ASMedia Showcases PCIe and USB4 Solutions at COMPUTEX 2025

[Nomad76]

ASMedia Technology is getting ready to unveil its latest PCIe Packet Switch solution (ASM58048) at Computex 2025, with sampling expected in Q3 2025. ASMedia product lineup covers PCIe generations 2 through 6, offering channel configurations that range from 3 to 80 channels. The upcoming PCIe Gen 5 product (ASM68080 series) development is underway, with a fully verified PHY design nearing completion. Following the ASM58048, ASMedia plans for multiple products featuring 24 to 80 channels, while work on PCIe Gen 6 is actively underway.

The company's USB4 ASM4242 host controller has found a home in high-end motherboards and laptops from a variety of brands being certified USB4 and Thunderbolt 4. The ASM2464PDX USB4 controller is tailored for edge AI computing, allowing for plug-and-play external GPUs by converting PCIe Gen 4 signals to USB4. This makes it easy to connect graphics cards, NVMe SSDs, and Wi-Fi modules externally, delivering performance that rivals internal setups.



Update May 22nd: Added more pictures

ASMedia's President Che-Wei Lin said, "ASMedia's PCIe solutions not only offer superior performance but also demonstrate unparalleled flexibility and compatibility, enhancing the user experience. From Gen 2 to Gen 6, our product roadmap comprehensively covers the needs of each stage, with a focus on the future demands of AI and edge computing. The PCIe Gen 5 ASM68080 series is poised for launch, and PCIe Gen 6 design is also within our development roadmap. This reflects our commitment to meeting the market's demand for high-speed designs. Furthermore, ASMedia's mass-produced USB4, through single-cable technology, achieves seamless integration of data, video, and power, redefining application possibilities across industries. In the future, ASMedia looks forward to collaborating with global partners to continuously promote and enhance the PCIe and USB4 ecosystems."



Pictures from TechPowerUp visit to ASMedia demo room


View at TechPowerUp Main Site | Source

 

[Tek-Check]

Where is USB4 v2 controller with 80 Gbps ports?

 

[TheLostSwede]

Where is USB4 v2 controller with 80 Gbps ports?

I will go see them Thursday, I'll ask, but I doubt there has been much progress.
This news post was based on information from a press release and they didn't show anything new at their stand located at the USB-IF booth in Nangang, as I walked past there today, but they have a suite in a hotel where the pics in the press release are from.
In fact, the only 80 Gbps stuff in the USB-IF booth, was cables.

 

[cal5582]

wait are they actually going to make pcie switch chips? someone actually taking up plx's mantle?

 

[Tigerfox]

wait are they actually going to make pcie switch chips? someone actually taking up plx's mantle?

Haven't they bought PLX?

 

[SiggiTheHun]

@ cal5582 & Tigerfox

2018:
---> https://www.anandtech.com/show/13022/asmedia-preps-asm2824-pcie-30-switch

 

[cal5582]

@ cal5582 & Tigerfox

2018:
---> https://www.anandtech.com/show/13022/asmedia-preps-asm2824-pcie-30-switch

no pricing listed. oof.
wonder how much those cost now. i need some actual expansion slots. x670e boards only having 3 pcie slots is kind of insanely sparse.

 

[_roman_]

It does not change the fact that the bandwidth is shared. There will be future topics my "nvme" speeds are slow. Like the one we had recently.

 

[cal5582]

yeah its limited by the uplink speed. alot of the cards i use are x1 though, they used to put x1 slots on motherboards.

 

[Cheeseball]

Where is USB4 v2 controller with 80 Gbps ports?

This guy ^ asking the right question LOL

 

[Tek-Check]

In fact, the only 80 Gbps stuff in the USB-IF booth, was cables.

That's interesting. Thanks for the feedback. We know they have been testing 80Gbps controller ASM2892A0 for ages, so we should hear at least what's the progress with that one.

 

[TheLostSwede]

no pricing listed. oof.
wonder how much those cost now. i need some actual expansion slots. x670e boards only having 3 pcie slots is kind of insanely sparse.

Their goal is to be the most affordable option in the market, but pricing depends on how many lanes you want to be able to switch.

That's interesting. Thanks for the feedback. We know they have been testing 80Gbps controller ASM2892A0 for ages, so we should hear at least what's the progress with that one.
View attachment 400526

This was apparently just them demoing a working PHY, it's not the full chip.

Where is USB4 v2 controller with 80 Gbps ports?

2027 expected availability from what I was told.

It's a mix between not having enough engineers, the USB IF not having finalised all the documentation/testing requirements, the company working on AMD's next gen chipset and a bunch of other products at the same time and finally no demand from their customers.
There is apparently not even a whole lot of demand for their device controllers outside of some storage products, although the somewhat recent Thunderbolt 4 certification and Intel's blessing (finally), might change that.
It's something of a chicken/egg effect, plus even most of the readership here doesn't seem all the interested in USB4 or Thunderbolt, at least that's what it feels like based on the comments here when we write about those kind of products.

 

[Tek-Check]

It's something of a chicken/egg effect, plus even most of the readership here doesn't seem all the interested in USB4 or Thunderbolt, at least that's what it feels like based on the comments here when we write about those kind of products.

Yes, it does feel like that at the moment. My impression is that OEMs are responsible for generating interest, in particular offering USB4/TB5 ports on displays and GPUs, so that users can actually start using those goodies that the interfaces provide. If they do not offer ports on entire ecosystem of devices, then the ecosystem cannot be created and people will not be as interested in using those connections.

 

[LabRat 891]

Finally! I was expecting to see (semi-affordable) Gen4 PCIe switches from ASmedia, years ago.
AFAIK, ASMedia-AMD's PROM21 (B650,B850,X670,X870) already had the basic functionality from day 1. So, I'm not quite sure what's taken them this long?

 

[Lianna]

2027 expected availability from what I was told.

It's a mix between not having enough engineers, the USB IF not having finalised all the documentation/testing requirements, the company working on AMD's next gen chipset and a bunch of other products at the same time and finally no demand from their customers.
There is apparently not even a whole lot of demand for their device controllers outside of some storage products, although the somewhat recent Thunderbolt 4 certification and Intel's blessing (finally), might change that.
It's something of a chicken/egg effect, plus even most of the readership here doesn't seem all the interested in USB4 or Thunderbolt, at least that's what it feels like based on the comments here when we write about those kind of products.

I'd guess that for the last 6-7 years most of the readership was accustomed to Thunderbolt (esp. 3) being one or more of:

- a security problem (DMA), being patched again and again, then some devices having default "BIOS" TB switches off, making some devices not work without providing a hint as to why;

- TB devices not being backwards compatible with USB (until late last year IIRC) or compatible (very rare), but with (much) lower performance on both interfaces (see an older TB SSD review on AT);

- compatibility problem, with some older TB devices needing to be "approved" manually (for security reasons, see DMA above), but notebook vendors failing to provide (working!) software (like Intel Thunderbolt Control Center or Thunderbolt Software) for that - I have a TB drive dock that was made to work on some Intel TB controllers and Win10, but not on Win11 ("does not need the Software, it has Settings for that", sadly, false) and not on USB4 (obviously Intel's Software does not recognize the controller, does ASMedia have software for device "approval"?);

- stability problem, some hosts and/or some OS versions (e.g. was OK, bad after OS update) bluescreening often with some TB/USB4 device(s) connected (and some not even once in their history without one);

- feature problem, some hosts and/or some OS versions e.g. repeatedly waking from sleep after a short while with some TB/USB4 device(s) connected;

- most (all?) host TB3/TB4 ports being backwards compatible with USB 3.2 Gen 2x1 only, not 2x2 - apparently ASMedia's USB4 host can do 2x2, is that right?

- TB3/TB4 ports reserving a lot of unused bandwidth, making them only somewhat faster for e.g. external SSDs than Gen 2x2 (USB4 fortunately does not exhibit this behaviour, see 2025 USB4 SSD reviews);

- some hosts using Intel-accepted 2-lane PCIe connection for TB host controller, negating any performance advantage over Gen 2x2 (slower, in fact);

- and that's without talking about prices, that are just recently coming down to somewhat sane levels...

So with quite a few hosts with TB3/TB4/USB4 and quite a few devices, the dock I prefer is a non-USB4 one, just feature complete (Gen 2 downstream, HDMI 2.x, PD65W+, preferably PD100W+).
New, preferably ASMedia-based, USB4 SSDs directly connected to (obviously ASMedia-based) USB4 hosts may change my mind, provided someone finally offers them in 8+ TB sizes.
It will take some time to change the overall stigma.

 

[TheLostSwede]

View attachment 400820
Finally! I was expecting to see (semi-affordable) Gen4 PCIe switches from ASmedia, years ago.
AFAIK, ASMedia-AMD's PROM21 (B650,B850,X670,X870) already had the basic functionality from day 1. So, I'm not quite sure what's taken them this long?

Not enough people, despite re-using IP blocks between products, but the AMD chipsets only goes up to blocks of 4 lanes, these products goes much further.

I'd guess that for the last 6-7 years most of the readership was accustomed to Thunderbolt (esp. 3) being one or more of:

- a security problem (DMA), being patched again and again, then some devices having default "BIOS" TB switches off, making some devices not work without providing a hint as to why;

- TB devices not being backwards compatible with USB (until late last year IIRC) or compatible (very rare), but with (much) lower performance on both interfaces (see an older TB SSD review on AT);

- compatibility problem, with some older TB devices needing to be "approved" manually (for security reasons, see DMA above), but notebook vendors failing to provide (working!) software (like Intel Thunderbolt Control Center or Thunderbolt Software) for that - I have a TB drive dock that was made to work on some Intel TB controllers and Win10, but not on Win11 ("does not need the Software, it has Settings for that", sadly, false) and not on USB4 (obviously Intel's Software does not recognize the controller, does ASMedia have software for device "approval"?);

- stability problem, some hosts and/or some OS versions (e.g. was OK, bad after OS update) bluescreening often with some TB/USB4 device(s) connected (and some not even once in their history without one);

- feature problem, some hosts and/or some OS versions e.g. repeatedly waking from sleep after a short while with some TB/USB4 device(s) connected;

- most (all?) host TB3/TB4 ports being backwards compatible with USB 3.2 Gen 2x1 only, not 2x2 - apparently ASMedia's USB4 host can do 2x2, is that right?

- TB3/TB4 ports reserving a lot of unused bandwidth, making them only somewhat faster for e.g. external SSDs than Gen 2x2 (USB4 fortunately does not exhibit this behaviour, see 2025 USB4 SSD reviews);

- some hosts using Intel-accepted 2-lane PCIe connection for TB host controller, negating any performance advantage over Gen 2x2 (slower, in fact);

- and that's without talking about prices, that are just recently coming down to somewhat sane levels...

So with quite a few hosts with TB3/TB4/USB4 and quite a few devices, the dock I prefer is a non-USB4 one, just feature complete (Gen 2 downstream, HDMI 2.x, PD65W+, preferably PD100W+).
New, preferably ASMedia-based, USB4 SSDs directly connected to (obviously ASMedia-based) USB4 hosts may change my mind, provided someone finally offers them in 8+ TB sizes.
It will take some time to change the overall stigma.

I wasn't aware that Thunderbolt had that many issues. Seems like Intel has done a good job covering up most of that.
Yes, USB4 is fully backwards compatible with all speed grades of USB 3.x, as that is a requirement by the USB-IF.
Do keep in mind that until recently, Thunderbolt was also limited to PCIe 3.0 x4, hence why the ASM4242 was faster. Only the latest gen of Thunderbolt 4/5 chips from Intel has a PCIe 4.0 x4 system interface.
The reason for that weird two lane PCIe interface for Thunderbolt was supposedly power saving, it was only meant for some laptops.

 

[Tek-Check]

I'd guess that for the last 6-7 years most of the readership was accustomed to Thunderbolt (esp. 3) being one or more of:
It will take some time to change the overall stigma.

This. A huge mess in the ecosystem, linked to compatibility, security and different feature-set, utterly confusing for buyers and purposefully obfuscated by vendors of devices. I'd add that they also manipulate with generic interface information and deceive the public.

Example of deception in advertisement. I bought Z390 ITX from Asrock years ago, with TB3 port, advertised, of course, as "40 Gbps". Little did I know that the controller supported only PCIe3 x2 link. I found out only later, after seeing low transfer speeds, that it was JHL6240... This was never listed on the product page or explained to buyers the speed of different protocols through generic 40 Gbps cable.

They do the same now with Thunderbolt 5. I can see listings "80 Gbps" and "120 Gbps" for external GPU enclosure, for example from Gigabyte at Computex, which is complete nonsense. PCIe data used by GPU to communicate with laptop's CPU/RAM will travel at 64 Gbps only. Nobody is saying this to buyers or listing it on their product pages. It's a massive deception and failure to be informative and honest with customers. Utter failure.

Reported by this website and well commented by a user

GIGABYTE Shows Off AORUS RTX 5090 eGPU Box Implementing 120 Gbps Thunderbolt 5

GIGABYTE at the 2025 Computex showed off the AORUS RTX 5090 AI Box, which it claims to be the world's fastest external GPU (eGPU). This is an external GPU box with an ATX 3.1 power supply, and a Thunderbolt 5 interface that also leverages asymmetric bandwidth boost—where the host to GPU transmit...

www.techpowerup.com

New, preferably ASMedia-based, USB4 SSDs directly connected to (obviously ASMedia-based) USB4 hosts may change my mind, provided someone finally offers them in 8+ TB sizes.

I have native ASMedia USB4 host port on X870E motherboard and their SSD controller in Akasa USB4 enclosure. They work fine together, at ~3.8 GB/s.

The reason for that weird two lane PCIe interface for Thunderbolt was supposedly power saving, it was only meant for some laptops.

Are we able to hold vendors more accountable for their deceptive advertisement practices, such as new Gigabyte eGPU enclosure?

It's sad to think that whenever we discover dubious practices and claims, we need to call GamersNexus or HardwareUnboxed to try to expose it.

 

[Lianna]

They do the same now with Thunderbolt 5. I can see listings "80 Gbps" and "120 Gbps" for external GPU enclosure, for example from Gigabyte at Computex, which is complete nonsense. PCIe data used by GPU to communicate with laptop's CPU/RAM will travel at 64 Gbps only. Nobody is saying this to buyers or listing it on their product pages. It's a massive deception and failure to be informative and honest with customers. Utter failure.

Reported by this website and well commented by a user

GIGABYTE Shows Off AORUS RTX 5090 eGPU Box Implementing 120 Gbps Thunderbolt 5

GIGABYTE at the 2025 Computex showed off the AORUS RTX 5090 AI Box, which it claims to be the world's fastest external GPU (eGPU). This is an external GPU box with an ATX 3.1 power supply, and a Thunderbolt 5 interface that also leverages asymmetric bandwidth boost—where the host to GPU transmit...

www.techpowerup.com

View attachment 400875

I read somewhere that Intel tries to make some PCIe lanes asymmetrical too on its own chipsets, so that one of four PCIe lanes changes direction for TB5, making it not 64/64 Gbps (TX/RX) but 80/48 IIRC? Still not enough and out-of-spec, too. Can't find the source, though.

In the list above I still forgot about add-on TB PCIe cards for desktops, which is another can of worms.

The really sad thing is that while for newer hosts and devices points 1, 2, 7, and 8 from my list above may no longer be a problem, the others are still present, experienced in the last year on Zen 4 / Pheonix (3, 4, and 5) and Zen 5 / Strix Halo (3 and 4).

I have native ASMedia USB4 host port on X870E motherboard and their SSD controller in Akasa USB4 enclosure. They work fine together, at ~3.8 GB/s.

Thanks for confirmation, hoped so.
Because of SSD power requirements I prefer ready-made external SSDs, as external M.2 enclosures were sometimes not respecting power limits, causing the SSD to turn off mid-work.
This should not be an issue with USB4/TB hosts, where power limit is higher, but if used in compatibility mode on, say, USB 3.2 Gen 2 ports (that I need) may still be relevant.
The new Corsair EX400U seems really good, but unfortunately stops at only 4 TB.

 

[Tek-Check]

I read somewhere that Intel tries to make some PCIe lanes asymmetrical too on its own chipsets, so that one of four PCIe lanes changes direction for TB5, making it not 64/64 Gbps (TX/RX) but 80/48 IIRC? Still not enough and out-of-spec, too. Can't find the source, though.

This could be a nightmare to implement conveniently. They would need to take x1 or x2 lanes from another x4 PHY, in addition to default x4 PHY, and then sync x5 or x6 traffic speed in one direction.

This is not a solution, as they cannot predict and dictate the bandwidth of directional traffic between external GPU and laptop's CPU/RAM. By making the traffic asynchronous, they could be bottlenecking either Tx or Rx side during gameplay, as one side would always have more. How the heck they would predict Tx and Rx traffic at any time during gameplay?

They need to implement Gen5 x4 link, minimally.

 

[TheLostSwede]

@Tek-Check with regards to your comments about Thunderbolt, it's mandated by Intel. It's nothing the device makers can really make, as Intel doesn't allow them to use the actual data transfer speeds, but rather only the combined bandwidth.
In a way, this is like the marketing speeds for WiFi routers, although no-one is mandating the numbers there, but everyone knows that bigger numbers sell better.

 

[Tek-Check]

In a way, this is like the marketing speeds for WiFi routers, although no-one is mandating the numbers there, but everyone knows that bigger numbers sell better

I know, but Intel doesn't need to treat us, members of public, as if we were ignorant and naive idiots. They can simply say that the cable's bandwidth is such and such, and specific protocol speeds are such and such. It's not a rocket science...

Educate and inform the public, don't deceive them. This is one of reasons why public trust in Intel has eroded in recent years. They lack integrity and others, like Gigabyte, copy the same nonsense approach.

Since Intel cannot be trusted to provide the public with marketing material on how fast are different protocols within TB5 cable, Gigabyte cannot be trusted too, by extension.

Also, we have tech journalists and youtubers who should push them hard to provide the above mentioned information in marketing material and product specifications, so that buyers are actually aware of what they are buying.

 

[TheLostSwede]

I know, but Intel doesn't need to treat us, members of public, as if we were ignorant and naive idiots. They can simply say that the cable's bandwidth is such and such, and specific protocol speeds are such and such. It's not a rocket science...

Educate and inform the public, don't deceive them. This is one of reasons why public trust in Intel has eroded in recent years. They lack integrity and others, like Gigabyte, copy the same nonsense approach.

Since Intel cannot be trusted to provide the public with marketing material on how fast are different protocols within TB5 cable, Gigabyte cannot be trusted too, by extension.

Also, we have tech journalists and youtubers who should push them hard to provide the above mentioned information in marketing material and product specifications, so that buyers are actually aware of what they are buying.

They don't copy, they simply have no choice, as Intel dictates the standard and how its partners market it. Don't blame the motherboard guys, everyone implementing thunderbolt have to follow Intel's marketing language, or they get zero money from intel.

Most people have as much interest in USB and thunderbolt as they have in watching paint dry.
I believe I'm one of one journalist that give a shit and actually talk to the companies about this tech.

I was afaik, the one and only person that did a comparison test between 90% of all USB 3.0 host controllers back in the day. I missed one or two as I couldn't get my hands on them. AMD sucked at the time. I ended up meeting the guy in charge of their USB 3.0 implementation in Taipei at an AMD event and wanted to ask him what went wrong, he told me he was aware of my test, but refused to talk to me about it.

This is often how it is when you give a shit. I'm not afraid to call out BS, I'm old as experienced enough at this point, but if no-one cares, what can I do?

I've started writing out the actual speed of USB interfaces in all my posts here, I'll make sure to do the same for thunderbolt, when relevant, but I can't do much more.

 

[Tek-Check]

I was afaik, the one and only person that did a comparison test between 90% of all USB 3.0 host controllers back in the day. I missed one or two as I couldn't get my hands on them. AMD sucked at the time. I ended up meeting the guy in charge of their USB 3.0 implementation in Taipei at an AMD event and wanted to ask him what went wrong, he told me he was aware of my test, but refused to talk to me about it.

This is often how it is when you give a shit. I'm not afraid to call out BS, I'm old as experienced enough at this point, but if no-one cares, what can I do?

Keep doing the great job on this front, keep asking questions, keep calling out BS, keep letting them know if they hide something from the public. I highly appreciate your content because it is informative and educational.

It is true that Intel in their TB5 brief for tech media mentioned 64 Gbps for PCIe data and 'two times more bandwidth for external SSDs and eGFX...', so they are formally on the "good side of the Force". However, at this moment of time, the key problem is marketing of devices that actually are meant to use 64 Gbps link for PCIe data. All we can see is this gigantic "80/120 Gbps" mantra, which melts my brain.

 

[Cheeseball]

Keep doing the great job on this front, keep asking questions, keep calling out BS, keep letting them know if they hide something from the public. I highly appreciate your content because it is informative and educational.

It is true that Intel in their TB5 brief for tech media mentioned 64 Gbps for PCIe data and 'two times more bandwidth for external SSDs and eGFX...', so they are formally on the "good side of the Force". However, at this moment of time, the key problem is marketing of devices that actually are meant to use 64 Gbps link for PCIe data. All we can see is this gigantic "80/120 Gbps" mantra, which melts my brain.

Technically the 80/120Gbps may still be correct, since its just not for the PCI-E tunnelling alone. No one is able to test this since a TB5-equipped eGPU isn't publicly out yet, but theoretically if one uses it without an external monitor, the DisplayPort data that goes back to the host device should not heavily bottleneck the rest of the connection now at 80 Gbps. 64 Gbps going to the eGPU then around 16 Gbps for DP data (which may be compressed with DSC) going back.

Just like with early TB3 where it was actually 32Gbps PCI-E 3.0 x4 + 8Gbps DP data for a total of 40Gbps. So maybe with TB5 now its 64Gbps PCI-E 4.0 x4 + 16 Gbps DP data?

I'm not sure how the 120 Gbps link up would play out, unless Intel has future plans for upgrading TB5 (just like how newer TB4 chipsets support PCI-E 4.0 x4 for 64Gbps) somehow.

Marketing like this sucks, but the total data bandwidth may more or less be correct.