We have with us the EVGA GeForce RTX 3090 FTW3 Ultra graphics card, the company's premium air-cooled custom-design RTX 3090 offering. The RTX 3090 FTW3 Ultra features EVGA's iCX3 cooling solution that approaches the task of cooling this 350 W GPU with not just a meaty heatsink ventilated by fans, but such innovations as cutouts in the PCB that let the airflow through, and a network of thermal sensors at critical areas to help the cooler respond better to heat. The GeForce RTX 3090 transcends the barrier between client gaming graphics and professional visualization, letting content creators leverage its awesome 24 GB video memory and nearly 1 TB/s memory bandwidth to deal with vast graphics rendering datasets, especially when paired with NVIDIA's GeForce Studio drivers. No doubt, it's better endowed than the RTX 3080, but NVIDIA believes that its gaming capabilities can't be seen in isolation to justify double the price of the RTX 3080. Hence, the company considers the RTX 3080 to be its "flagship" gaming product.
With the GeForce RTX 30-series "Ampere," NVIDIA has reshaped the top end of its product stack. The RTX 3090, although the meatiest GPU on paper and priced the highest, isn't considered to be the company's flagship graphics product. The RTX 3080 is capable of AAA gaming at 4K Ultra HD with RTX raytracing on, which means the RTX 3090 attains a "halo" status not unlike that of the TITAN series. While the previous-generation TITAN RTX was priced at $2,500, the RTX 3090 starts $1,000 cheaper, at $1,500. It's up to you what you want to do with an almost fully unlocked "GA102" GPU and 24 GB of blistering fast memory.
The GeForce "Ampere" architecture represents the second generation of NVIDIA's path-breaking RTX technology, which brings real-time raytracing to the consumer graphics segment. While we're still a bit far off from fully raytraced graphics, NVIDIA figured out a way to combine conventional raster 3D graphics with real-time raytraced components, such as lighting, shadows, reflections, ambient-occlusion, and global-illumination. With the 2nd generation RTX, NVIDIA is introducing raytraced motion blur, an effect that's extremely difficult to pull off in real-time and was until now emulated in post-processing.
The 2nd generation RTX is a combination of new "Ampere" CUDA cores that offer up to double the math throughput over the previous-generation using concurrent INT32 and FP32 operations; second generation RT cores that double the intersection throughput and feature fixed function hardware with temporal components to accelerate ray-traced motion blur; and third generation Tensor cores that leverage the sparsity phenomenon in deep-learning neural nets to accelerate AI inference performance by an order of magnitude. NVIDIA combines these with a blistering new memory technology it calls GDDR6X and co-developed with Micron Technology—it takes advantage of the new PAM4 signaling format to increase the data rate. The RTX 3090 features a 384-bit wide memory bus with 19.5 Gbps memory throughput, which works out to a stellar 940 GB/s of memory bandwidth. The GPU leverages the PCI-Express Gen 4 bus.
EVGA bolstered the GeForce RTX 3090 for the enthusiast gaming and overclocking market with the RTX 3090 FTW3 Ultra, which is priced at about $1,800, a $300 premium. The iCX3 triple-slot cooler leverages a large aluminium fin-stack heatsink with a high fin-surface area and three hydro-dynamic bearing fans, where the middle fan is pushed slightly out of alignment of the other fans. EVGA also innovated cutouts in the PCB at various places, so air from the cooler can flow right through, similar to the dual-axial flow-through cooling solution of the Founders Edition card. The FTW3 Ultra also pulls power through three 8-pin PCIe power connectors, giving the RTX 3090 sufficient power to sustain its GPU Boost frequencies better. EVGA gave the FTW3 Ultra a hearty overclock of 1800 MHz (compared to the 1695 MHz reference). In this review, we put the card through its paces against several other RTX 3090 cards we tested, as well as its nearest rival—the RTX 3080.
|GTX 1080 Ti||$650||3584||88||1481 MHz||1582 MHz||1376 MHz||GP102||12000M||11 GB, GDDR5X, 352-bit|
|RX 5700 XT||$370||2560||64||1605 MHz||1755 MHz||1750 MHz||Navi 10||10300M||8 GB, GDDR6, 256-bit|
|RTX 2070||$340||2304||64||1410 MHz||1620 MHz||1750 MHz||TU106||10800M||8 GB, GDDR6, 256-bit|
|RTX 2070 Super||$450||2560||64||1605 MHz||1770 MHz||1750 MHz||TU104||13600M||8 GB, GDDR6, 256-bit|
|Radeon VII||$680||3840||64||1802 MHz||N/A||1000 MHz||Vega 20||13230M||16 GB, HBM2, 4096-bit|
|RTX 2080||$600||2944||64||1515 MHz||1710 MHz||1750 MHz||TU104||13600M||8 GB, GDDR6, 256-bit|
|RTX 2080 Super||$690||3072||64||1650 MHz||1815 MHz||1940 MHz||TU104||13600M||8 GB, GDDR6, 256-bit|
|RTX 2080 Ti||$1000||4352||88||1350 MHz||1545 MHz||1750 MHz||TU102||18600M||11 GB, GDDR6, 352-bit|
|RTX 3070||$500||5888||64||1500 MHz||1725 MHz||1750 MHz||GA104||17400M||8 GB, GDDR6, 256-bit|
|RTX 3080||$700||8704||96||1440 MHz||1710 MHz||1188 MHz||GA102||28000M||10 GB, GDDR6X, 320-bit|
|RTX 3090||$1500||10496||112||1395 MHz||1695 MHz||1219 MHz||GA102||28000M||24 GB, GDDR6X, 384-bit|
|EVGA RTX 3090|
|$1800||10496||112||1395 MHz||1800 MHz||1219 MHz||GA102||28000M||24 GB, GDDR6X, 384-bit|
EVGA has completely revamped their design language with the RTX 30 series. Instead of a transparent cooler shroud we now get classic black that's paired with silver metal highlights and a red trim. I have to say I really like the red even though it clashes a bit with NVIDIA's "green" color theme. On the back, you'll find a high-quality metal backplate.
Dimensions of the card are 30 x 14 cm.
Installation requires three slots in your system.
Display connectivity options include three standard DisplayPort 1.4a and one HDMI 2.1. Interestingly, the USB-C port for VR headsets, which NVIDIA introduced on the Turing Founders Editions, has been removed—guess it didn't take off as planned. The DisplayPort 1.4a outputs support Display Stream Compression (DSC) 1.2a, which lets you connect 4K displays at 120 Hz and 8K displays at 60 Hz. Ampere can drive two 8K displays at 60 Hz with just one cable per display.
Ampere is the first GPU to support HDMI 2.1, which increases bandwidth to 48 Gbps to support higher resolutions, like 4K144 and 8K30, with a single cable. With DSC, this goes up to 4K240 and 8K120. NVIDIA's new NVENC/NVDEC video engine is optimized to handle video tasks with minimal CPU load. The highlight here is added support for AV1 decode. Just like on Turing, you may also decode MPEG-2, VC1, VP8, VP9, H.264, and H.265 natively, at up to 8K@12-bit.
The encoder is identical to Turing. It supports H.264, H.265, and lossless at up to 8K@10-bit.
This BIOS switch lets you toggle between the normal (default) BIOS and an "OC" BIOS. Both BIOSes run the same clocks and voltages. The only difference is that "OC" has no fan stop and the fan curve is a bit more aggressive.
Near the back of the card, you'll find two pin headers. The first one is to provide an RGB signal to other components, so they can be synced with the graphics card RGB. The second one lets you connect a case fan to the graphics card, making it spin at the same rate as the GPU fans. Since the graphics card is the highest heat output in most computers, this makes a lot of sense—idle fan stop during browsing and productivity and fans running when gaming.
Unlike the NVIDIA Founders Edition card that introduces the new 12-pin power input, EVGA sticks to the industry standard 8-pin PCIe power inputs, but there are three of them. Combined with PCIe slot power, this configuration is rated for 525 W. At reference specifications, the RTX 3090 is already shipping with 350 W typical board power, which maxes out power draw of two 8-pin connectors and begins to feed heavily on the PCIe slot. The third 8-pin input, which adds 150 W, is hence much needed for this card to have any meaningful power headroom.
The GeForce RTX 3090 supports SLI and features a newer-generation NVLink bridge interface, which means you can't use your NVLink bridge from your Turing cards. Be warned that with Ampere NVIDIA isn't supporting SLI as in implicit multi-GPU (SLI as you know it), but explicit multi-GPU that's developed and supported by game/application developers. With multi-GPU game support being pretty much non-existent, this basically means SLI is dead. Perhaps creative and 3D modeling applications that support explicit multi-GPU can benefit from SLI.
EVGA's heatsink uses a copper baseplate to spread the heat and six heatpipes that quickly move the heat away from the GPU.
The backplate is made out of metal and protects the card against damage during installation and handling. Note the heatpipes integrated in the backplate, and the thermal pads to soak up heat from the memory chips placed on this side of the card.
High-resolution PCB PicturesThese pictures are for the convenience of volt modders and people who would like to see all the finer details on the PCB. Feel free to link back to us and use these in your articles or forum posts.
High-res versions are also available (front, back).
Our Patreon Silver Supporters can read articles in single-page format.