Nvidia adds three new models with 20 to 32 GBytes of memory to its current lineup of Ada-generation RTX workstation graphics cards.
Nvidia's current graphics chips of the Ada-Lovelace generation have so far mainly been used on gaming graphics cards of the GeForce RTX 4000 series with its latest offspring, the RTX 4060. On the workstation graphics cards of the RTX series, the largest Ada chip AD102 has so far only been used on the top model RTX 6000 Ada with 48 GBytes, while Nvidia has continued to equip cheaper models with the graphics chips from the previous generation "Ampere".
During the opening speech of this year's Siggraph, Nvidia boss Jen-Hsun Huang now presented three more RTX Ada cards, which are supposed to bring the new technology into price segments below circa 9400 Euros for the RTX 6000 Ada. The Nvidia RTX 5000 Ada with 32 GB, the RTX 4500 Ada with 24 GB and the RTX 4000 Ada with 20 GB graphics memory. They not only succeed the RTX Ampere series, i.e. RTX A5000 & Co., but also have to compete against AMD's Radeon Pro W7000, of which four variants including the low-priced Radeon Pro W7600 and W7500 have been released so far.
High computing power and efficiency, slower memory
Nvidia claims that the cards are 50 to 100 percent faster in classic graphics, AI and rendering applications compared to their respective numeric predecessors with Ampere GPUs. In the Omniverse environment, which will also receive DLSS-3 frame generation with an update, they are said to be up to three times faster.
Even though they carry 20 to 33 percent more memory, the new ones have to take a back seat when it comes to memory transfer rates. Compared to the RTX A5000 with Ampere GPU, the RTX 5000 Ada has a transfer rate that is a quarter lower (576 to 768 GByte/s), the 4500s are minus 32.5 percent and the 4000 models are 20 percent less. As with the GeForce gaming graphics cards, Nvidia wants to compensate for this with the significantly larger Level 2 cache of the Ada chips - which has worked well in games so far.
The new RTX 5000 Ada is supposed to cost around 4000 US dollars without taxes. This puts it in the range of AMD's Radeon Pro W7900, but it adds 48 GB instead of just 32 GB. The RTX 4500 Ada comes with 24 GBytes for around 2250 US dollars plus taxes and thus has to compete against AMD's Radeon Pro W7800 with 32 GBytes. These two RTX cards have its dual-slot cooler and swallow nominally 250 and 210 watts, respectively, so they are a bit more frugal than their AMD counterparts. Compared to their own predecessors, the cards are thirstier by 20 (RTX 5000 Ada) and 10 watts (RTX 4500 Ada), respectively.
The RTX 4000 Ada is a significantly slimmed down version of the RTX 4500 Ada, whose TDP has been trimmed down to 130 watts, which is about 62 percent compared to the 4500 Ada. This value is sufficient for a single-slot cooler, which is also offered by the much cheaper Radeon Pro W7600, but it is also much slower. Compared to the Nvidia RTX A4000 from the previous generation, that is 10 watts less.
All new RTX Ada cards have the new 16-pin connector of the ATX 3.0 specification. There are four full-size display ports of version 1.4a in the slot bezels. That is sufficient (respectively) for example for 4K with 120 Hz and DSC, 5K with 60 Hz and DSC or for a total of two times 8K60 with DSC. Combinations like 4 × 5K (60 Hz), 2 × 8K (60 Hz) or 1 × 12K (120 Hz with DSC) are reserved for the Radeon Pro W7800/7900 with their DP2.1 ports.
|RTX 6000 Ada||RTX 5000 Ada||RTX 4500 Ada||RTX 4000 Ada|
|Shader multiprocessors / RT cores||142 / 142||100 / 100||60 / 60||48 / 48|
|Shader cores (boost clock, approx.)||18176 (2505 MHz)||12800 (2550 MHz)||7680 (2578 MHz)||6144 (2172 MHz)|
|Computing power FP32 / Tensor (FP8, Sparsity)||91,1 / 1457 TFlops||65,3 / 1044 TFlops||39,6 / 634 TFlops||26,7 / 427,6 TFlops|
|NVEnc/-Dec||3 / 3||2 / 2||2 / 2||2 / 2|
|Graphic memory||48 GB GDDR6 (ECC)||32 GB GDDR6 (ECC)||24 GB GDDR6 (ECC)||20 GB GDDR6 (ECC)|
|Memory interface||384 Bit||256 Bit||192 Bit||160 Bit|
|Memory transfer rate||960 GByte/s||576 GByte/s||432 GByte/s||360 GByte/s|
|TDP / Power connection||300 Watt / 16-Pol CEM5||250 Watt / 16-Pol CEM5||210 Watt / 16-Pol CEM5||130 Watt / 16-Pol CEM5|
|Display outputs||4 × DP1.4a|
|Price||ca. 9400 Euro||4000 US-Dollar||2250 US-Dollar||1250 US-Dollar|
The Nvidia RTX 5000 Ada should be available immediately, with the RTX 4000 Ada following in September and the RTX 4500 Ada in October. Unlike their predecessors, the new cards consistently no longer support pairing via NVLink.