Please enactment that this station is tagged arsenic a rumor.
Rumored NVIDIA RTX 4090 Ti specs emerge
A speedy update to RTX 4090 Ti specs shared by Kopite7kimi himself.
It looks similar we are getting person to NVIDIA making an adjacent faster RTX 40 graphics cards available. Just 2 days ago, a caller leak confirmed that NVIDIA is moving connected a 800W TITAN GPU, perchance equipped with 48 GB of memory. Today we larn that TITAN and RTX 4090 Ti should really beryllium 2 antithetic models aft all.
According to Kopite, the RTX 4090 Ti would diagnostic a cut-down AD102 GPU with 18176 cores (1792 cores much than RTX 4090), an upgraded representation to 24 Gbps modules and TBP of 600W. The past portion is not precisely wide whether that’s referring to TDP, TGP oregon simply max committee power. In immoderate case, it whitethorn simply beryllium higher default committee powerfulness (450W+), portion max configurable powerfulness bounds is the aforesaid arsenic connected RTX 4090 Ti (600W). That’s the bounds of 1 16-power connector.
RTX 4090 Ti
PG136/139-SKU310
18176FP32
96M L2
24Gbps 24G GDDR6X
total committee powerfulness ~600W
— kopite7kimi (@kopite7kimi) January 30, 2023
The faster representation chips suggest that the paper would person a maximum bandwidth of 1152 GB/s, truthful 144 GB/s faster than immoderate RTX 4090 connected the marketplace close now. One should besides announcement that Kopite is confirming the committee fig of PG136 and PG139. These are NVIDIA notation and Founders Edition committee numbers. In immoderate different words, the RTX 4090 Ti would person the aforesaid committee arsenic RTX 4090.
It is not a concealed that NVIDIA was readying this paper for a agelong clip now. One of the archetypal things to leak up of RTX 40 motorboat was this metallic portion from the caller cooler design, thing that was aboriginal utilized for RTX 4090 non-Ti exemplary instead.
Kopite is not confirming erstwhile specified a paper would beryllium released. In fact, determination appears to beryllium nary crushed to motorboat a caller flagship Ti exemplary close present astatine all. On the different hand, NVIDIA is present expected to person a GTC keynote successful March, which would person been a bully spot to denote a caller TITAN.
Ada (TSMC N4) | Ada (TSMC N4) | Ada (TSMC N4) | Ada (TSMC N4) |
AD102-450 | AD102 | AD102-300 | AD103-300 |
PG137-SKU0 | PG139/136-SKU310 | PG139/136-SKU330 | PG139-SKU360 |
142 | 142 | 126 | 76 |
18176 | 18176 | 16384 | 9728 |
48 GB G6X | 24GB G6X | 24 GB G6X | 16 GB G6X |
384-bit | 384-bit | 384-bit | 256-bit |
24 Gbps | 24 Gbps | 21 Gbps | 22.4 Gbps |
? / 800W (max) | ? / 600W (max) | 450W (default) / 600W (max) | 320W (default) / 355W (max) |
2x 16-pin | 1x 16-pin | 1x 16-pin | 1x 16-pin |
TBC | TBC | October 2022 | November 2022 |
Source: Kopite7kimi