nVidia could simply point to a "modest" 1.6x performance upgrade and aim to have its SoC available to phablets and larger smartphones.
Could be. Realistically the Shield tablet has already broken the 30fps "barrier" in Manhattan and it doesn't look like there's all that much in sight to surpass that sort of performance any time soon. The question would be who needs over twice as much even just in theory next year?
I'll admit that I don't know too much about the mobile gaming scene but is anyone actually asking for faster graphics on their phone or tablet? If they are asking for it my next question is what games are they playing.
nVidia's console mode is an interesting approach to driving demand for faster SoC's but for mainstream usage I don't quite get it. At some point good enough is going to be good enough.
@ Blazkovicz: Erista will need two SMM's (ie. 256 CUDA cores in total) to actually significantly improve in performance vs. Tegra K1. Theoretically one SMM is possible, but I just don't see them going that route @ the 20nm fab. process node where transistor density is vastly improved vs. 28nm HPM.
Are we ruling out the possibility of one Erista chip with 1 SMM and another with 2 SMMs?
So did Xiaomi sell 50K MiPads in a few hours because of anything else but "Keplar" inside?
If you look at the latest "highend" china-phones they start to use the new MT6595 a lot. Maybe the Antutu benchmark results of >45000 are enough for them.
yes, it seems nvidia didn't want to use Denver for servers. So they have the SoC but now instead of pushing it they stopped. This seems to imply that they have given up on servers.
Peak GPU clock operating frequency
Tegra K1: 852MHz
Tegra M1: 1038MHz
Performance improvement with Tegra M1 Erista: 1.218x
Overall graphics performance improvement with Tegra M1 Erista vs Tegra K1 Logan at the same power envelope: 2.27x !!!
Would Erista simply have one SMM?
Keep it simple for lowest complexity and power use.
Are we ruling out the possibility of one Erista chip with 1 SMM and another with 2 SMMs?
That's kind of his point..Nvidia has stated that they will not compete in this market and will target high end only..so a 1 SMM part is not good enough IMHO.Erm, a putative 1 SMM chip would still have extremely good GPU performance for the low-end to mid-range market, I'd have thought?
1 or 2 SMM doesn't really make all that much of a difference considering the whole SoC size. So, unless it's a whole different SoC, it's probably not worth the trouble unless the volume is extremely high. Just disable 1 SMM if you want to sell some lower end chip...
I agree this could be an option, I just didn't agree that two versions made sense.If NV keeps the same strategy as with K1 yes they probably used 2 clusters; if however they're aiming for a rather conservative overall performance increase but a quite high perf/W difference 1 cluster could also be a viable scenario.
FWIW I don't share the views of some here it will be a >2x perf/w improvement based on some extrapolated desktop numbers
The improvement in GFLOPS per watt is ~ 1.625x for Tegra M1 Erista compared to Tegra K1 Logan. So at the same power consumption as Tegra K1 Logan (which has a peak throughput of ~ 327 GFLOPS at a peak GPU clock operating frequency of ~ 852MHz with 192 Kepler CUDA cores), Tegra M1 Erista will have a peak throughput of ~ 531 GFLOPS at a peak GPU clock operating frequency of ~ 1038MHz with 256 Maxwell CUDA cores).
GM2x0 cores have 40% higher ALU efficiency than Kepler. All slides for GM107 claimed a 35% improvement. Yes it's hairsplitting but I think it's likelier they stay at first Maxwell generation level.Since a Maxwell CUDA core has ~ 1.4x higher graphics performance than a Kepler CUDA core, my estimate for graphics performance improvement of Tegra M1 Erista at the same power level as Tegra K1 Logan is ~ 2.27x.
That hopefully yes; and that obviously will benefit the GPU (amongst all other SoC units) irrelevant of the amount of clusters it'll have after all. Again the 2SMMs are very likely, however I'm not sure if they have something completely different in mind this time.On a side note, Tegra M1 Erista should also have significantly higher peak memory bandwidth and higher memory bandwidth efficiency vs. Tegra K1 Logan too.
It states single precision GFLOPs/W Normalized, for which Lord knows what normalized stands for. Just for the record's sake if they'd really count it as you do it would mean that with 327 GFLOPs and 40 GFLOPs/W it would consume almost 8.2W for the GPU alone.