Introduction

The NVIDIA RTX A2000 "Ampere" in today's review is a professional-visualization graphics card designed by NVIDIA for commercial desktops and workstations. It features a low-profile design making it fit for compact towers or rackmount workstations where space comes at a premium. The card comes in a half-height, dual-slot design and uses a self-contained air cooler (doesn't rely on rack airflow). NVIDIA hasn't stuck either the GeForce or Quadro label onto this card. The "A" in A2000 denotes the "Ampere" graphics architecture it is based on. Another important capability is that the card runs on PCI-Express slot power only—no external power connectors needed.

The RTX A2000 fits the need for a compact graphics card built with certain enterprise-grade features: the pro-vis support channel, plenty of display connectivity for a multi-monitor setup with up to four 4K displays, and contemporary software capabilities, including the full DirectX 12 Ultimate feature set. Within its segment, the A2000 can help with various 3D creativity applications, as well as help with AI-augmented applications. The reviewed RTX A2000 comes in two variants based on memory—6 GB and 12 GB. Both models feature ECC for the memory. We have with us the 6 GB variant, which we bought for €720 (including 20% VAT).



By now you must be wondering what a pro-vis graphics card is doing on TechPowerUp, and you'd be right in thinking so—this is technically the first time we're testing a professional graphics card, but this is not a professional graphics card review. I mentioned the price early on because it's the reason I bought this thing to test in the first place. The A2000 isn't a GeForce RTX 3050/3060 in a suit even though it's based on the same GA106 silicon as those two. It has its own unique hardware configuration we'll get to in a minute. At the price we bought this card for, you'll be getting an RTX 3050 in the wild (i.e. scalpers selling on eBay), whereas the card has a much stronger hardware configuration. What makes it so interesting is that for its tiny form factor, it offers a lot of performance.

The A2000 is carved out of the GA106 silicon by enabling 26 out of 30 streaming multiprocessors (SM) present on the silicon. This works out to 3,328 CUDA cores, 104 TMUs, 48 ROPs, 104 Tensor Cores, and 26 RT cores. In comparison, the RTX 3050 only gets 20 SM, 2,560 CUDA cores, 80 Tensor Cores, and 20 RT cores. The real ace up the A2000's sleeve has to be its full 192-bit GDDR6 memory bus in line with that of the RTX 3060. The RTX 3050 uses a 50% narrower 128-bit bus. Unlike the RTX 3050, the RTX A2000 uses the full PCI-Express x16 bus width.

Another major set of hardware features with the A2000 are with its power configuration. Despite having more muscle than the RTX 3050, the A2000 is optimized to meet a typical board power of just 70 W. You can be absolutely sure it's not sneakily pulling more power in short bursts because the card lacks any additional power connectors and relies entirely on the PCI-Express slot, which can deliver no more than 75 W. This is a professional-segment card, so it will strictly maintain its power limits so as to never overload the PCIe slot and damage an expensive workstation motherboard. The A2000 achieves 70 W using tighter clock speeds than its GeForce cousins, the most aggressive power-management (NVIDIA probably roped in some GeForce Max-Q engineering talent to work on this) and perhaps even highest bins of the GA106 silicon for the best electrical characteristics.

We plan to put the RTX A2000 6 GB through our usual review selection of gaming benchmarks for two reasons. First, it's the most powerful low-profile graphics card ever built and doesn't even need additional power connectors. Second, as we mentioned, under some circumstances, you may actually find the A2000 a bargain compared to scalper-priced RTX 3050 or RTX 3060 cards. Out of the box, the card comes with its low-profile expansion bracket installed, but you'll get a full-height bracket, and dongles that let you use the four mini-DisplayPort 1.4 as standard-size DisplayPorts.

Packaging

The Card

The RTX A2000 comes with an extremely clean design—the card is almost like a magic black box. The matte surface looks impeccable, if only it was a bit more resistant to fingerprints. What's breaking the look a bit is that no backplate is included—the cost increase should be negligible, especially in this price segment.


Compared to other cards, the RTX A2000 is tiny. We have the GeForce RTX 3060 Ti on the left and RTX 3090 on the right.


Dimensions of the card are 17.0 x 7.0 cm, and it weighs 304 g.


Installation requires two slots in your system.


Display connectivity options include four Mini-DisplayPort 1.4 ports. A Mini-DP to DP cable is included in the package.


The card has no external power inputs. All power is supplied via the PCI-Express slot, which is specified for up to 75 W.

Teardown

When disassembling the card, the first piece that will come off is the black cooler shroud with the fan.


Here you can see how the fan works—it's a radial blower design that sucks air in through the fan hub and blows it across the heatsink, out of the card.


Under the shroud, we'll find a simple heatsink without any fancy heatpipes.


This is really just a block of metal with fins. The base looks well-designed, though. It provides cooling for the GPU and memory chips.

High-Resolution PCB Pictures

These 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, videos or forum posts.



High-res versions are also available (front, back).

Circuit Board (PCB) Analysis

For GPU voltage, a three-phase VRM configuration is installed, driven by an Richtek RT8845A controller.


The DrMOS for the GPU is made by Alpha and Omega, using the AOZ5311NQI, which we've seen on many recent graphics cards.


Memory voltage is a two-phase design generated by a uPI uP1666Q controller.


The DrMOS for the memory is made by Alpha and Omega, too, using AOE6932 MOSFETs.


The GDDR6 memory chips are made by Samsung and carry the model number K4Z80325BC-HC14. They are specified to run at 1750 MHz (14 Gbps GDDR6 effective).


NVIDIA's GA106 graphics processor is the company's third GeForce Ampere architecture chip. It is produced on a 8 nanometer process at Samsung and has a transistor count of 13.3 billion with a die size of 276 mm².