The introduction of the new Radeons has paradoxically revived interest in the GeForce RTX 4080. We’ll be reviewing a variant that’s at the lower end of the lineup, but cooler performance can bravely match models costing several thousand more. .
At launch, we introduced the GeForce RTX 4080 Founders Edition. The RTX 4080 is the first card built on the smaller AD103 chip with Ada Lovelace architecture. We discussed the novelties and changes it brings during its presentation.
While the RTX 4090’s larger AD102 chip has 12 basic GDC blocks (whereas the RTX 4090 has eleven active), the AD103 only has seven. From this one can already guess that the difference in performance compared to the RTX 4090 will be large.
Nvidia thus pits the RTX 4090 against the older RTX 3080 Ti, which offered similar parameters and was much closer in price to the RTX 4080 than the RTX 3080. However, the old GeForce is significantly worse in terms of memory capacity, performance, efficiency, and a little worse in terms of consumption.
It’s disappointing that the RTX 4080 shrinks so significantly compared to the RTX 4090. In terms of the number of compute units, bus width, or memory capacity, it’s about two-thirds the RTX 4090, it’s only slightly better. At least that also applies to the consumption, the power consumption is limited to 320 W, while for the RTX 4090 it is 450 W. But at the same time, manufacturers can be expected to use this “reserve” for overclocked models.
Below are the parameters of the top models of the current and previous generations.
| Template | RTX Titan | RTX 3080Ti | RTX 3090 | RTX 3090Ti | RTX 4090 |
| Graphics chip | TU201-400-A1 | GA102-225-A1 | GA102-300-A1 | GA 102-350 | AD102 300-A1 |
| Productive process | TSMC 12nm | Samsung8nm8N | Samsung8nm8N | Samsung8nm8N | TSMC 4nm |
| Chip area | 754mm² | 628mm² | 628mm² | 628mm² | 608mm² |
| Number of transistors | 18.6 billion. | 28.3 billion. | 28.3 billion. | 28.3 billion. | 76.3 billion. |
| Graphics Processing Cluster (GPC) | 6 | 7 | 7 | 7 | 11 |
| Multiprocessor (SM) Streaming | 72 | 80 | 82 | 84 | 128 |
| Rasterization Unit (ROP) | 96 | 112 | 112 | 112 | 176 |
| Stream Processors (CUDA cores) | 4 608 | 10 240 | 10 496 | 10 752 | 16 384 |
| Texturizing Unit (TMU) | 288 | 320 | 328 | 336 | 512 |
| Core RT | 72 | 80 | 82 | 84 | 128 |
| Tensor of Jadra (tensor cores) | 576 | 320 | 328 | 336 | 512 |
| Basic beat | 1 350 MHz | 1 365 MHz | 1 395 MHz | 1 560 MHz | 2235Mhz |
| The push of touch | 1770Mhz | 1665Mhz | 1695Mhz | 1860Mhz | 2520Mhz |
| Memory capacity | 24GB | 12GB | 24GB | 24GB | 24GB |
| Memory type | GDDR6 | GDDR6X | GDDR6X | GDDR6X | GDDR6X |
| Bus width | 384 b | 384 b | 384 b | 384 b | 384 b |
| Memory speed | 14.0GB/sec | 19.0GB/sec | 19.5GB/sec | 21.0GB/sec | 21.0GB/sec |
| Memory beat | 1750Mhz | 1 188 MHz | 1 219 MHz | 1 313 MHz | 1 313 MHz |
| Total memory throughput | 672.0GB/sec | 912.4GB/sec | 936.2GB/sec | 1000.8GB/sec | 1000.8GB/sec |
| Interface | PCIe 3.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 |
| Additional power supply | 2×8 pins | 1 x 12-pin (2 x 8-pin) | 1 x 12-pin (2 x 8-pin) | 1 x 12-pin (3 x 8-pin) | 1 x 12-pin (4 x 8-pin) |
| Maximum card power | 280W | 350W | 350W | 450W | 450W |
| Introductory price | $2,499 | $1,199 | $1 499 | USD 1 999 | $1,699 |
| introductory price including VAT (about 25.00 CZK/USD) |
75,600 Czech crowns | 36,300 CZK | 45,300 CZK | 60,500 CZK | 48,500 Czech crowns |
Palit GeForce RTX 4080 GameRock
Palit’s RTX 4080 in GameRock design it has two BIOSes. This time it differs not only in the setting of the fan regulation, but also in the maximums that can be set for the power limit. However, both are set to the same beats. Performance is similar, so we’ll test it on the quieter one.
We have the non-overclocked version in the test, so the clock with boost is at the reference 2505 MHz.
Silent BIOS | Performance
Both bios have the power consumption limit increased from 320 to 340W, but it is also the maximum in silent mode. In terms of performance, you can increase it by another 10W.
Silent BIOS | Performance
With both BIOSes, the board is passively cooled with no load. For several generations, GeForces have been able to monitor and regulate two fans independently, for cards with three fans, two fans are usually connected to the first output and the third fan to the second. In addition, the diagnostic card shows the average GPU and memory temperatures, the hottest spot of the chip, the total power consumption of the card and the GPU power consumption, the consumption of the PCIe slot and the external power supply, the individual voltages and other values.
Silent BIOS | Performance
On the packaging, Palit draws attention to the “pure crystal design, which turns paper into a work of art when the backlight is off, and a sparkling kaleidoscope with lit ARGB LEDs.” And in action, it’s a very large wild, probably the brightest card I’ve tried so far.
More interesting is probably the caveat about Gale Hunter fans, which have special fins at the ends of the blades, shaping of the ribs for smoother airflow, and the ability to synchronize the board’s backlight with the rest of the components using the cable supplied without the need to install additional software.
Inside you’ll find, in addition to the almost essential three-eight-pin power supply for today’s GeForces, a backlight sync cable and a folding strut to support the heavy card in the slot.
An instructional video from the manufacturer shows what the strut looks like and how it is assembled. The nice thing is that the manufacturer also used the holes on the cover with thread for mounting, which are usually found on more powerful boards. So you don’t have to worry about detaching it when you move the cabinet, like with braces with the arm that the board rests on.
