Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
Nvidia GPU RTX 4xxx Lineup
  • Will be called "Ada Lovelace"

    AD102 to of the line chip

    • 12 GPC
    • 72 TPC
    • 144 SM
    • 18'432 FP32 units
    • 1.8Ghz
    • ~66 TFlops FP32 performance
    • Samsung production
  • 13 Replies sorted by
  • 3DCenter reports that Ada Lovelace GPUs in the GeForce RTX 40 series of video cards will receive a TDP of 424-450 watts.
    For comparison, the GeForce RTX 3070 Ti and RTX 3080 cards have values of 290-320 watts, respectively. The GeForce RTX 3080 Ti and RTX 3090 have a TDP of 350 watts. Past rumors have stated that the AD102 GPU will be manufactured in the 5nm process at TSMC's facilities. Its area will be approximately 600 mm2 against 628 mm2 for GA102. There may be 18432 CUDA cores here versus 10496 for the RTX 3090.

  • The next generation Nvidia Ada Lovelace AD102 GPU will be clocked at 2.2GHz. This chip will be included in the flagship video cards and will replace the existing GA102.

  • image

    image

    sa18754.jpg
    640 x 387 - 29K
    sa18755.jpg
    607 x 389 - 35K
  • Partners of NVIDIA, including among Taiwanese manufacturers of semiconductor components, have already begun preparations for the start of production of a new generation of NVIDIA gaming video cards based on the Ada Lovelace architecture. Their appearance is expected in the second half of 2022.

    According to the latest rumors, the flagship NVIDIA AD102 GPU (Ada Lovelace) will be clocked at up to 2.5GHz. It is credited with 18,432 CUDA cores comprising 144 SM blocks. This is twice as much compared to current generation graphics cards. Computing power (FP32) is expected at 85-92 teraflops, which is more than double that of the current flagship GeForce RTX 3090.

    According to the leaks, the GeForce RTX 4090 will be asked for about $ 3000. Considering that it is hardly possible to cope with the deficit in the current realities, such accelerators will be sold on the market for $ 5000-8000. GeForce RTX 4080 Ti will cost about $ 2000, GeForce RTX 4080 - $ 1200, GeForce RTX 4070 - $ 800.

  • I was looking for a new LGA1700 build for video editing.

    DDR5 RAM is not available yet.

    And the big dealers don't have a single GeForce card in stock.

  • Can't wait to see all the new cards that will be unavailable.

  • Insider Greymon55 confirmed the old assertion about the power of high-end RTX 40 series graphics cards based on the AD102 chip. A late 2020 leak suggested that cards with the Lovelace AD102 GPU could achieve 66.4 teraflops in FP32 mode at 1.8GHz. A later statement made by Greymon55 tells us that the card with the AD102 GPU could offer 85 to 92 teraflops at 2.3 to 2.5 GHz. Now, the same insider has posted a reminder of this, stating, "If Lovelace is overclocked to 2.5GHz, you'll get about 90 teraflops of FP32."

  • Main progress now is just to add more stuff:

    image

    sa19500.jpg
    673 x 185 - 25K
  • According to industry sources, Nvidia has shared its flagship RTX 4000 series GPU reference boards, which may be named RTX 4080, RTX 4080 Ti, RTX 4090, and RTX 4090 Ti, with its partners. The graphics processors themselves have not yet been delivered, but it is known that the boards are designed for consumption up to 600 watts.

    Several well-known insiders have recently stated that they expect the RTX 4080 and 4090 TBP to be at 600W. The higher end RTX 4090 Ti model can draw over 800W.

    Igor's Lab reports that Nvidia has asked manufacturers to test reference boards used to design their own PCBs with a total power of 600W. Not every RTX 4080 or RTX 4090 graphics card will necessarily draw 600W, but at least some of them will run at higher clock speeds.

    The board has a whopping 24 voltage converters, compared to 20 on the RTX 3090 Founder's Edition PCB. All of these converters and modules are powered by a single 12VHPWR connector, also known as a PCIe 5.0 power connector, that can deliver 600W. Igor's Lab notes that manufacturers are designing PCBs that use the full 600W, but this is the maximum they can do, since the PCIe 5.0 slot has just such a limitation. Each 12VHPWR connector comes with an adapter with four 6+2-pin Molex connectors, although most power supplies don't have that many cables.

  • RTX 4090 is 60% faster than RTX 3090 Ti and 77.5% than RTX 3090 in the same CUDA test. That’s indeed massive improvement for this type of test alone. Naturally, we cannot translate this performance to gaming, that’s an entirely different subject that will involve more than just raw GPU power.

  • image

    sa7327.jpg
    800 x 397 - 35K
  • Excellent on pricing and capitalist greed