Unreleased RTX 3050 Ti engineering sample appears in photos and benchmarks — the RTX 3060 alternative that never happened

GeForce RTX 3050 Ti
(Image credit: X/GOK)

Nvidia’s GeForce RTX 30 series (codenamed Ampere) firmly established itself among the best graphics cards of its time. Interestingly, it seems that Nvidia may have left some money on the table with Ampere. The latest photographs and benchmark results shared by hardware leaker Gok clearly show that the GeForce RTX 3050 Ti is real and has at least reached the prototype stage.

Nvidia did bring a GeForce RTX 3050 Ti to the market, but only for laptops, catering primarily to mobile gamers. What we're seeing today is the desktop version, which should not be confused with the GeForce RTX 3050 Ti Mobile or the GeForce RTX 3050 Ti Max-Q for laptops. Ampere was a popular and successful architecture for Nvidia. However, the downsides of Ampere's launch during the great graphics card shortage, first caused by the cryptocurrency mining boom and second by the COVID-19 pandemic, probably deterred Nvidia from releasing the GeForce RTX 3050 Ti.

The GeForce RTX 3050 Ti engineering sample, reportedly from a company named Robiny, leverages the PG190 SKU 40 design board. It employs the GA106 silicon, the same die utilized in other popular Ampere products, including the GeForce RTX 3050 and the GeForce RTX 3060. As you can tell from the model name, the GeForce RTX 3050 Ti slots in between the two aforementioned models. Once again, it's just an example of Nvidia's strategy to recycle high-end silicon to serve different market segments.

Latest Videos From

Nvidia GeForce RTX 3050 Ti Specifications

Swipe to scroll horizontally

Graphics Card

GeForce RTX 3060

GeForce RTX 3050 Ti*

GeForce RTX 3050

Architecture

GA106

GA106

GA106

Process Technology

Samsung 8N

Samsung 8N

Samsung 8N

Transistors (Billion)

12

12

12

Die size (mm²)

276

276

276

SMs

28

26

20

GPU Cores

3,584

3,328

2,560

Tensor Cores

112

104

80

RT Cores

28

26

20

Base Clock (MHz)

1,320

1,410

1,552

Boost Clock (MHz)

1,777

1,665

1,777

VRAM Speed (Gb/s)

15

14

14

VRAM (GB)

12

6

8

VRAM Bus Width

192

192

128

ROPs

48

48

48

TMUs

112

104

80

Bandwidth (GB/S)

360

336

224

TGP (watts)

170

?

130

Launch Date

February 2021

N/A

January 2022

Launch Price

$329

N/A

$249

*Specifications are unconfirmed by Nvidia.

A perfect GA106 silicon has 30 Streaming Multiprocessors (SMs), but the GeForce RTX 3050 Ti has only 26 enabled, resulting in approximately 87% die utilization. With 26 SMs active, the Ampere-based graphics card features 3,328 CUDA cores, which is about 7% fewer than the GeForce RTX 3060, yet 30% more than the GeForce RTX 3050. The configuration puts the GeForce RTX 3050 Ti in a sweet spot, balancing performance and power efficiency.

The clock speeds for the GeForce RTX 3050 Ti fall well within expectations. The base and boost clocks are higher than those of the GeForce RTX 3060 because the GeForce RTX 3050 Ti has fewer CUDA cores. Conversely, the clock speeds are somewhat lower than those of the standard GeForce RTX 3050, which features even fewer CUDA cores.

On the contrary, the memory subsystem of the GeForce RTX 3050 Ti is a cross between those of the GeForce RTX 3050 and the GeForce RTX 3060. While it retains the 14 Gb/s GDDR6 memory modules found in the former, the memory operates over a wider 192-bit interface, similar to the latter's. As a result, the GeForce RTX 3050 Ti offers 50% higher memory bandwidth than the GeForce RTX 3050 but trails behind the GeForce RTX 3060 by 7%.

The GeForce RTX 3050 Ti's memory capacity raises some eyebrows. The graphics card appears to carry 6GB of GDDR6 memory, half of what's available on the GeForce RTX 3060 and 2GB less than a GeForce RTX 3050. For reference, we’re comparing the original models of the GeForce RTX 3060 and GeForce RTX 3050, which shipped with 12GB and 8GB of memory, respectively. Nvidia has introduced different memory configurations over time, including the GeForce RTX 3060 8GB, GeForce RTX 3050 6GB, and GeForce RTX 3050 4GB, as well as a couple of variations with different silicon revisions.

Gok shared a screenshot of the GeForce RTX 3050 Ti achieving a graphics score of 7,787 points in 3DMark's Time Spy benchmark. A quick search shows the GeForce RTX 3060 scoring between 8,200 and 9,000 points on the same benchmark. Meanwhile, the GeForce RTX 3050 scores range from 5,300 to 6,400. If we use the maximums for comparison, the GeForce RTX 3050 Ti could be up to 22% faster than the GeForce RTX 3050 or up to 13% slower than a GeForce RTX 3060.

The GeForce RTX 3050 and GeForce RTX 3060 launched at $249 and $329, respectively. The GeForce RTX 3050 Ti could easily have filled the gap at $289. Then again, Ampere was around during difficult times where MSRPs didn't mean anything. The GeForce RTX 3060 still reigns as the most popular graphics card on Steam even to this day. Sadly, we can't rewind time to see what would have happened if the GeForce RTX 3050 Ti had come out. In any event, the graphics card would have further strengthened Nvidia's hold in the mid-range segment. With custom GeForce RTX 3060 starting at $479 nowadays, budget gamers could use another alternative, and something like the GeForce RTX 3050 Ti could certainly fill that spot.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Zhiye Liu
News Editor, RAM Reviewer & SSD Technician

Zhiye Liu is a news editor, memory reviewer, and SSD tester at Tom’s Hardware. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.

  • PEnns
    "The GDDR6 savior that budget gamers need"
    Absolutely, especially it's 6 gig of VRAM!!! If only there was a GPU, available, with 16 G VRAM for around $450.....wait, there is one but it is AMD so it doesn't show up on the radar.

    PS: I remember reviewers laughing their butts off at the 3050 series......but now we have desperate times and down is up and vice versa. Such articles can cause hypertension.
    Reply
  • usertests
    PEnns said:
    "The GDDR6 savior that budget gamers need"
    Absolutely, especially it's 6 gig of VRAM!!! If only there was a GPU, available, with 16 G VRAM for around $450.....wait, there is one but it is AMD so it doesn't show up on the radar.

    PS: I remember reviewers laughing their butts off at the 3050 series......but now we have desperate times and down is up and vice versa. Such articles can cause hypertension.
    The 3050 6GB has remained one of the best 75W options, and this could* have beaten it with 6 or 12 GB (clocks would need to go down, but it has +44.4% cores).

    With 12 GB, it could have made sense as a 3060 12GB and 3050 8GB alternative. Depends on the price.

    With 6 GB and 130-170W TDP, it would have been a tough sell vs. the 3050 8GB. Might make sense for old games, esports, or someone fiddling with settings.

    *It turns out that this prototype matches the silicon in a real product, the RTX A2000 PCIe workstation GPU:

    192-bit 6 GB or 12 GB (slower 12 Gbps and 288 GB/s)
    3328 CUDA, 104 tensor, 26 RT cores
    70W TDP and low 562-1200 MHz clocks (could go to 1500 MHz)

    mADFD1OVh1E
    keAFTWkGX0Q

    Used RTX A2000 prices appear to be in the $300-500 range depending on the VRAM. At that point, you might consider the faster Arc Pro B50 with 16 GB. Or hope for a better consumer/pro 75W/LP card in the future.
    Reply
  • PEnns
    usertests said:
    The 3050 6GB has remained one of the best 75W options, and this could* have beaten it with 6 or 12 GB (clocks would need to go down, but it has +44.4% cores).

    With 12 GB, it could have made sense as a 3060 12GB and 3050 8GB alternative. Depends on the price.

    With 6 GB and 130-170W TDP, it would have been a tough sell vs. the 3050 8GB. Might make sense for old games, esports, or someone fiddling with settings.

    *It turns out that this prototype matches the silicon in a real product, the RTX A2000 PCIe workstation GPU:

    192-bit 6 GB or 12 GB (slower 12 Gbps and 288 GB/s)
    3328 CUDA, 104 tensor, 26 RT cores
    70W TDP and low 562-1200 MHz clocks (could go to 1500 MHz)

    mADFD1OVh1E
    keAFTWkGX0Q

    Used RTX A2000 prices appear to be in the $300-500 range depending on the VRAM. At that point, you might consider the faster Arc Pro B50 with 16 GB. Or hope for a better consumer/pro 75W/LP card in the future.

    It never ceases to amaze me! TDP is such a moving target and it's either a huge issue or a non-issue.

    I remember when Intel and Nvidia hardware TDPs were all over the spectrum (by a HUGE margin) and the fans thought "a few / 100 " Watts s were not a big deal. What was that thing they loved to say?? Ah yes: "Power is cheap!"
    But if an AMD product needs 10 W more power then it is is major talking point / worth mentioning (especially in reviews!) and suddenly it is a huge consideration when buying!!
    Reply
  • usertests
    PEnns said:
    But if an AMD product needs 10 W more power then it is is major talking point / worth mentioning (especially in reviews!) and suddenly it is a huge consideration when buying!!
    I think most of the criticism I pay attention to has been fair and focused on the power efficiency rather than 10 Watts here or there. If one GPU is delivering similar performance to a competing card while using 20% less power, that's a serious competitive advantage.

    IIRC, the Pascal generation is one where Nvidia leapfrogged AMD in power efficiency, and more recently, RDNA3 fell behind, probably from badly missing performance targets after moving to chiplets.

    If you consider massive flagships, the high TDP can be justified by the amount of die area used (e.g. the RTX 5090's 575W for 750mm^2, vs. the 5080's 360W for 378mm^2). You simply need more power to remain on an optimal efficiency curve if the resources are drastically increased. However, reviews showed the 5090 using less than the TDP most of the time, likely because of CPU bottlenecks. And if you are hitting the maximum refresh rate of your monitor, you can be capped and underutilizing it.

    0-75W is its own segment because you can put the card in basically any system with a PCIe x16 slot, without a power connector or needing to replace the power supply (unless it's incredibly janky). This is also the segment where many of the low profile or single-slot cards reside since big/expensive coolers aren't required. I'll certainly complain about double digit watts if it takes a card just out of this range. For example, 6500 XT at 107-113W, or the RTX 5050 using 130W while being less efficient than the RTX 5060.
    Reply