Nvidia is expected to launch new GPUs this summer based on its Turing microarchitecture, and we’re already seeing specifications and capabilities popping up online. We recommend taking all such rumors with a grain of salt; we’ve already seen the rumor mill be wrong this year when it predicted Nvidia would launch new consumer GPUs in March or April, so it’s entirely possible that these specifications will change over time.
The GPU Wccftech describes is structurally similar to the GTX 1080 Ti, with a few significant departures. The rumored design is a 3584:224:64 architecture (GPU Cores:Texture Mapping Units:Render Outputs). For comparison, the GTX 1080 is a 256:160:64 configuration, while the GTX 1080 Ti is 3584:224:88. The GTX 1180 supposedly has a 256-bit memory bus, however, compared with the 352-bit bus on the GTX 1080 Ti.
At first glance, our hypothetical GTX 1180 may seem a bit underpowered compared with the GTX 1080 Ti. Remember, however, that Nvidia is expected to use GDDR6 memory for its new high-end cards, and GDDR6 can sustain substantially faster transfer rates. We don’t know which memory chips Nvidia will use, but the 1080 Ti is outfitted with 11Gbps memory. Even 14Gbps GDDR6 would represent a 1.27x bandwidth improvement, while 16Gbps chips offer 1.45x more bandwidth per IC. The GTX 1080 Ti’s bus is 1.38x wider than the GTX 1080’s, which means a GTX 1180 on a 256-bit memory bus with 16Gbps RAM would actually offer more bandwidth than the GTX 1080 Ti (512GB/s versus 484GB/s). Wccftech predicts that the GTX 1180 could offer performance 1.51x higher than the GTX 1080 and 1.18x faster than the GTX 1080 Ti. Such predictions should be taken with their own heaping tablespoon of salt, but the targets at least make sense. A 1.51x performance improvement over a bit more than two years is pretty reasonable, particularly in comparison with the CPU market, where AMD’s ~10 percent performance improvement with second-generation Ryzen is considered excellent for the market.
The rumored price of $699 for the GTX 1180 will sit less well with people, given that the GTX 1080 ran $600. Top-end GPUs have indisputably been getting more expensive, and it looks like that’s going to continue. Some of the price leap may be due to the cost of GDDR6, but Nvidia also saves money on a smaller die when it moves to 7nm. The more likely explanation for any cost increases (assuming they appear) is that Nvidia is taking more profit. Over the past year, AIBs and companies like Newegg and Amazon have made a killing jacking up the price of GPUs and pocketing the difference. If demand for cards continues to be strong due to the influence of the cryptocurrency market, it only makes sense that NV would attempt to pull some of that cash in for itself.
As for any competitive response from AMD, we have no reason to expect one. The only telegraphed new part from AMD this year is a 7nm die shrink of Vega focused on machine learning markets. It’s possible AMD could be playing its cards close to the chest and planning some kind of tweaked Vega launch, but it doesn’t seem likely. Assuming that there’s no low-hanging fruit AMD could pluck for a refresh, the best we could hope would be a modest clock bump and mild power efficiency improvement. A ~10 percent gain would be welcome, but it wouldn’t be enough to put AMD on top.
This doesn’t mean that AMD is without options. One thing we saw with the older Fury X family is how lowering performance targets slightly can dramatically improve power consumption and efficiency. A new Vega Nano targeting, say, Vega 56 performance, coupled with aggressive pricing, could make AMD an attractive upper midrange option, even if the GTX 1180 claims the top of the performance stack.
Should Spectre, Meltdown Be the Death Knell for the x86 Standard?
Spectre and Meltdown are serious CPU flaws, but do they warrant throwing out the entire closed-source CPU model?
What is Speculative Execution?
Speculative execution has been in the news of late, typically when discussing the Meltdown and Spectre bugs. We explain the topic.
Microsoft Pulls Spectre, Meltdown Patches for Older AMD Systems
Microsoft has halted the distribution of patches for Spectre, Meltdown on some AMD systems due to problems with BSODs and bricked systems.
Samsung Announces High-Speed HBM2 Breakthrough, Codenamed Aquabolt
Samsung has found a way to turn up the speed on HBM2 dramatically without needing a voltage increase. Could this be the turning point for the standard?