Nvidia GeForce GTX 1650 Ultra spotted with cut down TU106 GPU

by Mark Tyson on 30 June 2020, 10:21

Tags: NVIDIA (NASDAQ:NVDA), GALAX

Quick Link: HEXUS.net/qaemns

Add to My Vault: x

Chinese graphics card vendor SZ Galaxy has listed a previously unannounced graphics card dubbed the GeForce GTX 1650 Ultra. Twitter users Momomo_us discovered the new SKU, which joins the legion of GTX 1650 variants out there, but this one has something rather surprising under the bonnet - it is a new model based upon the TU106 GPU, the same GPU used for the original RTX 2070.

There are a lot of GTX 1650 variants out there already, a user replied to Momomo_us's Tweet to list 10 designs, including the new Ultra - there are permutations such as the Super, Mobile, Max-Q, Ti, and GDDR6 editions.

So, what is the GeForce GTX 1650 Ultra, and why does it exist? According to the official product page, this Galax branded card is powered by the TU106-125 GPU. This is a cut down version of the GPU behind the original RTX 2070, which was the TU-106-400. The key thing here is that the GTX 1650 Ultra's GPU has significantly fewer CUDA cores than a RTX 2070's GPU would provide; 896 compared to 2304.

  • TU106-125 GPU
  • 896 CUDA cores
  • 1,590MHz GPU boost clock
  • 4GB GDDR6 VRAM on 128bit bus
  • 12Gbps memory performance

The cores/clocks/memory configuration makes the GTX 1650 Ultra's nearest neighbour the GTX 1650 GDDR6 version (based upon the TU117-300 GPU). All the specs bullet pointed above align with this SKU, however the GTX 1650 GDDR6 has a TDP of 75W, compared to this Ultra model's 90W.

It is quite common for graphics card makers to make use of GPU dies that aren't up to snuff for the intended target product for lower tier products. In this case the TU106-125 GPU doesn't need functioning RTX or Tensor cores, and it has fewer than 60 per cent of the die's CUDA cores in action. Repurposing these GPUs helps minimise waste materials and production cost losses.

It will be interesting to see the performance of the GeForce GTX 1650 Ultra when these cards become available, but nothing in the spec suggests 'Ultra' performance, or any boost narrowing the margin between standard GTX 1650 and its 'Super' variants. Some may be irked by the use of the 'Ultra' suffix here, as one might consider 'ultra' to be an improvement upon 'super'. The GTX 1650 Super, is a completely different kettle of fish though, with its TU116-250 GPU boasting 1,280 CUDA cores and running at a boost clock of 1,725MHz out of the box.



HEXUS Forums :: 3 Comments

Login with Forum Account

Don't have an account? Register today!
Spring has sprung, and nvidia mid range GPUs have returned to their spawning grounds. The first few baby SKUs have already shown up, and soon we'll be up to our necks in them again
I'd be fine with the cut down CUDA cores as long as it has a proper Turing NVENC chip, that'd make it an ideal host graphics card for an UnRAID server.
good lord! Nvidia flooding the market with multiple gpus variations to create artificial “competition” and distort the fact they doubled prices in the last couple of years to fill their greedy hands with $$$. Can't wait for their £2k 3080ti. Nvidia is turning into Intel of gpus. Zero innovation (please don't even start with ray tracing), incremental updates at best all to milk customers for their $$$, that's what happens when you have a company run by greedy investors who couldn't care less about what the company is producing or (should be) representing.