VRAM on RTX 40s is as rare as life on Mars, and to get cards with a fair amount of memory, you’ll have to turn to the brand’s professional offering ?. Hairy troll aside, NVIDIA recently announced new graphics cards for professionals. These are the RTX 5000 Ada, RTX 4500 Ada and RTX 4000 Ada.
RTX 4500 Ada & RTX 4000 Ada: the smallest cards!
With its 4500 and 4000 models, the Chameleon offers us cards equipped with an AD104 GPU. However, depending on the model, the chip contains more or less cuda core. As you can see, with its 60 SM, the 4500 is the bigger of the two, containing 7680 cuda cores. On the other hand, we have the 4000 with 48 SM and 6144 cuda core. Finally, the other interesting data concerns the number of tensor cores, which are 240 and 192 respectively.
Compared to the GeForce series, we find a large amount of VRAM, which is logical given the intended use. The 4500 Ada offers 24 GB of VRAM, a 192-bit memory bus and a bandwidth of 432 GB/s via its 18 Gbps memory.
The 4000 Ada has 20 GB of memory and a small 160-bit bus. Bandwidth is de facto reduced: 360 GB/s. It’s a small card, all the more so as its GPU is the slower of the two: 2200 MHz vs. 2600 MHz for the 4500. On the other hand, it has the advantage in terms of power consumption/heating (130W vs. 210W) and format ( single slot). On the other hand, both models require a 12VHPWR power supply.
This RTX 4000 Ada is the full-size version of the RTX 4000 SFF Ada. It features the same GPU and memory configuration. However, performance is a good notch higher, thanks to a GPU less hampered by power consumption limits linked to its format. This is evident in its frequencies alone: 2200 MHz vs. 1560 MHz.
RTX 5000 Ada: the biggest of the three!
With this RTX 5000 Ada, we’re moving up a gear, since it’s an AD102 GPU rather than an AD104. As a result, it contains more cuda cores and tensor cores , with 12800 cuda cores and 400 tensor cores. With its 2500 MHz frequency, its FP32 performance is also higher: 65.3 TFLOP versus 39.6 TFLOP for the 4500 and 26.7 TFLOP for the 4000.
Memory is also important, with 32 GB GDDR6 running at 18 Gbps like the others. On the other hand, the 256-bit wide bus enables a bandwidth of 578 GB/s. Nevertheless, this is a far cry from the RTX 6000‘s 960 GB/s.
This time, we’re dealing with a dual-slot card cooled by a blower. The thermal envelope of this model is 250W and we’re still entitled to a 12VHPWR power socket.
Pricing:
Inevitably, as soon as you get into the professional range, the bill skyrockets. Here’s a short list of the cards in the series:
- RTX 6000 Ada: $6800
- RTX 5000 Ada: $4000
- RTX 4500 Ada: $2250
- RTX 4000 Ada : $1250