Few tech manufacturers have the marketing skills of Nvidia, whose work in AI and computer graphics is transforming industries now valued at more than $100trillion.
In terms of stoking expectation, the California-based inventor of the games processing unit (GPU) is surely peerless.
It’s handling of the imminent release of its new RTX 40 series of graphics cards has been a masterclass in information management, where less is certainly more.
Leak and rumour are the firm’s stock-in-trade, saying little, neither confirming nor denying anything and allowing its customers – of whom there are many – to whip themselves into a frenzy of anticipation.
Speculation is rife that the company’s enigmatic CEO Jensen Huang will use the upcoming GTC conference, on September 20, to make an announcement regarding the launch, at least, of the RTX 4090.
Typically, all that has been said to date about the boss’s contribution to the event, is that he will deliver a keynote speech on ‘the future of computing.’
The timing of the company’s switch to next-generation models has been complicated by a massive over-production of its existing GeForce RTX 30 series.
Despite it being hugely popular, there remains a large volume of unsold inventory sitting in Nvidia warehouses around the world, which the company wants to shift before unleashing an improved range of products on the market.
So, in lieu of any more definitive information, all we can say for sure at the moment is that the RTX 4090 is expected to be officially launched at the GTC, with a release date likely to be later in the year.
While it is hoped Huang may touch on other GPU models, such as the 4080 or 4080 Ti, during his address, no official launch or release dates for these models is expected.
The GeForce RTX 30 Series provided a significant jump in performance compared with the RTX 20 Series, and the expectation is that we could be seeing a similar jump to the RTX 40 Series.
This is fantastic news for us at G2 Digital and our customers as we plan to fit these new cards in our 2U, 3U & 4U units.
We’re particularly excited about using them in our new 4U PC which will be able to take 4x Dual width GPUs.
Leaks and rumours from Nvidia HQ abound and so any information currently online about the new series (codenamed Lovelace) should not be taken as gospel.
That said, despite the improvement in performance offered by the RTX 30 Series over the 20 Series, the RTX 40 Series is expected to offer a similar boost, compared with the current top-end RTX 3090 – a graphics card with an MSRP of £1399/$1499.
Little is known about the mid-tier next-gen graphics card, but it’s suggested the RTX 4070 could feature 10 or 12GB of slightly slower GDDR5 VRAM clocked at 18Gbps, with a 300W TDP.
That would represent a 2/4GB boost to overall VRAM compared with the 8GB GDDR6X 3070 Ti, though with slower memory and slightly slower clock speeds.
The GeForce RTX 4080 is a conflated configuration with the same cores as the existing GA102 part but tuned with much higher frequency or between 9000-1000 cores that allow some room for a ‘Ti’ variant in the future with the main configuration. The GPU will be equipped with 64 MB of L2 cache and up to 224 ROPs.
Clock speeds are expected to be between 2.0-3.0 GHz, with a higher than usual bump because of the two-node jump – Ampere GPUs with a Samsung 8nm node was, in reality, a 10nm process node with some optimizations.
Nvidia is skipping a 7nm and going straight for an optimised version of a 5nm node. With Pascal on the TSMC 16nm node, it delivered a huge frequency leap and we can expect a similar jump this time around too.
As for memory, the 4080 is expected to have 16 GB GDDR6X capacities at faster 21 Gbps speeds across a 256-bit bus interface, providing up to 672 GB/s of bandwidth.
All these boosted specifications will require a higher power draw with the 4080 expected to require TBP of around 420W. A single 16-pin Gen 5 connector should be enough for both reference and custom models.
We can assume to be true reports that a 5nm process node will run the 4090’s GPU architecture, offering a two-times jump in performance. What isn’t known for sure is whether it will be based on a standard N5 or N5P node.
It is also claimed that the GDDR6X memory speed on the 4090 has been increased from 21Gbps to 23Gbps, while retaining 16GB of VRAM. This implies a 384-bit memory bus and upwards of 1TB/s bandwidth
There are conflicting rumours regarding power consumption, with the latest suggesting total board power (TBP) for the 4090 has been reduced from 420W to 340W – lower than the 350W seen with the last-gen 3080.
It’s also rumoured that the new series could feature clock speeds between 2.2 and 2.5GHz (boosted), offering a decent jump from the current 1.7-1.9GHz clock speeds on offer from the current Ampere architecture.
Another leak suggests the 4090 will feature 126 Streaming Multiprocessors, giving 16,128 CUDA cores.
Despite the top-end spec, it is expected to feature the same 450W TDP as its predecessor, rather than the expected 600W. Another rumour suggests Nvidia could be keeping back an RTX 4090 Ti follow-up, with a rumoured 144 Streaming Multiprocessors bringing the CUDA core count up to 18,432.
The same source claimed the RTX 4090 desktop GPU scored higher than 19,000 points in 3DMark’s Time Spy Extreme benchmark test.
If this is accurate, it suggests the RTX 4090 is the most powerful Nvidia card to date, scoring almost double the standard RTX 3090 and outscoring the current number one mark achieved by an LN2-cooled EVGA RTX 3090 Ti Kingpin Edition by some 30%.