Nvidia's RTX 4000 SFF Ada Generation graphics card targets compact workstations, so it's unlikely that you'll find it on the list of best graphics cards for gaming. That doesn't mean that the graphics card sucks, though. A recent review from the Japanese publication Jisaku Hibi shows the RTX 4000 SFF performing very close to a GeForce RTX 3060 Ti, with 65% lower power consumption.
The RTX 4000 SFF comes with the AD104 silicon, which powers the GeForce RTX 4070 and GeForce RTX 4070 Ti. However, the RTX 4000 SFF's die only has 48 fully-enabled Streaming Multiprocessors (SMs) out of the 60. This puts the Ada Lovelace graphics card's CUDA core count at 6,144, 4.5% more than a GeForce RTX 4070 but 20% less than the GeForce RTX 4070 Ti. In addition, the RTX 4000 SFF is a workstation graphics card with a small form factor; therefore, gaming isn't its top priority.
At 70W, the RTX 4000 SFF comes with very modest clock speeds. The graphics card operates with a 1,290 MHz base clock and can boost up to 1,565 MHz. However, the amount of memory on the RTX 4000 SFF is its strongest suit. The graphics card comes with a whopping 20GB of GDDR6 memory.
Sadly, the memory bandwidth doesn't do the graphics card justice. Limited to a 160-bit memory interface, the RTX 4000 SFF's 14 Gbps GDDR6 memory chips can only deliver a memory bandwidth of up to 280 GB/s.
Jisaku Hibi's gaming-focused review on the RTX 4000 SFF puts the Ada-based graphics card through its paces in 15 titles, such as Cyberpunk 2077, Marvel's Guardians of the Galaxy, Death Stranding, and others. The publication tested the RTX 4000 SFF across three resolutions with different image fidelity settings.
Nvidia RTX 4000 SFF Ada Generation Benchmarks
|Graphics Card||1920 x 1080 Gaming||2560 x 1440 Gaming||3840 x 2160 Gaming|
|GeForce RTX 4070||155%||164%||170%|
|GeForce RTX 3070||121%||126%||133%|
|GeForce RTX 3060 Ti||110%||113%||0%|
|RTX 4000 SFF Ada Generation||100%||100%||100%|
|GeForce RTX 3060 12GB||83%||86%||0%|
|GeForce RTX 3050 (∼ RTX A2000)||61%||62%||0%|
The GeForce RTX 4070 outperformed the RTX 4000 SFF by a considerable margin. The last-generation GeForce RTX 3070 had no problems besting the Ada-powered workstation graphics card, either. The GeForce RTX 3060 Ti, a SKU Nvidia is rumored to be discontinued, was a bit faster than the RTX 4000 SFF, the margins were less than 15%. It shows goes to show the prowess of Nvidia’s Ada Lovelace architecture.
Remember that the the GeForce RTX 3060 Ti is a 200W graphics card. Therefore, it is very impressive how the margin between the RTX 4000 SFF and the GeForce RTX 3060 Ti wasn’t very significant despite the former having a 65% lower TDP.
The GeForce RTX 3060, which is a 170W SKU, isn’t a pushover, either. Nonetheless, the RTX 4000 SFF surpassed the Ampere graphics card by a comfortable margin (20% average). When it comes to generation-over-generation uplifts, the RTX 4000 SFF, on average, was about 70% faster than the RTX A2000, whose performance is similar to that of the mainstream GeForce RTX 3050.
The RTX 4000 SFF has a MSRP of $1,250 so no one in their mind would purchase it for gaming. Nonetheless, it’s good to know that the RTX 4000 SFF is a capable gaming graphics card when professionals need to take some R&R time from their work.
It appears that this card has different specs, especially the bandwidth value, which is stated as 320 GB/s in some spec sheets, whereas 280 on others. I checked multiple websites/vendor sites like PNY, Leadtek, and there are conflicting bandwidth values.
I think this could be due to card using either 14 Gbps GDDR6 memory chips, or 16 Gbps. Which would give us the exact bandwidth value of 280 or 320 GB/s. Maybe there are 2 SKUs Nvidia is selling.
But I'm sure there are 2 variants in the market. Which is kind of weird, since there has been no official confirmation from NVIDIA on the usage of different memory chips.
Here it says 280GB/s.
Also not clear: is this just a laptop part repackaged on a PCIe card?