Back in October, Nvidia launched the GeForce RTX 2070 to bring the power of the Turing architecture to more gamers. The RTX 2070 was the first card to use TU106, Nvidia's smallest Turing GPU so far. In its fully fledged form on the RTX 2070, TU106 offers 2304 shader ALUs spread across 36 Turing streaming multiprocessors, or SMs. Those SMs are further organized into three graphics processing clusters, or "GPCs" in Nvidia parlance.
Turing chips are some of the largest GPUs around, and TU106 is no different despite its place at the base of the Turing lineup. At 445 mm², this chip's die size closes in on the Vega 10 GPU that powers the Radeon RX Vega 56 and RX Vega 64. Area aside, however, the two teams' most modern graphics processors couldn't be more different.
For gamers, Turing introduced Nvidia's RT cores for acceleration of certain ray-tracing operations, along with tensor cores for neural-network processing. Even as the smallest of the Turing siblings, TU106 boasts a diverse bench of compute resources. The RTX 2070 has 36 SMs; each SM has eight tensor cores (288 total) and one RT core. Here's how TU106 stacks up with some well-known cards from AMD and Nvidia alike:
|RX Vega 56||1471||64||224/112||3584||2048||410 GB/s||8 GB|
|GTX 1070||1683||64||108/108||1920||256||259 GB/s||8 GB|
|RTX 2060 FE||1680||48||120/120||1920||192||336 GB/s||6 GB|
|RTX 2070 FE||1710||64||120/120||2304||256||448 GB/s||8 GB|
|GTX 1080||1733||64||160/160||2560||256||320 GB/s||8 GB|
|RX Vega 64||1546||64||256/128||4096||2048||484 GB/s||8 GB|
|RTX 2080 FE||1800||64||184/184||2944||256||448 GB/s||8 GB|
|GTX 1080 Ti||1582||88||224/224?||3584||352||484 GB/s||11 GB|
|RTX 2080 Ti FE||1635||88||272/272||4352||352||616 GB/s||11 GB|
|Titan Xp||1582||96||240/240||3840||384||547 GB/s||12 GB|
|Titan V||1455||96||320/320||5120||3072||653 GB/s||12 GB|
|RX Vega 56||94||330/165||5.9||10.5|
|RTX 2060 FE||81||202/202||5.0||6.5|
|RTX 2070 FE||109||246/246||5.1||7.9|
|RX Vega 64||99||396/198||6.2||12.7|
|GTX 1080 Ti||139||354/354||9.5||11.3|
|RTX 2080 Ti||144||473/473||9.8||14.2|
We're looking at TU106 today as implemented on Asus' ROG Strix RTX 2070 O8G Gaming. This is the company's halo factory-overclocked card, sitting atop a stack comprising the slightly less hot-clocked Strix A8G, the RTX 2070 Dual, and the blower-style RTX 2070 Turbo. Behold:
In its reference form, the RTX 2070 rings in at a 1410-MHz base clock and a 1620-MHz boost speed. The "factory-overclocked" RTX 2070 Founders Edition puts another 90 MHz on top of that reference boost spec. Asus has warmed up this Strix even further in its default "gaming mode" clock profile. This card has a nameplate 1815-MHz boost clock, and Turing's GPU Boost 4.0 dynamic voltage and frequency management will no doubt bring that figure even higher in practice. Folks who want to grab Asus' GPU Tweak II software can enable an "OC Mode" profile that raises the boost clock another 30 MHz, for what that's worth. Asus doesn't overclock the card's memory, but the RTX 2070 already has a rather impressive 448 GB/s of peak memory bandwidth to play with anyway, thanks to its 8 GB of GDDR6 memory running at 14 Gb/s per pin.
Asus' ROG Strix cards for Turing don't all tap the same cooler to keep their chips in check. The Strix RTX 2080 uses a brand-new cooler, fan design, and shroud, while the Strix RTX 2070 gets capped off with the same cooler and fans as those on the company's Strix GTX 1080 Ti. That's fine, however, since a 2.5-slot cooler on an ostensibly 175-W graphics card is still ridiculous overkill. Incidentally, you can find this same massive cooler on the company's ROG Strix RTX 2060, too.
As on past Strix cards, the RTX 2070 offers subtle, well-integrated RGB LED lighting for those who want to bedazzle their builds. The main illuminated features of the Strix come courtesy of a pair of light pipes flanking this card's trio of fans. These light pipes will go unseen in traditional builds with horizontally mounted graphics cards, but folks who go to the trouble of using a vertical mount will be rewarded by the Strix.
Flipping the card over reveals a backplate that will be familiar to observers of Strix cards as far back as the company's GTX 1080. The ROG logo lights up in RGB LED glory. If you don't like flash, a button at the rear of the card disables all of the Strix's RGB LEDs with one touch—no software required.
Around back, the Strix switches out the RTX 2070's standard complement of three DisplayPort outputs and one HDMI output for a pair each of DisplayPorts and HDMI outs. The card keeps the VirtualLink USB Type-C connector for as-yet-unannounced VR headsets, though.
The front edge of the Strix card boasts a pair of PWM fan connectors to link any connected spinners to changes in GPU temperature. These headers remain unique to Asus' Strix cards, as far as I can tell, and they're quite useful even as motherboards have begun to offer more and more temperature sources for connected fans. A system's GPU temperature tends to remain frustratingly hidden from motherboard makers' fan-control utilities, and Asus' software-free fan control approach here makes for the simplest method of ensuring the graphics card gets the fresh air it needs to stay cool.
The Strix's own fans operate using two separate profiles that you can control with a switch near the LED on/off button. The default "performance" BIOS lacks the zero-RPM-at-idle fan mode we've come to expect from modern high-end graphics cards, and its fan curve is more aggressive under load to maximize heat removal. The quiet BIOS, on the other hand, enables both a zero-RPM mode and a less aggressive fan curve than its performance counterpart in order to minimize noise under load. We'll examine the effects of these firmwares in greater depth during noise and thermal testing.
Stripping the Strix back to its bones gives us a better look at its massive cooler and impressive power-delivery circuitry, as well as a sturdy frame that both mates to the backplate to prevent card sag and helps cool the eight GDDR6 packages that ring the TU106 GPU.
Asus' cooler uses seven heat pipes running into a large fin stack. The power delivery circuitry gets coupled to this stack using a thermal pad of its own.
The GPU itself comes in contact with a polished, solid baseplate.
To power the Strix, Asus taps the common uPI uP9512 PWM controller. This chip talks to 10 TI NextFET CSD95481RWJ integrated power stages (possibly via some PWM doublers). These highly efficient power stages are rated for 60 A of maximum output each, and even eight of them dedicated to powering the GPU would rival some high-end motherboards for current-delivery potential. On the Strix, I expect these ICs will deliver smooth power while generating minimal heat.
Overall, Asus has built an undeniably high-end and well-considered take on the GeForce RTX 2070 with this ROG Strix card. For the privilege, the company asks $630 for its halo card at e-tail, making it one of the most expensive RTX 2070s around. Let's have an in-depth look at the performance this card delivers now.