Rumors about Nvidia graphics cards: Pictures show alleged RTX 4090 Ti or Titan Ada

0
88

Newly emerged images of a four-slot high slot cover with unusually oriented connectors are fueling new rumors about a potential GeForce RTX 4090 Ti or a new Nvidia Titan based on the AD102 graphics chip. Again, there is talk of a power consumption of up to 800 watts.

Speculations about a supposed superlative Ada Lovelace graphics card are not new: rumors about a new graphics accelerator of the Titan class were already circulating in April 2022, which was last used at the end of 2018 by the Titan RTX (test) based on the Turing architecture. In this context, higher and higher TDP classifications of up to 900 watts were mentioned over the course of the year. The GeForce RTX 4090 (test), which was finally released in October, proved to be much more economical, but does not rely on the full expansion of the AD102 graphics chip. The rumors and ultimately the wishful thinking about an even more powerful gaming graphics card with Ada Lovelace architecture continued under the name GeForce RTX 4090 Ti.

A graphics card with a vertical PCB

Images now circulated by Twitter user @Zed__Wang aka MEGAsizeGPU are fueling speculation again. You can see the four-slot high slot bezel of a presumed new Nvidia top model, mounted on an equally large heat sink in one of the pictures. Three aspects are now attracting attention:

  1. The Founders Edition of the GeForce RTX 4090 is only three slots high with a TDP of 450 watts and the heat sink can do that easily dissipate power loss. The use of a four-slot high cooler therefore indicates a significantly higher power consumption.
  2. The heat sink shown has a golden coloration, which is known in this form from Nvidia's professional compute solutions, but has also adorned Titan graphics cards in the past.
  3. Three DisplayPort connections and an HDMI socket indicate that it is actually a desktop graphics card. Their arrangement is interesting: The stacked sockets indicate a vertically oriented PCB, which has already been discussed in the past for an Ada Lovelace top model and supported by recordings, as VideoCardz notes. A conceivable advantage of the unusual design would be an optimized air flow.
Slot cover of an alleged GeForce RTX 4090 Ti or Titan (image: MEGAsizeGPU)
Massive heat sink of a suspected GeForce RTX 4090 Ti or Titan (Image: Chiphell)
Slot panel of a presumed GeForce RTX 4090 Ti or Titan (image: MEGAsizeGPU)

The assumption that Nvidia would use an AD102-450 for such a graphics card is still current. Previous rumors certify the rumored graphics chip a total of 142 streaming multiprocessors and thus also 142 raytracing cores and 18,176 FP32 execution units, although it is still just not the AD102 full expansion.

A hybrid of GeForce RTX 4090 and Nvidia L40?

Meanwhile, GPUs with those specifications are already in use with the Nvidia RTX 6000 professional graphics card and the Nvidia L40 and L40G server graphics cards. The latter offers up to 90 TFLOPS FP32 computing power and provides the hardware for the RTX 4080 SuperPods, which provide GeForce Now Ultimate with the performance of the eponymous desktop graphics card. However, a GeForce RTX 4090 Ti or Titan Ada would differ fundamentally from those products in two respects.

  1. While the professional solutions have a maximum TDP specification of 300 watts, the GeForce RTX 4090 Ti is traded with a maximum power limit of 800 watts. With the GeForce RTX 4090 with AD102-300, this is 600 watts, but is usually not exhausted even with overclocking, while the TDP is “only” 450 watts – the GeForce RTX 4090 Ti should behave similarly. In addition, the difference can be attributed to higher clock rates and different memory compared to the professional solutions.
  2. Speaking of memory: Instead of GDDR6 (ECC), the GeForce RTX 4090 Ti should use GDDR6X type graphics memory, which is also offered by the GeForce RTX 4000 graphics cards presented to date. For the first time, however, components with a throughput of 24 GB/s are used. Manufacturer Micron does list the corresponding memory chips, but they have not been used in the RTX 4000 so far. With a 384-bit interface, the bandwidth would increase to 1,152 GB/s.

It is also worth noting that the GeForce RTX 4090 Ti is rumored to have 48 GB of GDDR6X graphics memory, which in turn is another explanation for the noticeably higher power limit in relation to the GeForce RTX 4090 with 24 GB of GDDR6X. The editors have compiled the alleged specifications of the rumored Ada Lovelace top model in the following table on the basis of information obtained in the past, the data sheet of the Nvidia L40 and projections based on GeForce graphics cards that have already been released.

Specifications of selected Nvidia graphics cards Nvidia L40 RTX 4090 Ti/Titan* RTX 4090 RTX 4080 RTX 4070 Ti RTX 3090 Ti Architecture Ada Ampere GPU AD102-895 AD102-450 AD102-300 AD103-300/301 AD104-400 GA102-350 Manufacturing TSMC N4 Samsung 8nm transistors 76.3B 45.9B 35.8B 28B Chip size 608mm² 379mm² 295mm² 628mm² SM 142 128 76 60 84 FP32 ALUs 18,176 16,384 9,728 7,680 10,752 RT cores 142. 3rd Gen 128, 3rd Gen 76, 3rd Gen 60, 3rd Gen 84, 2nd Gen Tensor Cores 568, 4th Gen 512, 4th Gen 304, 4th Gen 240, 4th Gen 336, 3rd Gen Base Clock 735MHz ? 2230MHz 2210MHz 2310MHz 1560MHz Boost Clock 2490MHz ? 2520MHz 2510MHz 2610MHz 1860MHz FP32 processing power 90.5 TFLOPS 82.6 TFLOPS 48.8 TFLOPS 40.1 TFLOPS 40.0 TFLOPS FP16 processing power 90.5 TFLOPS 82.6 TFLOPS 48.8 TFLOPS 40.1 TFLOPS 40.0 TFLOPS FP16 over tensor ? 330 Tflops 195 Tflops 164 Tflops 156 Tflops Texture units 568 512 300 336 ROPS 192 80 112 L1-Cache 18.176 KB 16.384 KB 7.680 KB 10.752 KB L2-Cache 96 MB 66 MB 6 MB storage 48 GB GDDR6 (EEC) 48 GB GDDR6X 24 GB GDDR6X 16 GB GDDR6X 12 GB GDDR6X 24 GB GDDR6X Memory throughput 18 Gbps 24 Gbps 21 Gbps 22.4 Gbps 21 Gbps Memory interface 384-bit 256-bit 192-bit 384-bit Memory bandwidth 864 GB/s 1152 GB/s 1152 GB/s8 GB/s 717 GB/s 504 GB/s 1,008 GB/s Slot connection PCIe 4.0 ×16 video engine ? 2x NVENC (8th Gen)
1x NVDEC (5th Gen) 1x NVENC (7th Gen)
1 × NVDEC (5th Gen) TDP 300 Watts < 800 watts 450 watts 320 watts 285 watts 450 watts *Not confirmed

Such a GeForce RTX 4090 Ti or Titan Ada would offer a raw performance increase of around 10-15 percent compared to a GeForce RTX 4090 – depending on how much the faster memory comes into play. It should be noted, however, that a GeForce RTX 4090 already has a 70 percent increase in raw performance compared to the GeForce RTX 4080 (test) on paper, but even in the editors' UHD performance rating it can only set itself apart by around 35 percent – half of the raw power falls by the wayside. With a GeForce RTX 4090 Ti it would probably be no different compared to the GeForce RTX 4090.

Information should be treated with caution

And there follows another note that is obligatory for such reports: Ultimately, the rumors about the GeForce RTX 4090 Ti or Titan Ada are by no means reliable information that should be treated with particular caution. A graphics card that will be released in the future can certainly be seen in the images at hand – but it is also possible that they are prototypes based on an old concept that Nvidia has since rejected.

It is also doubtful that Nvidia will follow a top gaming model that has only been on the market for a few months with a more powerful variant, although the performance leap of the GeForce RTX 4090 was already huge and the graphics card is unrivaled anyway. After all, there was a year and a half between the market launch of the GeForce RTX 3090 (test) and the GeForce RTX 3090 Ti (test).

Overview of rumors about future ones Nvidia graphics cards

  • RTX 4070, 4060 Ti & 4090 Ti: Rumors speak of two more AD104 graphics cards
  • Nvidia graphics cards rumors: GeForce RTX 4070 Ti, 4070 and 4060 Ti become more concrete
  • Nvidia GeForce RTX 4070: Rumors speak of 5,888 ALUs and 250 watt TDP
  • Nvidia GeForce RTX 4060 Ti: rumors speak of 3070 performance for USD 500
  • GeForce RTX 4060: rumors circulating are unfounded