Nvidia RTX 3090, RTX 3080 and RTX Titan specifications leaked – GDDR6X?

Specifications for Nvidia’s planned Ampere series of graphics cards have started appearing online, many of which detail a new memory type called GDDR6X, and a graphics card called the RTX 3090. 

At this time, we cannot confirm the validity of these specifications. Details of the Ampere series remain unknown to most Nvidia employees, making the best response from practically all Nvidia representatives be the industry standard “we don’t comment on rumours”. Take these specifications with a grain of salt, as while they have been independently reported by two sources, @KkatCorgi and @kolite7kimi (as detailed by Videocardz), it is possible that these specifications originated from a single, possibly false, source. 

Why is there no RTX 3080 Ti? 

Rumour has it that Nvidia plans to kill off its XX80 Ti branding. Simply put, the RTX Super series set a confusing precedent for consumers. Is the RTX 2080 Super faster, or is the RTX 2080 Ti faster? While hardware enthusiasts will know the difference, most consumers will not, and this has caused Nvidia to rethink its Ti branding. 

With multi-GPU being an almost dead technology, Nvidia has room to re-adopt their X90 series brand name to replace their confusing Ti branding. This streamlines Nvidia’s marketing and allows future “Super” edition cards to be released without as much consumer confusion. 

GDDR6X

At this time, no memory manufacturer has revealed GDDR6 series memory modules. That said, GDDR6X could simply be a brand name for faster GDDR6 memory modules, similar to HBM2E. The fastest GDDR6 memory chips inside today’s consumer graphics cards are rated at 16Gbps, though Samsung claims to have 18Gbps modules available.    

Faster graphics cards will require more memory bandwidth, and there are two ways to achieve this. GPU makers can either go wide or go tall. Going wide requires more memory chips over a larger memory bus and going tall requires faster memory chips over a smaller memory bus. With GDDR6X, Nvidia is going tall by using a small memory bus with faster GDDR6X memory chips. 

Right now, we do not know if GDDR6X even exists, and this fact casts a huge shadow over these rumoured specifications. That said, who heard about GDDR5X before Nvidia revealed its GTX 10 series of graphics cards? 

RTX 3080

Looking at the chart below, we can see that Nvidia’s rumoured RTX 3080 offers gamers the same number of CUDA cores as Nvidia’s RTX 2080 Ti and delivers gamers significantly more memory bandwidth.

With Nvidia’s RTX 2080 Ti being both bandwidth and power-constrained in many of today’s games, it is easy to see how Ampere has the potential to deliver significant performance improvements over Turing. 7nm manufacturing and architectural enhancements will increase Ampere’s efficiency, performance per clock cycle, and raw clock speeds over Nvidia’s Turing-based RTX 2080 Ti. These factors should make Ampere a lot more powerful than Turing, at least on the high-end.   

 
 

Â	RTX Titan (Ampere)	RTX 3090 /3080 Ti	RTX 3080	Titan RTX (Turing)	RTX 2080 Ti	RTX 2080 Super	RTX 2080
GPU	GA102-400	GA102-300	GA102-200	TU102-400	TU102-300	TU104-450	TU104-400
CUDA Cores	5376	5248	4352	4652	4352	3072	2944
Memory	24GB GDDR6X	12GB GDDR6X	10GB GDDR6X	24GB GDDR6	11GB GDDR6	8GB GDDR6	8GB GDDR6
Memory Speed	17Gbps	21Gbps	19Gbps	14Gbps	16Gbps	15.5 Gbps	14 Gbps
Memory Bus	384-bit	384-bit	320-bit	384-bit	352-bit	256-bit	256-bit
Memory Bandwidth	816GB/s	1008GB/s	760 GB/s	672 GB/s	616 GB/s	496 GB/s	448GB/s