Asus unveils an RTX 4060 Ti with an M.2 slot for PCIe 5.0 SSDs or external GPUs

Daniel Sims

Posts: 1,292   +43
Staff
Not that you would, but you could: When details regarding a graphics card with an integrated M.2 slot emerged earlier this year, the purpose of the unusual feature was unclear. This week, a proper unveiling from Asus shed light on the advantages of connecting additional motherboard components to a GPU.

Asus's new Dual GeForce RTX 4060 Ti SSD mostly looks and behaves like a typical mid-range Ada Lovelace graphics card. However, its embedded M.2. port leverages the component's unique motherboard placement to improve SSD performance and support other hardware.

The notable feature is possible on the mainstream GPU because its graphics processing functionality only uses 8 out of 16 PCIe lanes. The remaining lanes, which higher-end cards can't spare without suffering performance degradation, go toward the M.2 slot, allowing mid-range GPU users to take full advantage of their motherboard.

An additional NVMe SSD port is an obvious advantage, especially in PCs with a limited number of them. Other vendors have released add-in-boards dedicated to hosting SSDs to squeeze mass storage options into compact setups. However, a graphics card's cooling apparatus offers another benefit.

Although the Dual RTX 4060 Ti is a PCIe 4.0 GPU, it can operate an SSD at PCIe 5.0 speeds if the drive and motherboard support PCIe 5.0. In the above video, Asus demonstrated an integrated SSD reaching read/write speeds of around 12GB/s.

To attain such performance while connected directly to the motherboard, PCIe 5.0 SSDs require enormous heatsinks. Otherwise, they can throttle to speeds reminiscent of disk drives. Meanwhile, the company's demonstration showed the GPU-connected drive maintaining a stable 50 degrees Celsius with no discernable slowdown.

Asus further demoed the card's versatility by connecting the M.2 slot to an external RTX 4090. Although the flagship card is shown operating with fewer PCIe lanes than usual, the PCIe 5.0 connection appears to handle it well. The video shows the setup running Cyberpunk 2077 at around 150 fps (with unknown settings).

It isn't clear why anyone would want to devote an extra M.2 slot to a second GPU, but the capability is there if needed. Perhaps the external connection could help users with PCs too small to contain the monstrous 4090. Another more sensible use might be attaching an external capture card.

Asus warns that engaging the Dual RTX 4060 Ti's integrated M.2 slot requires PCIe lane splitting, which users must manually activate in the BIOS. Furthermore, the only Intel motherboards that support the feature are the Z790, Z690, and H770, while all AMD boards include splitting.

The Dual GeForce RTX 4060 Ti SSD is set to become available this month, but Asus didn't specify a price.

Permalink to story.

 
While it's a great idea for some builds I still think that ryzen CPUs don't come with enough PCI-e lanes. This might be a good idea on some mATX builds but I have a hard time seeing the usefulness of an m.2 slot on a GPU outside of some niche setups. There should be a middle ground between ryzen CPUs and threadripper and they could have limited it only to their X570 chipsets or something. Now we're stuck with the amount of PCIe lanes that the socket can support.

Now I'm an AMD guy and it was cool seeing 16 cores on a system with support for 128GB of ram but AM5 has a lot holding it back. It doesn't have ECC support, it could easily support 256GB if they wanted and it's limited on PCI-e lanes.

I look forward to my first AM5 build but it seems like a bit of a missed opportunity. The difference between more PCI lanes and ECC support shouldn't be $3000. If you really need that stuff as a hobbyist you can always pick up used server hardware for cheap but that opens up a whole other can of worms entirely.
 
Zen 4 already has a issues with boot times because of memory training. ECC will only add more time to this.

This pcie split is a niche thing, me I dont see any use for it.
 
I like how they cool the SSD, by using the cooler of the graphics card itself with the small cutout on the PCB and the thermal pads. That's really smart tbh. And a much better prospect in terms of access, performance, and cooling compared to the SSD on my 7 year old SFF build, where my m.2 SSD is hidden on the back of the motherboard with no airflow (I added four 1-inch chipset heatsinks to reduce the throttling issues and this is just a PCIe 3.0 SSD).

It is certainly a niche product without question, but if I were to try another SFF build (custom enclosure, how small can I goooooooo) I could see myself potentially using such a card if I needed the additional storage but lacked the physical capacity to add it to the motherboard.
 
Utterly useless, but very marketable to the utterly uninformed.
As a programmer if it's just extra storage then its still a bonus... but I get the sneaking suspicion that it could be used for offloading extra work to. Maybe used as an interface between my code and the GPU. Maybe put some code that runs along with the AI use that as a buffer. Could be a lot of use.
 
pDs7WRn.jpg
 
I've been out of the performance and gaming scene for too long, so I'm not understanding what this is for. I would think this be ideal to put game assets on the GPU itself so maybe getting that data to video memory can be faster. This may help in games like GTA where you could have stutters when trans-versing large portions of the map that need to load into vRAM. Is that not what it is for?
 
I've been out of the performance and gaming scene for too long, so I'm not understanding what this is for. I would think this be ideal to put game assets on the GPU itself so maybe getting that data to video memory can be faster. This may help in games like GTA where you could have stutters when trans-versing large portions of the map that need to load into vRAM. Is that not what it is for?
Basically what I was saying I bet you will be able to offload code to it quicker then if it was in normal ram.
 
Back