Thunderbolt 4 & NVMe M.2 External Storage: Read/Write Speeds?

[Ploki]

Lexar NM790 4TB empty vs WD SN850X 4TB (50% full) in ASM2464.

I thought that DRAM vs DRAM-less would show in benchmarks, but it doesn't really.
I guess the limiting factor is the data port itself already.

edit: deleted benchmarks because something is off

 

[Melbourne Park]

Here are my AmorphousDiskMark results from my newly delivered Maiwo K1717 on my Mac studio M1 Max base model. I've fitted a 2TB WD_Black SN770 into the K1717.

Temperatures reported by iStat Menus never went over 36 degrees C whilst testing.

For comparison my previous OWC Envoy Express enclosure averaged 1600/1200 MB/s Read/Write speeds.

View attachment 2335274View attachment 2335275View attachment 2335276

I don't see the point in posting the internal speeds of the external drive when the T-3 bottlenecks the drive to a best around 2,000 MB/s.

 

[steve123]

I don't see the point in posting the internal speeds of the external drive when the T-3 bottlenecks the drive to a best around 2,000 MB/s.

I do not understand? What are you referring to when you say "T-3 bottlenecks"?

Those benchmarks appear to illustrate the performance of an externally connected TB drive where the bandwidth of file system read and write operations far exceeds 2000 MB/s?

 

[Melbourne Park]

I do not understand? What are you referring to when you say "T-3 bottlenecks"?

Those benchmarks appear to illustrate the performance of an externally connected TB drive where the bandwidth of file system read and write operations far exceeds 2000 MB/s?

Because I asked the maker of my internal RAID card, Highpoint, what the speed would be via Thunderbolt 3-4. They said the bottleneck is Thunderbolt, and their tests show around 2,000 due to the Thunderbolt bottleneck. Their tests though are not Black Magic etc speed performance checks, they examine real usage tests. My card runs with its not fast (I am just combining their capacity) at a bit over 4,000. Faster speeds with half the capacity are available. But the company says expect a 2,000 MB/s real world Thunderbolt bottleneck. They would understand real world Thunderbolt I am presuming. The bottleneck is not the PCI slot, its not the card either.

As far as non RAID and Thunderbolt 5, Sabrent have tested their prototype external T-5 drive and they are seeing 6,000 MB/s reads and not sure about writes. They are also saying around 4,000 on T3-4 connections. Real world? Well following on from Highpoint, I guess real world for T-5 would be 4,000 ie double what they say T-3 is. Not really surprising. Of course T-5 will not be cheap for some time ...

I'm not an expert, but I did ask a company who knows about this kind of stuff. And speed tests are useful, but real world is what its about if one actually needs the working performance.

 

[Chancha]

Because I asked the maker of my internal RAID card, Highpoint, what the speed would be via Thunderbolt 3-4. They said the bottleneck is Thunderbolt, and their tests show around 2,000 due to the Thunderbolt bottleneck. Their tests though are not Black Magic etc speed performance checks, they examine real usage tests. My card runs with its not fast (I am just combining their capacity) at a bit over 4,000. Faster speeds with half the capacity are available. But the company says expect a 2,000 MB/s real world Thunderbolt bottleneck. They would understand real world Thunderbolt I am presuming. The bottleneck is not the PCI slot, its not the card either.

As far as non RAID and Thunderbolt 5, Sabrent have tested their prototype external T-5 drive and they are seeing 6,000 MB/s reads and not sure about writes. They are also saying around 4,000 on T3-4 connections. Real world? Well following on from Highpoint, I guess real world for T-5 would be 4,000 ie double what they say T-3 is. Not really surprising. Of course T-5 will not be cheap for some time ...

I'm not an expert, but I did ask a company who knows about this kind of stuff. And speed tests are useful, but real world is what its about if one actually needs the working performance.

The actual "Thunderbolt 3/4 bottleneck" across *all available* devices out there, is as far as I know somewhere around 2800MB/s real world large transfer, and 3000MB/s short burst benchmarking. I am literally getting that speed from a JHL7440 chip equipped NVMe enclosure.

That specific maker that you have consulted with probably are only concerned with their own product and thus answered you within that context.

Then, the post that you quoted, the guy was benchmarking with a USB4 NVMe enclosure using an ASMedia chip. This means it does not use Thunderbolt at all. It is already widely known and tested for that chip to easily exceed 3000MB/s on Macs, or more if you get an optimized combo of specific SSD model and other factors.

 

[Ploki]

USB4 bottleneck is about 3400mb/s.
Thunderbolt3 2800mb/s max
(I have both connected all the time)

This is direct to computer connection.
Speeds slow down via daisy chaining.

 

[Sharky II]

Bought this recently and put a WD Black SN850X 2GB into it: https://www.amazon.co.uk/dp/B0CLV3D3H6?ref=ppx_yo2ov_dt_b_fed_asin_title

It looks/feels premium and I'm getting good speeds on a Mac Studio M1 Ultra (see screenshot). However, it gets hot and the fan ramps up and is noisy. I mainly work with audio and I couldn't use this while working on audio in the same room - it's louder than my entire Mac Studio which is a bit silly. It should be the type of flat fan similar to what's in a MacBook rather than a useless normal 2cm fan crammed in there. The ramping up and down of the fan is also annoying. It's just sat there after a file transfer with the fan ramping up and down, not even in use. I'm pretty sure the fan isn't helping things much, which seems like a waste. If it's going to make noise, at least make it do something.

Thankfully, I bought this for video editing, where I don't really care about noise as much. So I'll leave it unplugged unless I'm working on video. And no, there is no way to turn the fan off.

Further, the drive sits a slight angle so the included thermal pad doesn't really make much more than 30% contact with the chassis. I eventually added some more of my own that I had lying around, no idea if it's helping or not.

While I'm happy with the speed, I wouldn't buy another unless it was for video or photo editing.

This whole situation is so frustrating. I'd be happy with USB 3.2 but they all seem to have dropouts and disconnections. Next time I'm probably just going for an off the shelf Samsung T9 or whatever.

So difficult to find fast, reliable and quiet external storage.

 

[Sharky II]

Hi folks,

Bought an Akasis TBU405Pro M1 as it has a switch on the fan. It has a 1TB WD Black 850 (non X) inside. Unfortunately I'm only getting ~1300 write speed, with 2800 read. Not as fast as the UGREEN above, but can be silent. The fan is noisy and whiney and even puts out a high pitch noise when the drive is being used! Silent when the fan is off.

Anyway, any ideas on why the write speed is so slow?

The drive is has 35-40% free space.

 

[brandair]

I've had two Acasis TBU405 (without fan) running in RAID 0, each with a WD Black SN850X, for the past year, and had very rarely issues with high temps (avg. 55 C). My only issue was slow write speeds. Usually read/write speeds are more or less equal with comparable hardware - but mine are always 20-25% slower.

Recently, the TBU405's were replaced with the new and much larger NVME U.2 Thunderbolt drives (have their own power supply) from Acasis as here it's cheaper and more reliable to buy 4TB U.2 NVME's - in this case Samsung PM9A3. So, new enclosure and new disk, but again same problem - slower writes. Exchanging various TB 3 and 4 cables did not help, neither the Mac Studio trick where you plug a TB3/4 dock or a TB3/4 to DisplayPort cable (connected to a monitor of course) into the second TB back port to get better write speeds.

Both Acasis cases use the same Titan Ridge TB3 chipset, so maybe Acasis is responsible for the slow write speeds, and/or Apple's M1/M2 TB architecture. I don't know at this point, but the issue is a little annoying when investing for max. speed. What makes me wonder are a few user reports claiming to get equal read/write speeds with the same setup (Mac Studio M1, Acasis TB3 NVME & WD SN850X).

As to the loud fan, the Chinese Acasis NVME U.2 TB3 enclosures come with a constant fan, which unfortunately is so loud that you could not believe a specialized company would manufacture something like that. In the end, I carefully unplugged the fan from the controller board and was greeted with silence. U.2 disk temps only rose by 2-3 Celsius degrees (avg. 45-55 C) when monitoring them for a few days, so the fan is actually rather useless anyway. With the NVME blade cases, however, you would need to ensure that sufficient thermal paste is applied before disconnecting the fan.

Hi folks,

Bought an Akasis TBU405Pro M1 as it has a switch on the fan. It has a 1TB WD Black 850 (non X) inside. Unfortunately I'm only getting ~1300 write speed, with 2800 read. Not as fast as the UGREEN above, but can be silent. The fan is noisy and whiney and even puts out a high pitch noise when the drive is being used! Silent when the fan is off.

Anyway, any ideas on why the write speed is so slow?

The drive is has 35-40% free space.

 

[Gr1f]

Some interesting results here:
I had planned to update the 2 X Acasis TB3 enclosures to TB4 at some point if I spot a flash sale.

Acasis TBU405ProMax TB4 with 2 x Lexar NM790 2TB Set to Raid 0 (just to get one volume)

2 X Acasis TB3 enclosures with Samsung 970 Pro 1TB in Raid 0

 

[Chancha]

Some interesting results here:
I had planned to update the 2 X Acasis TB3 enclosures to TB4 at some point if I spot a flash sale.

Acasis TBU405ProMax TB4 with 2 x Lexar NM790 2TB Set to Raid 0 (just to get one volume)
View attachment 2448354

2 X Acasis TB3 enclosures with Samsung 970 Pro 1TB in Raid 0
View attachment 2448355

The 970 EVO Plus was on the incompatible list of Acasis if I recall. Not sure if the same applies to the 970 PRO.

 

[MrGimper]

Bought this recently and put a WD Black SN850X 2GB into it: https://www.amazon.co.uk/dp/B0CLV3D3H6?ref=ppx_yo2ov_dt_b_fed_asin_title

It looks/feels premium and I'm getting good speeds on a Mac Studio M1 Ultra (see screenshot). However, it gets hot and the fan ramps up and is noisy. I mainly work with audio and I couldn't use this while working on audio in the same room - it's louder than my entire Mac Studio which is a bit silly. It should be the type of flat fan similar to what's in a MacBook rather than a useless normal 2cm fan crammed in there. The ramping up and down of the fan is also annoying. It's just sat there after a file transfer with the fan ramping up and down, not even in use. I'm pretty sure the fan isn't helping things much, which seems like a waste. If it's going to make noise, at least make it do something.

Thankfully, I bought this for video editing, where I don't really care about noise as much. So I'll leave it unplugged unless I'm working on video. And no, there is no way to turn the fan off.

Further, the drive sits a slight angle so the included thermal pad doesn't really make much more than 30% contact with the chassis. I eventually added some more of my own that I had lying around, no idea if it's helping or not.

While I'm happy with the speed, I wouldn't buy another unless it was for video or photo editing.

This whole situation is so frustrating. I'd be happy with USB 3.2 but they all seem to have dropouts and disconnections. Next time I'm probably just going for an off the shelf Samsung T9 or whatever.

So difficult to find fast, reliable and quiet external storage.

Bought 2 of those from Amazon at 2 different times and although it’s premium and fast, the fans on both clicked and whined.

Really ruined it for me. Ugreen support was terrible, led me a merry dance with irrelevant questions and I ultimately returned to Amazon.

I was getting 3200 read and write with a 2TB Hynix P31.

Edit: Forgot to mention that I did try jamming something in the tiny fan blades to stop it spinning and that worked, so…

 

[Melbourne Park]

Some interesting results here:
I had planned to update the 2 X Acasis TB3 enclosures to TB4 at some point if I spot a flash sale.

Acasis TBU405ProMax TB4 with 2 x Lexar NM790 2TB Set to Raid 0 (just to get one volume)
View attachment 2448354

2 X Acasis TB3 enclosures with Samsung 970 Pro 1TB in Raid 0
View attachment 2448355

The speed difference may not be due to T4 v T3 ... as I thought for data, they have the same speeds. But the controllers etc maybe the bottleneck?

 

[Pressure]

The speed difference may not be due to T4 v T3 ... as I thought for data, they have the same speeds. But the controllers etc maybe the bottleneck?

They don't have the same speed.

Thunderbolt 3 only has 22Gbps for data while Thunderbolt 4 has 32Gbps.

The slow write performance for the Samsung 970 Pro is probably due to incompatibility with the enclosure as well.

 

[PaulD-UK]

@Pressure "Thunderbolt 3 only has 22Gbps for data while Thunderbolt 4 has 32Gbps."

But how can you use this extra 'dynamic 'allocation of bandwidth?

'The key differences between Thunderbolt 4 and Thunderbolt 3 are a minimum bandwidth requirement of 32 Gbit/s for PCIe link...' (source Intel).

But Intel (before TB5) haven't produced a Device Controller chip (JHL series) for devices to connect to that 32 Gbps.
Intel has only produced the TB4 Goshen Ridge JHL8440 hub controller chip, which only allows 1 lane of PCIe (~800MB/s) to be allocated to internal NVMe devices (SSDs).
The rest is allocated to up to 3 downstream TB3/4 ports, which share the 32Gbps.
With DP video, USB3.*, Ethernet, audio etc bringing the dynamic allocation up to the nominal 40Gbps.

So to use a TB4 computer port for connecting SSDs, the fastest individual TB connection is still TB3 (22Gbps).

It's only USB4 devices that can be faster than TB3.

"Thunderbolt 4 is an implementation of USB4 "40 Gbps". Thunderbolt 4 mandates some features that are optional in USB4 including: backwards compatibility to Thunderbolt 3, minimum PCIe ("32 Gbps") and DP capabilities..." (source Intel).

"USB4 is based on the Thunderbolt 3 protocol; however, the implementation of Thunderbolt 3 protocol is mandatory only for hubs"(source USB-IF).

The USB4 protocol does allow up to the full 32Gbps to go to a single attached device.
Until TB5 is fully available...

Which poses an interesting possibility.
Will existing TB4 computer ports work at the higher (USB4 32Gbps) data rate with a TB5 SSD attached?

 

[Pressure]

@Pressure "Thunderbolt 3 only has 22Gbps for data while Thunderbolt 4 has 32Gbps."

But how can you use this extra 'dynamic 'allocation of bandwidth?

'The key differences between Thunderbolt 4 and Thunderbolt 3 are a minimum bandwidth requirement of 32 Gbit/s for PCIe link...' (source Intel).

But Intel (before TB5) haven't produced a Device Controller chip (JHL series) for devices to connect to that 32 Gbps.
Intel has only produced the TB4 Goshen Ridge JHL8440 hub controller chip, which only allows 1 lane of PCIe (~800MB/s) to be allocated to internal NVMe devices (SSDs).
The rest is allocated to up to 3 downstream TB3/4 ports, which share the 32Gbps.
With DP video, USB3.*, Ethernet, audio etc bringing the dynamic allocation up to the nominal 40Gbps.

So to use a TB4 computer port for connecting SSDs, the fastest individual TB connection is still TB3 (22Gbps).

It's only USB4 devices that can be faster than TB3.

"Thunderbolt 4 is an implementation of USB4 "40 Gbps". Thunderbolt 4 mandates some features that are optional in USB4 including: backwards compatibility to Thunderbolt 3, minimum PCIe ("32 Gbps") and DP capabilities..." (source Intel).

"USB4 is based on the Thunderbolt 3 protocol; however, the implementation of Thunderbolt 3 protocol is mandatory only for hubs"(source USB-IF).

The USB4 protocol does allow up to the full 32Gbps to go to a single attached device.
Until TB5 is fully available...

Which poses an interesting possibility.
Will existing TB4 computer ports work at the higher (USB4 32Gbps) data rate with a TB5 SSD attached?

You will probably laugh at the solution but TB4 to PCIe 3.0 x 4 docks enable 32Gbps.

 

[PaulD-UK]

@ Pressure "TB4 to PCIe 3.0 x 4 docks enable 32Gbps."

Do you have a link?

 

[Pressure]

@ Pressure "TB4 to PCIe 3.0 x 4 docks enable 32Gbps."

Do you have a link?

You can search for any number of TB to GPU dock.

They talk about it a lot at eGPU.io.

 

[Melbourne Park]

They don't have the same speed.

Thunderbolt 3 only has 22Gbps for data while Thunderbolt 4 has 32Gbps.

The slow write performance for the Samsung 970 Pro is probably due to incompatibility with the enclosure as well.

I have a PCIe RAID card, made by HighPoint, an SSD7101A-1 NVME RAID controller card, which takes up to 4 NVME cards. I queried HighPoint and they said the data transfer was similar from T-3 to T-4 for an external device connected to a T-4 computer. ie an external PCIe device (popular for GPUs but can also operate other PCIe hardware depending on the OS etc). Using a single cable though. They said expect 3,000 Mb/s overall.

Sonnet says for their Echo II DV Expansion Module (2-Slot PCIe card expansion module for Echo and DuoModo expansion enclosures) $900 duel PCIe Thunderbolt device: This external works with Thunderbolt 4 devices, but with Mac, it seems Thunderbolt 3 on the computer, provides a little bit less performance:

Speeds:

• 40Gbps Thunderbolt Interface Offers 3000 MB/s of PCIe Bandwidth with Thunderbolt 4 computers; 2570 MB/s with Thunderbolt 3 computers

Which matches what High Point told me I would expect in speed if my card was connected to a Mac Studio.

While T-4 does promise 32 Gb/s it seems the manufacturers all settle on 3,000. There are other computer bottlenecks with sustained transfer speeds I presume. Windows gets tricky as sometimes one controller only for two ports. And curiously T-3 is said to be 2570 from Sonnet rather than your 2200.

OWC though have said their T-5 external shows around 6,000. Curiously the MacBook Pro m3 40 core I am testing at the moment only has 1 TB, and its black magic score is 6350 Write and 4300 Read. Faster than I had expected.