Does anyone remember Palit's GTX 1550 Ti KalmX? I still had this card in the test for Tom's hardware at the time and the conclusion was rather mixed, because the card did not manage to survive even without any housing ventilation. The new GTX 1650 KalmX from Palit , I can spoil it before, manages this feat quite, albeit only in the second attempt after fine tuning by the manufacturer. But it is significantly faster than its predecessor and hardly more expensive.
This is exactly what led me to take another look at the test of this card, with and without the support of a very slow-turning housing fan. So almost total slack against quite a slight hint of nothing. Both are inaudible to the normal ear, but the fingertips applied to the heatpipes immediately notice the difference. I have that in the picture later, so please be patient. Palit is currently lying with approx. 170 Euro street price not in the absolute bargain window, but if you want it quietly, you may suffer something. Think about it. But let's get to the basics of the GeForce GTX 1650.
The TU117-300 is not a real full-end
Well, let's look again officially at what Nvidia offers us here as a chip, which Palit then installs with RAM and we come to the interior as well as the construction of the TU117-300. While Turing's streaming multiprocessors consist of fewer CUDA cores than Pascal's, the design compensates for this in part by distributing more streaming multiprocessors (SM) on the GPU.
The Turing architecture assigns each set of 16 CUDA cores, one scheduler, and one disposition unit per 16 CUDA cores (such as Pascal). Four of these 16-core groupings include the SM, 96 KB cache, which can be configured as 64 KB L1/32 KB of shared memory, or vice versa, and four texture units. As a result, the TU117-300 of the GeForce GTX 1650 with its 896 CUDA cores has such 14 SM. If you followed Nvidia's nomenclature for the chip designation, a larger TU117-400 would only be logical.
But the theory also confirms the fact that the TU117-300 is not the maximum expansion. I have modified and shortened a scheme of the TU116-300 of the GeForce GTX 1660 to fit exactly for the TU117-300 of the GeForce GTX 1650. We see very nicely the gaps in the two GPC, so that the TU117-400, if it will exist, could appear with 16 SM, which should also run out as a result of 1024 CUDA-Cores. Unfortunately, Nvidia does not reveal whether this could be a GeForce GTX 1650 ("We generally don't talk about unannounced products").
Because Turing has twice as many schedulers as Pascal, only one statement needs to be issued to the CUDA cores in every second cycle. In between, there is enough room to send a different instruction to any other device, including the INT32 cores. This is no different with the GTX 1650 than with the other larger Turing cards. Nvidia splits these 14 SM into two graphics processing clusters. In addition, 14 SMs, each with four associated texture units, total56 texture units for the entire GPU.
In addition to the Shaders and the Unified Cache of the Turing Architecture, the TU117 also supports a pair of algorithms called Content Adaptive Shading and Motion Adaptive Shading, which together are called variable-rate shading. I have already written a longer introduction to the launch of the GeForce RTX 2080 (Ti). Nvidia has also revealed that it is replacing the tensor cores with dedicated FP16 cores to enable the GeForce GTX 1660 Ti to process semi-precision operations at twice the rate of FP32. Nvidia has also noted that the TU117 is the smallest Turing chip that will be able to use the DXR share in the drivers.
Four 32-bit memory controllers end up giving the TU117 an aggregated 128-bit bus that operates the four GDDR5 modules at up to 128 GB/s. However, this is a considerable bandwidth disadvantage compared to the GTX 1660 and is even lower than the level of the old GeForce GTX 1060, but also significantly higher than that of the GeForce GTX 1050 Ti.
More about the Palit GeForce GTX 1650 KalmX 4GB
There is no real reference card from Nvidia, but at least default values. And Palit has fully adhered to them, at least electrically. The rest, however, is refreshingly different, because the relatively small passive cooler still drives some effort. But I will come back to that in a very late place. The card, which weighs only 426 grams, measures 17.5 cm gross in its length from the outer edge of the slot panel to the end of the radiator cover. It is 3.5 cm thick and 13.5 cm high (from the top of the motherboard slot to the top edge of the cooler). There are no LEDs, why?
With an HDMI 2.0 port and twice DisplayPort -1.4 port, there are enough variants to connect to the monitor.
|Installation length (gross)||17.5 cm|
|Installation height (gross)||13.5 cm|
|Installation depth front (gross)||3.5 cm|
|Installation depth rear (gross)||0.3 cm|
|Connections:||1x HDMI 2.0
2x DisplayPort 1.4
A first overview of the electrical data is provided here by the latest version of GPU-Z, whereby the base clock and the boost correspond to the reference specification:
The following table shows the relevant models and a fictitious GeForce GTX 1650 Ti:
GeForce GTX 1060
GeForce GTX 1660 Ti 6 GB
GeForce GTX 1660 6 GB
GeForce GTX 1650
GeForce GTX 1650
GeForce GTX 1650 Ti
|Base Clock Rate
||1506 MHz||1500 MHz||1530 MHz||1485 MHz||1485 MHz
|GPU Boost Rate
||1708 MHz||1770 MHz||1785 MHZ||1665 MHz||1665 MHz
||6 GB GDDR5||6 GB GDDR6||6 GB GDDR5||4GB GDDR5||4GB GDDR5||4GB GDDR5|
||4000 MHz||6000 MHz||4000 MHz||4000 MHz||4000 MHz
||1.5 MB||1.5 MB||1.5 MB||1 MB||1 MB||1 MB|
||120 W||120 W||120 W||75 W||75 W||?|
||4.4 billion||6.6 billion||6.6 billion||4.7 billion||4.7 billion||4.7 billion|
Test system and evaluation software
As always, I benchmark with my own PresentMon GUI and the interpreter software, which fills the Excel charts and the graphical output (and only the) so extra- or so. interpolates that all graphs are exactly long, creating a uniform and comparable time line. The remaining evaluations up to percentiles and bar graphics are based on the real raw data from the measurement. The benchmark system is still relatively new and has been upgraded in some areas at the end of 2019.
I have also summarized the individual components of the test system in tabular form.
|Test System and Equipment
Intel Core i9-9900 K
4x 8GB G.Skill FlareX DDR4 3200
||Alphacool Ice Block XPX (1151), XPX Pro (AM4, 2066)
Alphacool Ice Grinder (modified)
Thermal Grizzly Kryonaut
||Lian Li T70 FormatC Edition (Special Build)
Open Benchtable, Closed Case
Non-contact direct current measurement on PCIe slot (riser card)
||1x Optris PI640 + 2x Xi400 Thermal Imagers
Pix Connect Software
Type K Class 1 thermal sensors (up to 4 channels)
||NTI Audio M2211 (with calibration file)
Steinberg UR12 (with phantom power for the microphones)
Creative X7, Smaart v.7
Own anechoic chamber, 3.5 x 1.8 x 2.2 m (LxTxH)
Axial measurements, perpendicular to the center of the sound source(s), measuring distance 50 cm
Noise emission in dBA (slow) as RTA measurement
Frequency spectrum as graphic
|Os:||Windows 10 Pro (1909, all Updates)|
Palit GeForce GTX 1650 KalmX, 4GB GDDR5, HDMI, 2x DP (NE5165001BG1-1170H)