throughput. Silicon Motion's pricier eight-channel SM2262/2262EN controllers are faster, but you'll only find those on more expensive SSDs.

​

 

[csendesmark]

Oh, nice story.
When I get a new part I always do the burn in performance tests so I can return them if something is off.

 

[Chrispy_]

Best example I can think of was the ADATA 8100 series NVME drives.

I have two different ADATA SX8200 Pros

Both are fine, but they are not using the same hardware nor do they offer the same performance. However, they do both provide ~3GB/s and decent IOPS when half full of data , so I'm getting the advertised speeds with the slower one, even if those aren't the "as reviewed" speeds.

The bait-and-switches that really irk me are when a TLC drive becomes a QLC drive. When QLC drives run out of cache, they are so much garbage that you're almost better off using spinning rust. For your casual-user laptop drive, it's fine, but when you're expecting a TLC drive to use as a write cache, the entire storage array times out and goes offline because of a manufacturer cutting corners and being deceptive about it....

 

[kapone32]

I have two different ADATA SX8200 Pros

Both are fine, but they are not using the same hardware nor do they offer the same performance. However, they do both provide ~3GB/s and decent IOPS when half full of data , so I'm getting the advertised speeds with the slower one, even if those aren't the "as reviewed" speeds.

The bait-and-switches that really irk me are when a TLC drive becomes a QLC drive. When QLC drives run out of cache, they are so much garbage that you're almost better off using spinning rust. For your casual-user laptop drive, it's fine, but when you're expecting a TLC drive to use as a write cache, the entire storage array times out and goes offline because of a manufacturer cutting corners and being deceptive about it....

The 8100 series was a totally different scenario. The 8200 Pros were always their top drives but the 8100 went from TLC to QLC with a weaker controller when the 8200 launched. At most there was a 6 month gap between the launch of the 2.