Nvidia's Turing microarchitecture is a little long in the tooth. Still, the chipmaker isn't ready to leave Turing behind. According to VideoCardz, Nvidia will reportedly give Turing the last ride in the shape of the GeForce GTX 1630, rumored to launch on May 31.
The GeForce GTX 1630 would mark the first time that Nvidia has relinquished the GTX branding to a lower-tier SKU. The x30-series, such as the GT 730 or GT 1030 models, have typically assumed the GT moniker. The model name alone is enough to tell you that the GeForce GTX 1630 is an entry-level graphics card that slots right below the GeForce GTX 1650, the previous rock-bottom GTX offering. The new Turing graphics card will share the battlefield with AMD's Navi 24-powered models, such as the Radeon RX 6500 XT and Radeon RX 6400.
The leaked specifications, which you should approach with caution, claim that the GeForce GTX 1630 will wield Nvidia's TU117 silicon, which sounds plausible. The TU117 is the smallest Turing die, and Nvidia has used variants for its other SKUs, such as the GeForce GTX 1650 or GeForce MX550. Nvidia will play surgeon with the TU117 die inside the GeForce GTX 1630 to avoid cannibalization within its ranks.
The TU117 has 16 streaming multiprocessors (SMs), equivalent to 1,024 CUDA cores. The GeForce GTX 1650 has 14 enabled SMs (896 CUDA cores), meaning the GeForce GTX 1630 will have far less. VideoCardz believes that the GeForce GTX 1630 only comes with 8 SMs or 512 CUDA cores. That's precisely half of what TU117 offers and 43% less than what the GeForce GTX 1650 has onboard.
Nvidia GeForce GTX 1630 Specifications
|Header Cell - Column 0||Radeon RX 6400||GeForce GTX 1650||GeForce GTX 1630*|
|Process Technology||TSMC N6||TSMC 12FFN||TSMC 12FFN|
|Die size (mm^2)||107||200||200|
|SMs / CUs||16||16||8|
|Base Clock (MHz)||1,923||1,485||?|
|Boost Clock (MHz)||2,321||1,665||1,800|
|VRAM Speed (Gbps)||16||8||12|
|VRAM Bus Width||64||128||64|
|TFLOPS FP32 (Boost)||3.5||2.9||1.8|
|Launch Date||Jan 2022||Apr 2019||May 2022|
*Specifications are unconfirmed.
The GeForce GTX 1630 could sport a 1,800 MHz boost clock speed, but it's unknown whether the quoted value is the reference specification or an aftermarket model with a factory overclock. Single-precision numbers aren't the best metric to measure gaming performance. For comparison sake, the GeForce GTX 1650 delivers up to 61% higher FP32 performance than the GeForce GTX 1630, whereas the Radeon RX 6400 takes the margin up to 94%. Early reviews showed the Radeon RX 6400 on par with the GeForce GTX 1650 in terms of performance, so the GeForce GTX 1630 is likely slower than the Navi 24 graphics card. The question is just how much slower.
It's comforting to know that the GeForce GTX 1630 will reportedly wield 4GB of 12 Gbps GDDR6 memory. The sad part is that Nvidia will likely restrict it to a 64-bit memory interface. Therefore, the GeForce GTX 1630 can only offer memory bandwidth of up to 96 GBps, 25% lower than the GeForce GTX 1650 and Radeon RX 6400. The original GeForce GTX 1650 debuted with 8 Gbps GDDR5 memory, but Nvidia subsequently released three new variants with 12 Gbps GDDR6, bumping the memory bandwidth from 128 GBps to 192 GBps. So if we compare the GeForce GTX 1630 to the GDDR6 alternatives, we're looking at 50% less memory bandwidth.
Despite having fewer CUDA cores, the GeForce GTX 1630 may retain the same 75W TDP as the faster GeForce GTX 1650. The extra headroom is probably why the GeForce GTX 1630 could feature higher boost clock speeds. However, the GeForce GTX 1650 looks demanding besides the Radeon RX 6400's 53W rating.
The GeForce GTX 1630's exact MSRP is still up in the air, but the Turing-based graphics card will likely have a sub-$150 price tag. The GeForce GTX 1650 and Radeon RX 6400 have a $148 and $159 MSRP, respectively, so the GeForce GTX 1630 will have to go very low to compete. However, we would love to see the GeForce GTX 1630 retail between $100 to $120 since we haven't seen a GeForce GTX graphics card in that price range since Pascal (GeForce GTX 1050 and GTX 1050 Ti).
Side note: Now that I'm reminded of the multiple GTX 1650 GDDR6 variants - was it ever determined if the TU106, TU116, and TU117 versions of the 1650 GDDR6 all performed the same? Or did any variant have a particular advantage?
I was curious myself how the various 1650 compare. GDDR6 models have faster memory but lower clock speeds than the GDDR5 version.
If the reviews of the new Xe graphics cards are anything to go by, Intel drivers are horrendous. A lot of game titles just won't launch or they have terrible render issues and glitches. I hope they can figure it out though. I am mildly interested in their lowest end card to replace my GT1030 so I can have an HDMI 2.1 port without getting an RTX card or an RX 6400.
If it has a 75 watt TDP, it may be a bit pickier about the low-wattage power supplies found in some prebuilt systems compared to the 20-30 watt GT 1030, and cooling might also be more of a concern under load.
1030 was also last card to be featured in the upgradable laptop format size. I wonder if 1630 will be as well. We would be able to make interesting frankenstein laptops....
Well, the GDDR6 version definitely outperformed the GDDR5 version, despite the lower clock speeds. My thought is that the impetus to move to GDDR6 was because of the embarrassment of the 1650 GDDR5 underperforming the 1060 3GB (where the amount of VRAM wasn't an issue), and definitely under the RX 570 4GB.
The update to GDDR6, along with lower clock speeds, brought it up to nipping at the 570 4GB's heels. That definitely suggests that the 1650 was choked on VRAM bandwidth.
As to the 3 different variants of the 1650 GDDR6, I'm curious as to whether there's any real world difference in performance given the different chips involved.