The intro of GeForce Titan puts to rest the doubts whether Nvidia would be able to release – at least in limited quantities – a gaming version of the Tesla-focused GK110. What impact does it have on the competition?
Firstly,yes, it is expected to equal or beat AMD’s HD7970 GHz Edition in almost every performance aspect: from raw GPU and FP (both SP and DP) performance to memory capacity and bandwidth as well. After all, this is the first ‘standard issue’ single GPU card for PCs with 6 GB memory as standard, if you exclude Sapphire HD7970 Toxic 6GB, which was a very special edition.
Second, with Nvidia finally playing fair on the DP FP issue and not crippling the raw DP FP rate compared to the Tesla siblings, the general purpose high-end PC users can also now enjoy many of the benefits of CUDA & OpenCL accelerated apps in both single and double precision. The amount of memory on board will particularly help GPU compute, as there’ll be far less need to rely on performance-sapping PCIe comms once most or all of code and data is in GPU memory.
Third, if the yields improve further over time for the GK110, and Nvidia becomes able to launch at least a limited edition of All-Cores-Enabled version (GK110 GPU contains 15SMX clusters for a total of 2880 units), a la GTX580 on the Fermi side, there’s still another 5% – 10% performance uplift potential there depending on the clock speeds attainable with further steppings.
Since, as of now, AMD has no plans to release a high end ‘Sea Islands’ follow on to the HD7970 within at least next six months (they have to get rid of all of stock), the rollout of Titan gives back Nvidia the undisputed GPU-for-GPU performance title for that period. The often seen ‘waterfall’ effect in the GPU marketplace, where having the top level record (even for small quantities) helps your lower end GPUs sell better too, will likely help Nvidia in this case, unless AMD decides to be very aggressive and either fights a price war with mainstream ‘Sea Islands’ GPUs or brings forward the HD89X0 high end GPU intro.
This all matters a lot since, in 4 months time when the Haswell CPUs roll out, with their vastly improved integrated GPU, the low end of the discrete GPU market will be under immense pressure – having some performance to show off on high end will help Nvidia justify further the discrete GPU approach overall in the PC realm.