Gaming PC

PNY RTX 4070 Review: Single 8-Pin, No 16-Pin 12VHPWR

In our first Nvidia RTX 4070 Founders Edition launch review, we showed the benchmarks of the PNY GeForce RTX 4070 Verto Dual. Since then, there have been many others, including the launch of the RTX 4060 Ti and RX 7600, but now they are under various third-party reviews. These are some of the best graphics cards for gaming and other uses, and the RTX 4070 seems to be the best in terms of specs and price.

With 2GB more VRAM, the RTX 4070 essentially performs nearly the same as the previous generation RTX 3080, despite having a much narrower 192-bit memory interface. This is thanks to the Nvidia Ada Lovelace architecture, specifically the built-in L2 cache of up to 8MB per 32-bit chunk of the memory interface. More memory helps in some cases, but higher resolutions also tend to favor the increased raw bandwidth of the 3080’s 320-bit interface.

Before we get into the details of the PNY card, let’s start with the usual specs table.

Swipe to scroll horizontally
Nvidia and AMD GPU specifications
graphics card RTX4070 RTX4060Ti RTX4070Ti RTX4080 RX7600 RX7900XT RX6750XT RX6800 RX6800XT RX6950XT
architecture AD104 AD106 AD104 AD103 Navi 33 Navi 31 Navi 22 Navi 21 Navi 21 Navi 21
process technology TSMC 4N TSMC 4N TSMC 4N TSMC 4N TSMC N6 TSMC N5+N6 TSMC N7 TSMC N7 TSMC N7 TSMC N7
Transistor (billion) 32 22.9 35.8 45.9 13.3 45.6 + 5x 2.05 17.2 26.8 26.8 26.8
Die size (mm^2) 294.5 187.8 294.5 378.6 204 300+225 336 519 519 519
SM/CU 46 34 60 76 32 84 40 60 72 80
GPU core (shader) 5888 4352 7680 9728 2048 5376 2560 3840 4608 5120
Tensor / AI core 184 136 240 304 64 168 N/A N/A N/A N/A
Ray Tracing “Core” 46 34 60 76 32 84 40 60 72 80
Boost clock (MHz) 2475 2535 2610 2505 2625 2400 2600 2105 2250 2310
VRAM Speed ​​(Gbps) twenty one 18 twenty one 22.4 18 20 18 16 16 18
VRAM (GB) 12 8 12 16 8 20 12 16 16 16
VRAM bus width 192 128 192 256 128 320 192 256 256 256
L2 / Infinity Cache 36 32 48 64 32 80 96 128 128 128
ROP 64 48 80 112 64 192 64 96 128 128
TMU 184 136 240 304 128 336 160 240 288 320
TFLOPS FP32 (Boost) 29.1 22.1 40.1 48.7 21.5 51.6 13.3 16.2 20.7 23.7
TFLOPS FP16 (FP8) 233 (466) 177 (353) 321 (641) 390 (780) 43.0 103.2 26.6 32.4 41.4 47.4
Bandwidth (GBps) 504 288 504 717 288 800 432 512 512 576
TDP (Watts) 200 160 285 320 165 315 250 250 300 335
release date April 2023 May 2023 January 2023 November 2022 May 2023 December 2022 May 2022 November 2020 November 2020 May 2022
Release price $599 $399 $799 $1,199 $269 $899 $549 $579 $649 $1,099
Current price $585 $400 $790 $1,108 $257 $780 $330 $490 $500 $630

The spec table above shows the 10 GPUs used in this review. The PNY RTX 4070 Verto Dual and Nvidia RTX 4070 Founders Edition both have identical specs as they use a reference clock. Nothing more can be added.

The RTX 4070 has a MSRP of $599.99, but there are cards regularly available for less. The RTX 4070 is currently available at: Asus, gigabyte, MSIagain Zotak Use the code “ZIPDADS” and you’ll get an instant rebate of $50 (Father’s Day, if that’s confusing). Unfortunately PNY’s card is not included in the deal. Also, it’s “as long as the funds last” so you may or may not be able to take advantage of that offer.

Deals of this sort say a lot about graphics card sales in general. For example, the past few Black Fridays and Prime Days have had few notable GPU sales, but this is where graphics cards launched within two months of their release get $50 off directly from Newegg. That’s what it means. It’s possible that Nvidia could drop the price of all GPUs by around $50 without much pain.

Short-term sales aside, I can’t help but feel that the RTX 4070 should sell for a lot less than the MSRP. The previous generation he is faster than the RTX 3070 but also has a significantly smaller die size. The TSMC 4N’s 294.5 mm^2 compared to the Samsung 8N’s 392.5 mm^2 probably means his AD104 chips cost more, but he still believes Nvidia has about 190 chips per wafer. should be able to get Even at $19,000 per 4N wafer, it’s only $100 per chip (and he probably paid Nvidia less because of how many wafers Nvidia orders).

Related Articles

Back to top button