USB on M1 Macs isn't actually 10Gb/s? (Also definitely not USB4)

[Jzjz2021]

Please see attached.
View attachment 1718679 View attachment 1718680

Numbers look identical, I can't find any differences that would cause the lower speeds.
So that narrows it down to the nvme drive itself or is it a Mac mini vs MacBook Pro thing or is the OS causing general speed issues and flakiness mentioned is this thread.
Next step is to see what the Big Sur 11.2 update changes?

 

[theotherphil]

I just came across this thread and I apologise for not reading it all. I did however read an article a few months back which took some searching to track down again.

The gist of it is that M1 Macs have 2x USB4 controllers...one for each port, so each will have a theoretical 2,800MB/s bandwidth. The authors tested this theory and placed 4x NVMe SSD’s in a RAID 0 config and measured the performance. They achieved 3,500MB/s which validated the theory.

Thunderbolt on the M1 Mac mini – When 2 Actually Does Equal 4

I have read a lot of complaints about the new M1 based Macs only having two Thunderbolt ports. Having only two ports doesn't seem like enough for professional use. Both my 2019 16 inch MacBook Pro and my 2018 Mac mini have four Thunderbolt ports, and I can't imagine using a Mac with only two...

eshop.macsales.com

Edit: Just finished reading the thread so whilst this is not new information, it adds to the story.

 

[BeatCrazy]

Numbers look identical, I can't find any differences that would cause the lower speeds.
So that narrows it down to the nvme drive itself or is it a Mac mini vs MacBook Pro thing or is the OS causing general speed issues and flakiness mentioned is this thread.
Next step is to see what the Big Sur 11.2 update changes?

So I went and picked up a 1TB Samsung 970 EVO Plus, and put it in the same ACASIS USB4 enclosure.

Read speeds are much better. What is going on with write speeds?

 

[Jzjz2021]

So I went and picked up a 1TB Samsung 970 EVO Plus, and put it in the same ACASIS USB4 enclosure.

Read speeds are much better. What is going on with write speeds?
View attachment 1718814

Those read speed are looking good?
They should be the max of what Thunderbolt can offer according to what orpheus1120 said.
The write speeds are perplexing. Looking at all the numbers in this whole thread, they are all over the place and mostly on the slow end. It must be the OS then or the benchmark, we will see when 11.2 is released.
Also maybe try diskmark and sensei apps to compare write speeds.
At least the read speeds numbers are looking proper ? compared to your Adata nvme stick. I was just about to order the evo+ also to test, so you saved me the hassle. Thanks.?
I think the enclosure is a keeper, it seems like a solid piece of hardware. I hope I pointed you in the right direction at least. We will keep trying. I have to keep remembering these speeds are 10x-20x of what current spinning hard drives offer, so the results are still good. We are just trying to squeeze every bit of performance out of these drives and enclosures, also doing it on brand new untested hardware ie. M1 macs. We are sailing in uncharted waters.

 

[BeatCrazy]

Those read speed are looking good?
They should be the max of what Thunderbolt can offer according to what orpheus1120 said.
The write speeds are perplexing. Looking at all the numbers in this whole thread, they are all over the place and mostly on the slow end. It must be the OS then or the benchmark, we will see when 11.2 is released.
Also maybe try diskmark and sensei apps to compare write speeds.
At least the read speeds numbers are looking proper ? compared to your Adata nvme stick. I was just about to order the evo+ also to test, so you saved me the hassle. Thanks.?
I think the enclosure is a keeper, it seems like a solid piece of hardware. I hope I pointed you in the right direction at least. We will keep trying. I have to keep remembering these speeds are 10x-20x of what current spinning hard drives offer, so the results are still good. We are just trying to squeeze every bit of performance out of these drives and enclosures, also doing it on brand new untested hardware ie. M1 macs. We are sailing in uncharted waters.

Well, here's some crazy stuff. Remember how my 1TB ADATA XPG SX8200 performed so poorly in this new ACASIS enclosure? Well oddly enough, I have a Windows PC with a spare non-boot NVMe slot, so I measured both the 970 EVO Plus, as well as the ADATA.

970 EVO Plus:

Here's the ADATA in the same machine:

So what is the problem with these M1 machines?
RTL9210 enclosure behaves differently with different drives? JHL7440 enclosure behaves way different with different drives? Each of those drives behave as expected when connected to NVMe slots in a W10 PC.

 

[Jzjz2021]

Well, here's some crazy stuff. Remember how my 1TB ADATA XPG SX8200 performed so poorly in this new ACASIS enclosure? Well oddly enough, I have a Windows PC with a spare non-boot NVMe slot, so I measured both the 970 EVO Plus, as well as the ADATA.

970 EVO Plus:
View attachment 1718877
Here's the ADATA in the same machine:
View attachment 1718878


So what is the problem with these M1 machines?
RTL9210 enclosure behaves differently with different drives? JHL7440 enclosure behaves way different with different drives? Each of those drives behave as expected when connected to NVMe slots in a W10 PC.

You gotta love technology for quirks like this. ?
I was just doing the same thing, testing in a win10 machine. The nvme drives and a few high end flash drives I tested all lose there speed, especially write speeds when connected to a M1 Mac. In Windows they achieve there rated speeds. It also seems like the faster they go, especially above 1000MB/s, the more they lose there top end speeds. ie..Sandisk extreme Flash drive rated at 500MB/s gets 500 in windows, but only 300 connected to M1, a Samsung 960 pro nvme rated at 3500, gets around that in Windows but only 1500 connected to the Mac.
Losing 200MB/s on the flash drive is ok but losing 2000MB/s on the nvme hurts. Hopefully it is just a Big Sur OS bug and it will get fixed.
I guess we are just figuring out what this thread has been saying all along, USB is not working correctly on M1 Mac’s. The devices work which is good but they definitely do not get there rated speeds.

 

[mi7chy]

Probably insufficient PCIe lanes.

 

[joevt]

You gotta love technology for quirks like this. ?
I was just doing the same thing, testing in a win10 machine. The nvme drives and a few high end flash drives I tested all lose there speed, especially write speeds when connected to a M1 Mac. In Windows they achieve there rated speeds. It also seems like the faster they go, especially above 1000MB/s, the more they lose there top end speeds. ie..Sandisk extreme Flash drive rated at 500MB/s gets 500 in windows, but only 300 connected to M1, a Samsung 960 pro nvme rated at 3500, gets around that in Windows but only 1500 connected to the Mac.
Losing 200MB/s on the flash drive is ok but losing 2000MB/s on the nvme hurts. Hopefully it is just a Big Sur OS bug and it will get fixed.
I guess we are just figuring out what this thread has been saying all along, USB is not working correctly on M1 Mac’s. The devices work which is good but they definitely do not get there rated speeds.

PCIe gen 4 NVMe drives usually give saturated Thunderbolt numbers. I know using gen 4 in a gen 3 slot is a bit of a waste but it's the easiest way to get near 3000 MB/s read/write from a Thunderbolt 3 device to test the functioning of a Thunderbolt host.

Use AmorphousDiskMark to get the highest numbers.

Does APFS vs HFS+ make a difference? I always use HFS+.

Probably insufficient PCIe lanes.

Anything over 2000 MB/s means four lanes of PCIe 3.0 is working.

 

[BeatCrazy]

PCIe gen 4 NVMe drives usually give saturated Thunderbolt numbers. I know using gen 4 in a gen 3 slot is a bit of a waste but it's the easiest way to get near 3000 MB/s read/write from a Thunderbolt 3 device to test the functioning of a Thunderbolt host.

Use AmorphousDiskMark to get the highest numbers.

Does APFS vs HFS+ make a difference? I always use HFS+.


Anything over 2000 MB/s means four lanes of PCIe 3.0 is working.

I tried testing by formatting as APFS, HFS+, exFAT... no noticeable difference.

 

[orpheus1120]

If you have a thunderbolt 3 hub like the CalDigit T3 Plus, and connect to their T3 port, results should be better.

 

[BeatCrazy]

If you have a thunderbolt 3 hub like the CalDigit T3 Plus, and connect to their T3 port, results should be better.

I don't have one of those, but have been considering the OWC TB4 hub. So you're saying that hanging off a hub will have better speed than directly attaching to the M1?

 

[orpheus1120]

I think so. I suggest you read through the entire thread.

 

[Deccr]

The higher speeds when using a TB hub like the CalDigit TS3+, is for those using USB-based external SSDs, not TB external SSDs.

I have both the TS3+ and OWC TB4 dock. I’ve found external TB SSD drives have slightly faster speeds when connected directly to my M1 Mac mini, in comparison to being used through either of those docks.

 

[Slartibart]

„ Like previous Macs, the new M1 Macs can handle Thunderbolt 3 and also contain a USB 3.1 host controller for Gen 2 data transfers at 10 Gbit / s. On the motherboard, however, there are JHL8040R components that Intel specifies as "Thunderbolt 4 Retimer". Such retimers improve the signal quality, but do not contain any protocol logic: They rely on Thunderbolt logic in the SoC processor (or elsewhere). Thunderbolt 3 logic was included in the tenth generation of Core i processors, the newer Thunderbolt 4 in the recently introduced eleventh Core i generation. Apple also has Thunderbolt 3 logic in the M1 SoC.

Thunderbolt 3 and Thunderbolt 4 have the same user data rate of 40 GBit / s, so Thunderbolt 4 would not offer any advantages in this regard - and the same retimer chips work for both.

The fact that there is no new speed level is due to the fact that the further development of Thunderbolt has been discontinued as a separate standard: Intel has passed the specification to the USB standardization committee USB-IF, which has standardized USB 4 based on it - with the one known from Thunderbolt, but for USB new maximum data rate of 40 GBit / s. The USB-IF has changed certain technical subtleties, which is why pure Thunderbolt 3 controllers are not fully compatible with USB 4. ...

The attentive reader will be forced to ask a question mark at this point: Why does Apple advertise the USB-C sockets of the M1-Macs with USB 4? ....
"Thunderbolt 3 with better screen support than provided in the standard and USB with 3.1 speed" would be more precise - but it is much bulkier and doesn't sound so modern. ...“

Much more details and the why and how here (in German).

 

[theSeb]

This video has an ok and simple to understand explanation of what is going on with the speeds of the ports on M1 Macs.

 

[legato01]

So I got the ACASIS USB4.0 Mobile M.2 Nvme Enclosure today and I installed my 1TB ADATA XPG SX8200.

Speeds are much less than I was expecting. Write speeds often dip below 1000MB/s.

I tried with both a Belkin TB3, and the included ACASIS TB3 cable, no difference.

Same.

ACASIS USB4.0 arrived and was excited to test it out.

A little disappointed as I was expecting twice the performance.

 

[BeatCrazy]

Same.

ACASIS USB4.0 arrived and was excited to test it out.

A little disappointed as I was expecting twice the performance.

Yep. Maybe 11.2 will help? You probably saw, I doubled my read speeds when using the Samsung 970 EVO Plus. But both the Samsung and the ADATA performed essentially the same when put in a normal PCIe NVMe slot on a motherboard. So it's a problem with either macOS, or the enclosure.

 

[joevt]

Yep. Maybe 11.2 will help? You probably saw, I doubled my read speeds when using the Samsung 970 EVO Plus. But both the Samsung and the ADATA performed essentially the same when put in a normal PCIe NVMe slot on a motherboard. So it's a problem with either macOS, or the enclosure.

You didn't rule out a problem with the NVMe by testing it with an Intel Mac or Windows or a different enclosure or a different NVMe.

 

[BeatCrazy]

You didn't rule out a problem with the NVMe by testing it with an Intel Mac or Windows or a different enclosure or a different NVMe.

I don't recall if I mentioned, but I did test with my 2019 MacBook Pro 13" (Four TB3 ports) and got the same results, with both SSDs. I couldn't get my Windows PC with TB3 to recognize the TB3 interface of the ACASIS enclosure. I only have one Windows PC with TB3.

 

[joevt]

I don't recall if I mentioned, but I did test with my 2019 MacBook Pro 13" (Four TB3 ports) and got the same results, with both SSDs. I couldn't get my Windows PC with TB3 to recognize the TB3 interface of the ACASIS enclosure. I only have one Windows PC with TB3.

Ok, so it's not just M1 Macs but all Macs running macOS. Boot Camp on the MacBook Pro would determine if the problem is in macOS. A faster NVMe (like a PCIe 4.0 NVMe) known to work at full speed in a Thunderbolt enclosure can determine if the problem is with the enclosure. I believe the problem is a combination of the NVMe and Thunderbolt connection (because we know the NVMe is faster when connected directly to the motherboard and that other NVMe's have no problem in a Thunderbolt enclosure).

 

[BeatCrazy]

Ok, so it's not just M1 Macs but all Macs running macOS. Boot Camp on the MacBook Pro would determine if the problem is in macOS. A faster NVMe (like a PCIe 4.0 NVMe) known to work at full speed in a Thunderbolt enclosure can determine if the problem is with the enclosure. I believe the problem is a combination of the NVMe and Thunderbolt connection (because we know the NVMe is faster when connected directly to the motherboard and that other NVMe's have no problem in a Thunderbolt enclosure).

My 2019 MacBook Pro is company-issued, so I'm not comfortable installing BootCamp on it

I almost bought a Samsung 980 Pro 500GB yesterday when Amazon had them for $112, but that still felt a bit rich for my blood. If I had a way to use PCIe 4.0 as a boot drive in one of my machines, I probably could have justified it, and repurposed it after testing.

I'll keep thinking of ways to test. But since we know my drives are "good", and there is a substantial difference in Read speeds depending on which drive I put in the ACASIS enclosure w/JHL7440, I'm hesitant to blame the enclosure. I'd say Read is working close to expected. It's just Write that I think the M1 + TB3 has a problem with.

 

[ca$hman]

Finally at 10Gbps!

They were all tested with the same Sabrent 1TB SB-ROCKET-1TB NVME M.2 Drive (Non-Q)

View attachment 1683814

Started off by testing from the USB port first and the the Thunderbolt..

ORICO Thunderbolt 3 TCM2T3-G40-BK-BP 40Gbps, only tried with TB and it connected at 40 giving the best results although most expensive at £99.99!

View attachment 1683756

What SSD did you used in the ORICO to get these write speeds? I am using a samsung 970 EVO Plus 1TB but can't seem to get more than 850Mbps write speed on my MBP M1. On my MBP Intel I do get a 1.250Mbps speed which keeps my brains busy wondering how to get higher speed on the M1.

 

[adonis3k]

What SSD did you used in the ORICO to get these write speeds? I am using a samsung 970 EVO Plus 1TB but can't seem to get more than 850Mbps write speed on my MBP M1. On my MBP Intel I do get a 1.250Mbps speed which keeps my brains busy wondering how to get higher speed on the M1.

Are you using the Orico TB3 one? I have a Sabrent 1TB Rocket (not the Q version)

 

[ca$hman]

Are you using the Orico TB3 one? I have a Sabrent 1TB Rocket (not the Q version)

Yes, I have the 40Gbps TB3 one, TOM2T3-G40 to be precise. Very disappointed about the write speed.

 

[adonis3k]

Yes, I have the 40Gbps TB3 one, TOM2T3-G40 to be precise. Very disappointed about the write speed.

Are you using the supplied cable in the box?