The GeForce RTX 5090 on a 512-bit bus, the RTX 5080 half as powerful?


Nerces

Hardware and Gaming Specialist

March 13, 2024 at 11:23 a.m.

15

Hypothetical representation of a GeForce RTX 50 © VideoCardz

Hypothetical representation of a GeForce RTX 50 © VideoCardz

The next generation of graphics cards NVIDIA is starting to make a name for itself: the RTX 5090 could be clearly above the rest.

Will you hear some of the rumors surrounding the next generation of NVIDIA graphics cards? A few months before their supposed launch, the rumor machine is racing.

The latest concerns the interface bus which determines the maximum authorized memory bandwidth, guaranteeing rapid exchanges between the graphics processor and video memory.

No 384-bit RTX 50 card?

At the beginning of the month, JEDEC finally validated the technical specifications of the future GDDR7, video memory expected to be used by NVIDIA on its GeForce RTX 50 series.

It was enough to trigger a wave of publications around the memory subsystem of these upcoming GeForces. Thus, at the very beginning of the week, it was the indestructible Kopite7kimi who intervened to discuss the memory bus used on the future spearhead of the range, the GeForce RTX 5090 and its GB202 GPU, a 512-bit interface just that.

Remember that the GeForce RTX 4090 only benefited from a 384-bit bus, a sign that NVIDIA should greatly boost its next “beachhead”. Not to spoil anything, the GB202 should integrate 192 stream multiprocessors while the A102 of the RTX 4090 made do with 144.

The RTX 5080 half as powerful as the 5090?

In the wake of his previous revelations – and although he each time expresses caution, “NVIDIA can still change certain data” – the same Kopite7kimi gave details concerning the GB203.

A lower-ranking graphics processor, it should equip future GeForce RTX 5080s and, logically, confirm a certain move downmarket. This would, however, be more brutal than on the RTX 40 series. Indeed, Kopite7kimi indicates that the memory interface bus would only be 256-bit on this RTX 5080, as much as on the RTX 4080.

The gap between RTX 5090 and RTX 5080 would therefore be greater than between RTX 4090 and RTX 4080. The number of stream multiprocessors confirms this hypothesis: there would be “only” 96 on the RTX 5080, half as many than on the RTX 5090. Remember that we went from 84 to 144 between the RTX 4080 and the RTX 4090.

Supposed specs of the GeForce RTX 50 series lineup © VideoCardz

Supposed specs of the GeForce RTX 50 series lineup © VideoCardz

Just for good measure, Kopite7kimi takes the opportunity to slip in some information about the GB205 of the future RTX 5070: with a 192-bit memory bus, it is at the level of that of the RTX 4070.

Source : VideoCardz



Source link -99