Nvidia has outdone itself this time, even by the chipmaker's own standards. AIDA64 (via @momomo_us) has just added support for two new unreleased variants of the GeForce GTX 1650 graphics card, increasing the number of different versions to four.
The original GeForce GTX 1650, which landed last year, debuted with the TU117 silicon. The graphics card's 896 CUDA cores were complemented by 4GB of 8 Gbps GDDR5 memory across a 128-bit memory interface.
A little less than a year later, Nvidia rolled out the GeForce GTX 1650 GDDR6, and, as the name implies, the graphics card came with faster 12Gbps GDDR6 memory. Although the GDDR6 variant features lower clock speeds, the swap to GDDR6 memory allowed the graphics card to deliver up to 50% higher memory bandwidth. The upgrade contributed to a real-world performance improvement of up to 14% over the vanilla version.
AIDA64's latest changelog shows that Nvidia isn't quite done with the GeForce GTX 1650 yet.
Nvidia GeForce GTX 1650 Specifications
|Header Cell - Column 0||GeForce GTX 1650 TU106*||GeForce GTX 1650 TU116*||GeForce GTX 1650 GDDR6||GeForce GTX 1650|
|GPU Architecture||Turing (TU106)||Turing (TU116)||Turing (TU117)||Turing (TU117)|
|Base Clock Rate||1,410 MHz||1,410 MHz||1,410 MHz||1,485 MHz|
|Boost Clock Rate||1,590 MHz||1,590 MHz||1,590 MHz||1,665 MHz|
|Memory Clock||12 Gbps||12 Gbps||12 Gbps||8 Gbps|
|Memory Capacity||4GB GDDR6||4GB GDDR6||4GB GDDR6||4GB GDDR5|
|Memory Bandwidth||192 GBps||192 GBps||192 GBps||128 GBps|
|Transistor Count||10.8 billion||6.6 billion||4.7 billion||4.7 billion|
|Die Size||445 mm²||284 mm²||200 mm²||200 mm²|
*Specifications are unconfirmed.
AIDA64 refers to the two new variants as the GeForce GTX 1650 TU116 and GeForce GTX 1650 TU106. It's pretty much the same formula as the GDDR6 version that denotes the type of memory in the model name. On this occasion, the suffixes designate the silicon that's used inside each refreshed model.
In the case of the GeForce GTX 1650 TU116, the TU116 die is the same one that Nvidia employs in the the other GeForce GTX 16-series SKUs, including the GTX 1650 Super, GTX 1660, GTX 1660 Super and GTX 1660 Ti.
The TU106 die, on the other hand, is a very interesting selection, and the one that raises many questions. For context, the TU106 is present inside Nvidia's GeForce RTX 20-series models, such as the GeForce RTX 2060, RTX 2060 Super and RTX 2070. Unlike the TU116 and TU117 silicons, the TU106 does house Nvidia's Tensor and RT cores. Has Nvidia finally caved in and decided to bring ray tracing to a GTX 16-series graphics card? Unfortunately, we don't know for the time being. AIDA64 didn't give any hints on the specifications. Nonetheless, the base specifications should be the same for both new variants.
The new versions of the GeForce GTX 1650 probably aren't a coincidence. Nvidia is likely just reprocessing defective dies that don't meet the requirements for higher tier models and using them in the GeForce GTX 1650. Alternatively, the chipmaker might just be getting rid of leftover Turing silicon to make space for Ampere, which is heavily rumored to debut in September. In any event, it's easy to see why Nvidia chose the GeForce GTX 1650 in particular.
According to the Steam Hardware Survey for May 2020, the GeForce GTX 1650 was the fifth most popular graphics cards among Steam gamers. If there was a fast way to dispose of unused silicon, the GeForce GTX 1650 would be the best candidate.
You must have missed the various versions of the GT 730, and GTX 1060. lol
NOW I REMEMBER THEM AGAIN!
The butcher can make them out of whatever they have and the crowd will buy them because they are fun and priced right.
I could understand it if they wanted some entry-level quadro card with RTX or Tensor enabled, or maybe they are playing with the idea of some kind of RTX 2050/2030... But calling it a 1650 without also finding a way to squeeze more out of the silicon for a higher asking price? That isn't really passing the smell test to me.
But if it does exist at the same price point, that is at least possibly exciting for overclocking or maybe even firmware hacking.
Kind of proves the point I was making months ago that GPUs only cost as much as they currently do because we have a stagnant duopoly situation where both vendors have implicitly agreed on arbitrarily inflated prices.
Also, i'm pretty sure nVidia will hike their prices again for Ampere, so lowering the price now for the RTX models would work against them.
Even the 1650 GDDR6 (non-Super) is matched/slightly exceeded by the older and cheaper RX 570.
If the use of TU106/TU116 somehow boosts its performance enough, AND they keep it priced lower than the 1650 Super, down at RX 570 levels or below, then it'll sell well.
The confusing naming for multiple slightly different products might help them rake in a few bucks, but as things stand, the 1650 Super is the only place where Nvidia is outdoing the archaic AMD cards in price/performance. I'm curious to see what the performance of these new 1650 variants are like, and what steps, if any, Nvidia takes to make sure a customer will know what performance level/which 1650 they're getting.