News Asus Flaunts GeForce RTX 4060 Ti with M.2 Slots for SSDs

[Admin]

Asustek's prototype GeForce RTX 4060 Ti can also store data.

Asus Flaunts GeForce RTX 4060 Ti with M.2 Slots for SSDs : Read more

 

[InvalidError]

"PCIe switches are not cheap."

Good thing they aren't needed here since the 4060 leaves x8 lanes unused that can be permanently routed to the 4.0x4 NVMe slots instead of using any switches.

 

[Deleted member 2731765]

In addition to Nvidia's AD102 GPU with 4352 CUDA cores and 8GB of GDDR6 memory, the prototype Asus GeForce RTX 4060 Ti graphics card has two M.2-2280 slots with a PCIe 4.0 x4 interface for SSDs, one on the front and one of the back side of the PCB.

Er, typo ? It's AD106 GPU, not 102.

 

[ezst036]

At first glance this seems to be an odd pairing, but then I realized this is a really good place for a PCIe SSD because the companies who make those things want to put heatsinks on them that are like over 90 inches tall.

 

[Deleted member 2731765]

BTW, AMD actually also made a similar approach with its Radeon SSG graphics cards, but rather than utilizing the additional SSDs for traditional storage, the SSG made use of the NVMe storage to boost bandwidth and transfer rates for the GPU itself.

The tech allowed AMD to feature large flash memory (M.2) on the graphics board, 2TB, along with the local memory that the GPU can send large data sets to directly rather than having it sent over the system memory which has slower processing capabilities. Sounds familiar ?

AMD Radeon Pro SSG Specs

AMD Vega 10, 1500 MHz, 4096 Cores, 256 TMUs, 64 ROPs, 16384 MB HBM2, 945 MHz, 2048 bit

www.techpowerup.com

 

[Roland Of Gilead]

Would there be any benefit for Direct Storage, given the ssd's are on the card?

 

[InvalidError]

Would there be any benefit for Direct Storage, given the ssd's are on the card?

No, because the SSDs are connected to the otherwise unused PCIe lanes on the x16 slot and still have to go through the CPU to reach the GPU despite being on the same physical card.

 

[Deleted member 2731765]

Would there be any benefit for Direct Storage, given the ssd's are on the card?

Nope.

 

[Alvar "Miles" Udell]

An avid enthusiast would ask legitimate questions about cooling and supplying power for both the GPU and the drives since the card only has one eight-pin auxiliary PCIe power connector, which can officially provide up to 150W of power — the GeForce RTX 4060 Ti is rated for 160W. As it turns out, the drives can fetch the power from the slot.

Really? I'd say a novice would ask those questions, anyone calling themselves and enthusiast already knows the answers to those questions.

 

[Roland Of Gilead]

No, because the SSDs are connected to the otherwise unused PCIe lanes on the x16 slot and still have to go through the CPU to reach the GPU despite being on the same physical card.

Nope.

Ah, okay. Makes sense.

 

[atomicWAR]

Really? I'd say a novice would ask those questions, anyone calling themselves and enthusiast already knows the answers to those questions.

Yes the pcie slot provides 75 watts of power, the eight pin 150W and you have the card eating 160W ...leaving an eye watering 65W by ssd standards. More than enough power to run 6 M.2 drives at full tilt (assuming max 10W on writes) if there had been enough PCIe lanes left to use that excess wattage. But not everyone identifying as an enthusiast is going to know all those exact numbers off the top of their head...though they should have some guesstimates in the ball park to be fair. Yet that's also not including those readers who aren't enthusiasts and don't know any better. Point being I'd say its a fair call on the author's part to include such language all things considered. But I do get where you're coming from don't get me wrong.

That said, its not a very interesting product imo. Something with a direct connection to the the GPU would have perked my ears but this is just adding an extra m.2 slot by leeching of the vacant lanes left from a overly gimped gpu. Niche at best and some what of a curiosity therefore but there's not much interesting beyond that. To be fair though we'd need Nvidia involved directly to get something like a direct GPU to onboard m.2 slot/ssd interface, not just an AIB partner.

 

[InvalidError]

That said, its not a very interesting product imo. Something with a direct connection to the the GPU would have perked my ears but this is just adding an extra m.2 slot by leeching of the vacant lanes left from a overly gimped gpu. Niche at best and some what of a curiosity therefore but there's not much interesting beyond that.

If motherboard manufacturers won't put the hardware on the motherboard to route the excess PCIe lanes elsewhere when not used by the GPU or just ditch the x16 hard-wired slot for only x8 and use the other lanes elsewhere, then GPUs putting the extra unused lanes to some other use looks like a great idea to me. I may not need the extra NVMe slots today but might some number of years down the line, especially as more games enter the 150+GB install size arena and 1TB SSDs continue offering some of the best $/TB.

I can imagine this becoming a popular value-add on GPUs with only x8 PCIe. Given a choice of two similar GPUs with the 2xNVMe model costing significantly less than a dedicated x8-to-2xNVMe card, I'd probably get the NVMe GPU as I despise the idea of having completely dead-ended IO in my PC.

 

[H4UnT3R]

So when I have 26gbps pcie throughout on 3060 12gb, how much will be missing for 4060ti which has more CUDA cores etc? :-D Well, seems I'll skip 4xxx and we'll see about 5xxx :-/ hope AMD will have better AI support, because CUDA still rules...

 

[sfjuocekr]

You did not know that the PCI-Express slot is able to deliver 75 Watt of power?

This is not something the drivers do.

 

[PEnns]

Hmmm, a GPU that has 2 M.2 SSD slots......sounds really worth buying now.

Nahhh, just kidding. Wouldn't buy it even if it made a mean double espresso and Canoli cream cake as well!!

 

[Zerk2012]

The extra cost to make the card will keep it from selling, most boards today have at least 2X M.2 slots already.
I could be wrong I was one time I'm sure.

 

[LabRat 891]

I'm curious if the GPU has PCIe switching capabilities embedded, or if these are just bifurcated traces ran to the M.2s.

Great idea, if Consumer/Enthusiast motherboards were not so inconsistent in fully exposing bifurcation to the User.

 

[atomicWAR]

If motherboard manufacturers won't put the hardware on the motherboard to route the excess PCIe lanes elsewhere when not used by the GPU or just ditch the x16 hard-wired slot for only x8 and use the other lanes elsewhere, then GPUs putting the extra unused lanes to some other use looks like a great idea to me. I may not need the extra NVMe slots today but might some number of years down the line, especially as more games enter the 150+GB install size arena and 1TB SSDs continue offering some of the best $/TB.

I can imagine this becoming a popular value-add on GPUs with only x8 PCIe. Given a choice of two similar GPUs with the 2xNVMe model costing significantly less than a dedicated x8-to-2xNVMe card, I'd probably get the NVMe GPU as I despise the idea of having completely dead-ended IO in my PC.

I didn't mean it was a bad idea, only I suspect it will be niche. Thus not to interesting to myself. I could be wrong on that though I admit and it could become wildly popular. Heck maybe if it did Nvidia might consider making it a standard thing, maybe going as far as including a direct interface. Time will tell.

 

[purpleduggy]

The extra cost to make the card will keep it from selling, most boards today have at least 2X M.2 slots already.
I could be wrong I was one time I'm sure.

extra cost? an nvme m.2 port is like less a dollar, can buy 10x Foxconn branded NVMe M.2 ports for 5 USD in Shenzhen (ASUS buys in bulk so is even way cheaper) with some caps and power regulation. PCIe switch is like a few dollars per chip. Asus has all these parts available from motherboard manufacturing anyway. would probably be less than 5 dollars difference in cost. because the RTX4060 is not even using half of the PCIe lanes. Rest is pcb design. Plus its ASUS so they're gonna be 20% more expensive anyway, NVME drive or not.

NVMe on GPUs is going to become commonplace.

 

[thestryker]

"PCIe switches are not cheap."

Good thing they aren't needed here since the 4060 leaves x8 lanes unused that can be permanently routed to the 4.0x4 NVMe slots instead of using any switches.

It's going to need bifurcation built in because Intel and AMD do not do this on client platforms. Unless there's some other way to expose the SSDs to the system that I'm unaware of.

 

[InvalidError]

Heck maybe if it did Nvidia might consider making it a standard thing, maybe going as far as including a direct interface. Time will tell.

If Nvidia axes the PCIe interface from x16 to x8 to save money, I don't think it is going to add x4/x8 back to its chips on top of PCIe switching/routing logic in the GPU to add direct-connected NVMe to entry-level GPUs. It would be cheaper and simpler to just put the x16 interface back and allow the GPU to access NVMe storage wherever else it may be.

I don't expect any help with this from penny-grubbing GPU chip manufacturers obsessed in paring their products down to the absolute minimum viable product people will still buy at the highest profit margins they are able to squeeze out of the market.

 

[Zerk2012]

extra cost? an nvme m.2 port is like less a dollar, can buy 10x Foxconn branded NVMe M.2 ports for 5 USD in Shenzhen (ASUS buys in bulk so is even way cheaper) with some caps and power regulation. PCIe switch is like a few dollars per chip. Asus has all these parts available from motherboard manufacturing anyway. would probably be less than 5 dollars difference in cost. because the RTX4060 is not even using half of the PCIe lanes. Rest is pcb design. Plus its ASUS so they're gonna be 20% more expensive anyway, NVME drive or not.

NVMe on GPUs is going to become commonplace.

No you have the extra cost of designing the board and added manufacturing cost.
That would probably push the cost close to a 4070 with better performance.

Plus its ASUS

Another reason not to even consider it. Asus is nothing but gimmicks unless you look at the way overpriced products that you can buy just as good of a product from another manufacture for less money AI suit people actually pay for a gimmick software that can cause all kinds of problems.

Asus was a decent brand about 10 years ago.

 

[deesider]

The extra cost to make the card will keep it from selling, most boards today have at least 2X M.2 slots already.
I could be wrong I was one time I'm sure.

Looks like it could make sense on a mini-ITX board for someone who has used up all of their m2 allocation and doesn't have another PCIe slot. Other than that it seems more of a novelty.

 

[sitehostplus]

The more I think about it, the more it sounds like pure genius.

With the right architecture, these SSD's could be a way for the user to boost the graphics memory on the card.

If nothing else, it reduces latency by a boatload.

 

[palladin9479]

I can see this happening in some niche scenarios, mostly because newer NVME manufactures are putting these heatsinks on them that make it hard to fit them on the board. The back of the GPU has plenty of room.