NVIDIA Releases GeForce MX450 with PCI-Express 4.0 Interface

[btarunr]

NVIDIA released a mysterious new mobile GPU that has us scratching our heads over the silicon that could be driving it. The new GeForce MX450 is an entry-mainstream mobile GPU that apparently ships with a PCI-Express gen 4.0 bus interface, something only NVIDIA's "Ampere" GPUs feature. The product page for the MX450 doesn't list out any other specs, than its memory type support including new GDDR6 memory (supported only on NVIDIA architectures "Turing" or later). Interestingly, it also lists GDDR5 as one of its memory options. PCI-Express 4.0 is prominently listed as one of its specs.

Upon digging some more among device IDs, we've come across the ID of the GDDR5 variant, with the ASIC code "GP107-670-A1," and the silicon is based on the much older "Pascal" architecture, which lacks PCIe gen 4 support. The GDDR6 variant eludes us. This is the SKU which could be based on a newer architecture, given its support for GDDR6 and PCIe gen 4. NVIDIA's GeForce MX line of entry-mainstream mobile GPUs are built to performance/power targets, and wildly vary with the underlying tech. They've been historically a means for NVIDIA to clear inventory of older generation ASICs to notebook manufacturers, who get put the NVIDIA logo on their products, and advertise discrete graphics. Given this, the use of a newer (even unreleased) generation of GPUs comes as a surprise.



View at TechPowerUp Main Site

 

[Verpal]

Pcie 4.0?

Sounds like left over Volta to me.

 

[ExcuseMeWtf]

A bit weird place to start, given TPU's own testing shows astounding 2% difference between even PCI-E 2.0 and 4.0 when using 5700XT.

But hey, gotta start somewhere I suppose...

 

[Vayra86]

A bit weird place to start, given TPU's own testing shows astounding 2% difference between even PCI-E 2.0 and 4.0 when using 5700XT.

But hey, gotta start somewhere I suppose...

2%!!! That's the difference between Solitaire and Minesweeper playable, man!

 

[medi01]

A bit weird place to start, given TPU's own testing shows astounding 2% difference between even PCI-E 2.0 and 4.0 when using 5700XT.

But hey, gotta start somewhere I suppose...

Others found it to be 4-5% when going from PCIE 3 and 4, but with AMD 39xx CPU.

PS
On the subject, given how amazing the 4xxx series mobile AMD chips are, NV needed to boost MX line, to stay relevant.
Although, I"m afraid, that relevance is still related only to the, at this point, hilariously overpiced Intel chips.

 

[BorisDG]

Knowing 2080 Ti barely moved saturation from x8 to x16, I doubt 3090 will fill up the entire x16. My guess is RTX 4000/5000 still won't be bottlenecked in 3.0. PCIE 4.0 for now will be for some really highend SSDs.

 

[W1zzard]

Pcie 4.0?

Sounds like left over Volta to me.

It's just marketing value to be aboard the hype train when Tiger Lake laptops come out this year

 

[Kaotik]

Pcie 4.0?

Sounds like left over Volta to me.

Left over Volta? Aside the fact that Volta supports neither any sort of GDDR-memory or PCIe4, there's only one discrete Volta-based GPU and that V100 is huge.

 

[olstyle]

It's just marketing value to be aboard the hype train when Tiger Lake laptops come out this year

Or maybe Tiger Lake will cut the lanes in the laptop segment to compensate for PCIe 4.0 power draw. So it will be a 8 or even 4 Lane connection.

 

[W1zzard]

Or maybe Tiger Lake will cut the lanes in the laptop segment to compensate for PCIe 4.0 power draw. So it will be a 8 or even 4 Lane connection.

Correct it could be x4 4.0, possibly even through some kind of bridge chip that converts it back to x8 3.0

 

[R-T-B]

Others found it to be 4-5% when going from PCIE 3 and 4, but with AMD 39xx CPU.

Yeah, that's what TPU tested with. Care to validate this vague reference to, at the moment, nothing?

 

[king of swag187]

 

[watzupken]

Correct it could be x4 4.0, possibly even through some kind of bridge chip that converts it back to x8 3.0

My current Lenovo laptop with the MX350 is using just PCI-E 3.0 x4. I believe most other manufacturers will do the same due to limited bandwidth, and another benefit may be power savings. All these low end graphics don't require high bandwidth in the first place.