Review of the Nvidia GeForce GTX 1650

For

• Over 50 percent faster than GTX 1050
• No extra power required on many models
• Great for small form factor builds

Against

• Costs more than the GTX 1050 it replaces
• Struggles in more demanding games
• Budget GPUs often aren’t the best value

The introduction of the GeForce GTX 1650 was a foregone conclusion, as no other Turing GPUs adequately serve as a budget alternative to the top-tier graphics cards. This model is expected to be the last iteration of the Turing architecture, at least on the 12nm process. The new TU117 chip allows Nvidia to offer a complete range, from the high-end GeForce RTX 2080 Ti to the more affordable RTX 2060 for those interested in ray tracing, while the GTX 1660 Ti and GTX 1660 omit the RT and Tensor cores to provide more cost-effective options.

The GTX 1650 is equipped with the new TU117 GPU, which represents a smaller and more cost-effective version of the TU116 found in the GTX 1660 and 1660 Ti models. The primary distinctions when compared to the 1660 series lie in the memory configuration and the number of Streaming Multiprocessors (SMs), which subsequently affects the count of CUDA cores, texture units, and ROPs. This GPU continues to utilize TSMC’s 12nm lithography, while AMD’s Radeon VII currently employs a 7nm process. Consequently, the die size is approximately one-third smaller than that of the TU116, containing 4.7 billion transistors.

As anticipated, the GTX 1650 is equipped with 4GB of GDDR5 memory, operating at a speed of 8GT/s, which is identical to that of the GTX 1660 and the earlier GTX 1060 models. It features four active memory controllers on a 128-bit bus, providing a bandwidth of 128GB/s, which is marginally higher than that of the GTX 1050 Ti. Additionally, it includes 32 Render Output processors (ROPs).

The TU117 GPU core, utilized in the GTX 1650, comprises 14 Streaming Multiprocessors (SMs), resulting in a total of 896 CUDA cores and 56 texture units. Similar to other Turing architecture GPUs, the GTX 1650 is capable of performing concurrent FP32 and INT calculations, which can enhance gaming performance by approximately 15-35 percent, depending on the specific game, when compared to the prior Pascal architecture. It is important to note that the desktop version of the GTX 1650 does not utilize a fully enabled TU117, as there exists a mobile variant featuring 16 SMs and 1024 CUDA cores. Consequently, the introduction of a GTX 1650 Ti in the future seems plausible, and one might reasonably anticipate its arrival.

Nvidia generally adopts a conservative approach regarding its reported boost clocks, with the majority of its graphics cards operating at speeds significantly exceeding the stated figures. The standard GTX 1650 features a boost clock of 1665MHz, which translates to a theoretical performance of 2984 GFLOPS. While this performance is lower than that of the GTX 1060 series, it is approximately 50 percent greater than that of the GTX 1050. Additionally, the GTX 1650 is engineered to function without a 6-pin PCIe power connector; however, factory overclocked variants, such as the MSI GTX 1650 Gaming X 4G that I am currently utilizing, possess elevated clock speeds and necessitate a 6-pin PEG connector.

Ultimately, the speculation regarding pricing turned out to be somewhat inflated, which is a positive development. The suggested retail price for the base models is set at for the GTX 1650, making it only marginally pricier than the GTX 1050 Ti. This is the theoretical perspective; however, the 1050 Ti has recently been available for and above. Additionally, there remains the possibility of introducing a GTX 1650 Ti variant priced at.

Factory overclocked variants, such as the Asus and MSI graphics cards utilized in my testing, are indeed priced higher than their standard counterparts. Nevertheless, if you seek a performance upgrade beyond the base GTX 1650, it would be advisable to consider the GTX 1660 or AMD’s RX 570/580, or even the earlier GTX 1060 model. These options demand greater power than the 1650; however, any power supply unit equipped with the necessary 6-pin connector should be adequate.

Performance of the GeForce GTX 1650

Nvidia asserts that the GTX 1650 will perform up to twice as quickly as the GTX 950 and 50 percent faster than the GTX 1050, which appears to be a reasonable assessment, particularly considering that both of those graphics cards are equipped with only 2GB of VRAM. Based on its specifications, the GTX 1650 is expected to be approximately 25-30 percent faster than the GTX 1050 Ti; however, this also suggests that it may be slower than the GTX 1060 variants.

The enhancement in performance is attributed to multiple modifications. Firstly, the 1650 features increased memory bandwidth and a greater number of CUDA cores in comparison to the 1050 and 1050 Ti. Secondly, it operates at a significantly higher clock speed. Lastly, the Turing architecture facilitates simultaneous FP32 and INT computations, potentially increasing performance by an additional 10-30 percent relative to the Pascal GPUs, contingent upon the specific game and settings. Now, let us proceed directly to the performance results.

In summary, the GTX 1650 performs as anticipated. It demonstrates a significant improvement over the GTX 1050, achieving a 57 percent increase in performance at 1080p medium settings and a 73 percent increase at 1080p ultra during my evaluations. This enhancement can largely be attributed to the limited VRAM of the GTX 1050, as the GTX 1050 Ti shows performance levels that are much closer to the 1650, which is approximately 30 percent faster. Additionally, the GTX 1060 3GB outperforms the 1650 by 17 percent at 1080p medium and 9 percent at 1080p ultra, while the RX 570 4GB occupies a comparable position.

An additional aspect to consider is that the GTX 970, released in 2014, exhibits only a marginal performance advantage (averaging between 1-3 percent) over the GTX 1650, primarily attributable to advancements in architecture over the last two generations. Contemporary games generally perform better on the GTX 1650, whereas older titles tend to show a preference for the GTX 970.

To put it differently, a premium 145W graphics card priced at from 4.5 years ago has transformed into a more affordable 75W card costing today. While this development appears favorable, it undeniably indicates a deceleration in the rate of performance enhancements. Four and a half years prior to the release of the GTX 970, Nvidia introduced the GTX 470, with the GTX 970 offering performance that was up to three times greater.

It is important to note that my testing was conducted using a significantly overclocked MSI GTX 1650 Gaming X 4G, which operates at a boost frequency of 1860MHz, compared to the reference boost clock of 1665MHz. In contrast, GTX 1650 models available at Nvidia’s suggested retail price of typically feature boost clocks ranging from 1665MHz to 1750MHz, resulting in a performance reduction of approximately 5 to 10 percent.

The value proposition of the GeForce GTX 1650.

In conclusion, the charts presented above illustrate two distinct perspectives on performance relative to monetary investment, utilizing Germany’s Euro pricing as a reference. I have determined the “value” for these midrange, budget, and select high-end models based on the most competitive pricing available as of April 23. While prices may vary, the majority of these graphics cards can be found at the prices indicated.

It is noteworthy that in both the United States and the United Kingdom, the Nvidia 1060 3GB is offered at a highly competitive price, whereas in Europe, the GTX 1650 currently represents the best value from Nvidia. However, when considering performance relative to cost, AMD’s RX 570 4GB remains unparalleled. I have anticipated that the availability of this card would diminish, resulting in increased prices, yet it has consistently been priced at or below since early 2019. The primary factor that may lead to the discontinuation of the RX 570 (and 580) from the market is the anticipated release of AMD’s Navi, which is rumored to occur on July 7; however, we will need to observe developments closely.

The issue at hand is that GPUs do not operate independently. When constructing a gaming PC, the expenses associated with the other components must also be considered. I have utilized a midrange build priced at approximately, excluding the GPU, to assess the overall gaming value of each GPU within the context of the complete system. From this analysis, it becomes evident that the higher-performing graphics cards are significantly more favorable recommendations.

The GTX 1650 is equipped solely with the previous generation of NVENC technology.

A recent development following the initial launch of the GTX 1650 reveals that the Turing TU117 GPU it employs does not feature the upgraded Turing NVENC block; rather, it utilizes the Volta version, which is essentially equivalent to the Pascal version. What is the purpose of NVENC? It is responsible for video encoding, facilitating streaming, Nvidia ShadowPlay, and Nvidia Highlights. Nvidia has promoted the Turing architecture as offering enhanced quality and performance for NVENC, claiming it surpasses the quality of the x264 Fast setting commonly used by many streamers, while also being approximately 15 percent faster than the NVENC found in Pascal and Volta architectures.

The impact is not significant, considering that individuals utilizing budget GPUs are generally less inclined to engage in streaming activities; however, it is somewhat peculiar. One might wonder how many transistors were eliminated from the die size by excluding the latest NVENC. It is likely a reduction of a few hundred million at most, translating to merely a 2-4 percent increase in die size. This situation highlights the limitations faced by budget GPUs, where every cent is crucial.

The Volta NVENC block performs adequately and is capable of managing the demands of a budget gamer. Nevertheless, it represents a regression and provides further justification for considering higher-end GPUs, such as the GTX 1660 and its superior models.