Date: Tuesday , August 07, 2018
In July we began a journey looking into the past and comparing GPU performance from generation to generation. We began with our article: NVIDIA GPU Generational Performance Part 1 focusing on the GeForce GTX 780, GeForce GTX 980, and GeForce GTX 1080 video cards. If you have not read that evaluation, the introduction goes over very thoroughly about what this series of reviews is all about. We will briefly touch on our goals again, just as a reminder.
The topic of review is being divided into multiple parts so we can best facilitate testing. We are covering a wide range of GPUs, games, and game settings. We are covering NVIDIA and AMD, starting with NVIDIA. The purpose of this series is to find out how GPUs have evolved in terms of real-world gaming performance over the last five years. It should tell us if the upgrading we’ve done over the years actually means something and how much it actually means in terms of performance.
In our NVIDIA GPU Generational Performance Part 1 we found out many interesting things. The difference between GeForce GTX 980 and GeForce GTX 780 performance pales in comparison to what the GeForce GTX 1080 brought to the table. Even though Pascal was a refresh of Maxwell, the gains in performance were much larger than the jump from Kepler to Maxwell.
We also experienced that as newer games were released that were more demanding, the more GeForce GTX 980 got to stretch its legs. If you still have a GeForce 780 series video card, upgrading to GeForce GTX 1080 would be a massive improvement.
In today’s article, NVIDIA GPU Generational Performance Part 2, we are going to focus on the the GeForce GTX 780 Ti, GeForce GTX 980 Ti and GeForce GTX 1080 Ti.
The GeForce GTX 780 was launched on May 23rd, 2013 with an MSRP of $649. The GeForce GTX 780 Ti was launched six months later on November 7th, 2013 with an MSRP of $699. The GeForce GTX 780 Ti was the Kepler GK110 GPU fully realized with all 15 SMX’s enabled, beating the specs of the then GTX TITAN with only 14 SMX’s enabled.
The GeForce GTX 780 Ti is based on the Kepler architecture on a 28nm process. The GeForce GTX 780 Ti consists of 2880 CUDA Cores, 240 Texture Units and 48 ROPs. It runs at a baseclock of 875MHz and a boost clock of 928MHz. It has 3GB of GDDR5 on a 384-bit memory bus running at 7GHz. This gives It 336GB/sec of memory bandwidth. The TDP is 250W.
Three factors affect the effectiveness of the bump in CUDA Cores and Texture Units. One is that the ROPs remain at 48 just like the GTX 780. The L2 Cache also remains the same. Finally, the same limiting 3GB of VRAM capacity is on board just like the GTX 780. Though it has more memory bandwidth, with the ROP count, L2 Cache size and VRAM limitation, it is bottlenecked at certain points.
The GeForce GTX 980 was launched on September 18th, 2014 with an MSRP of $549. The GeForce GTX 980 Ti was launched eight months later on May 31st, 2015 with an MSRP of $649. Though this is a high-end GM200 Maxwell GPU, it sits below the GTX TITAN X of the time which is actually the GM200 Maxwell GPU fully utilized. This is different than what NVIDIA did above with the GeForce GTX 780 Ti being the fully realized Kepler GPU.
The GeForce GTX 980 Ti is utilizing 22 SMMs versus 24 SMMs on GTX TITAN X. This means more CUDA Cores and Texture Units on the GTX TITAN X. Therefore, now the GTX TITAN X is actually the fully realized Maxwell GPU, while the GTX 980 Ti is a slightly cut-down version. However, the price points are very different. GTX TITAN X was a $999 card while GTX 980 Ti was $649 and thus compares to the GTX 780 Ti in price as the natural upgrade path.
The GeForce GTX 980 Ti is based on the Maxwell architecture on a 28nm process. The GeForce GTX 980 Ti consists of 2816 CUDA Cores, 176 Texture Units and 96 ROPs. It runs at a baseclock of 1000MHz and a boost clock of 1075MHz. It has 6GB of GDDR5 on a 384-bit memory bus running at 7GHz. This gives It 336GB/sec of memory bandwidth. The TDP is 250W.
Its strengths are definitely the increase in ROPs to 96 versus the GTX 980’s 64. The bump in CUDA Cores and Texture Units also helps, even though it has a reduction in clock speed compared to GTX 980. Compared to GTX 780 Ti it has the same memory bandwidth, but an increased VRAM capacity of 6GB helps it more.
The GeForce GTX 1080 was launched on May 17th, 2016 with an MSRP of $599 or $699 for the Founders Edition. The GeForce GTX 1080 Ti was launched ten months later on March 9th, 2017 with an MSRP of $699. Compared to the GTX 780 Ti and GTX 980 Ti, the GTX 1080 Ti is a different beast. Instead of being based on GTX 1080’s GP104 GPU, GTX 1080 Ti is an entirely bigger GP102 GPU made to deliver a lot of performance. It is still the same Pascal architecture though.
The GeForce GTX 1080 Ti matches the GTX TITAN X (Pascal) video card in CUDA Cores and Texture Units. However, the memory subsystem of the GeForce GTX 1080 Ti is cut-down from the TITAN X (Pascal) GP102 GPU. In fact, if you take GTX TITAN X and cut out one ROP block cutting back on its ROPs, L2 Cache and memory bus width and VRAM capacity, you then get the GeForce GTX 1080 Ti. Priced at $699 it is the direct upgrade path from GeForce GTX 980 Ti and comparable.
The GeForce GTX 1080 Ti is based on the Pascal architecture on a 16nm FinFET process. The GeForce GTX 1080 Ti consists of 3584 CUDA Cores, 224 Texture Units and 88 ROPs. It runs at a baseclock of 1481MHz and a boost clock of 1582MHz. It has 11GB of GDDR5X on a 352-bit memory bus running at 11GHz. This gives It 484GB/sec of memory bandwidth. The TDP is 250W.
We are using the same games from Part 1. The games we have chosen are, from earliest to latest: Crysis 3 2013, Tomb Raider 2013, Grand Theft Auto V 2013, Far Cry 4 2014, The Witcher 3 2015, Fallout 4 2015, Rise of the Tomb Raider 2016, DOOM 2016, Deus EX Mankind Divided 2016, Battlefield 1 2016, Sniper Elite 4 2017, Mass Effect Andromeda 2017, Kingdom Come Deliverance 2018, Far Cry 5 2018.
We are going to test both 1440p and 4K resolutions in each game, on each video card.