11.3 TNvidiaFlops ~= 15 TAmdFlops ?1080TI today is 11.3 TF.
7nm better be 1 hell of a node shrink.
Don't kill me, it's something I read on the internets.
11.3 TNvidiaFlops ~= 15 TAmdFlops ?1080TI today is 11.3 TF.
7nm better be 1 hell of a node shrink.
They will always develop hardware. They just want to ensure that anyone and everyone can access its content.Saw a headline the other day that MS intends to deploy game streaming in a couple of years.
Does that mean they will no longer develop new console hardware?
I guess it will depend on the sales of the X1X over the next few years.
no lol11.3 TNvidiaFlops ~= 15 TAmdFlops ?
Don't kill me, it's something I read on the internets.
Wait what’s a “today’s FLOP”? Aren’t flops all created equal?
I'm trying some 2019-2020 cost projections based loosely on past BOMs.
$120 for a 15TF SoC
$5/GB if using GDDR6
$5/GB if using HBM (samsung's low cost hbm)
$10 for an interposer
$150 rest of BOM, psu, odd, hdd, m/b, cooling, etc...
(A) Total BOM using GDDR6 384 bits (768GB/s):
18GB: $360
24GB: $390
36GB: $450
(B) Total BOM using low-cost HBM 4 stacks (800GB/s)
No idea what the interposer/assembly yield problem looks like, but I thought it would be fun to try a range of 90% to 99% per die. (4 stacks being 5 dies on the interposer)
90% yield: (^5 = 59%)
16GB: $505
32GB: $641
95% yield: (^5 = 77%)
16GB: $422
32GB: $526
98% yield: (^5 = 90%)
16GB: $394
32GB: $472
99% yield: (^5 = 95%)
16GB: $381
32GB: $455
16 GB HBM @ 800GB/s and an SSD (+HDD) would be very functional I think, and nicely progressable with larger SSD/flash options through the generation alongside HDD increases, and/or be user upgradeable for power users, and should cost reduce well (SSDs gaining the benefit of process shrinks as SOc and HBM would). If I was in charge, I'd be seriously considering that option. In fact I'm finding it hard to seriously justify another design without an SSD. Gobs of RAM and a slow ass HDD will be a terrible user experience.95% yield: (^5 = 77%)
16GB: $422
32GB: $526
98% yield: (^5 = 90%)
16GB: $394
32GB: $472
99% yield: (^5 = 95%)
16GB: $381
32GB: $455
11.3 TNvidiaFlops ~= 15 TAmdFlops ?
Don't kill me, it's something I read on the internets.
Probably (and hopefully) chips that are still in the making.
Unless you mean a *new* 3ds type upgrado ; so a *new* Nintendo switch. For which I could see them simply using tegra X2 and adding some ram.
For an out and out successor id expect something between ps4 and the pro (docked) with a better CPU. Assuming Nintendo keeps up with modern designs.
16 GB HBM @ 800GB/s and an SSD (+HDD) would be very functional I think, and nicely progressable with larger SSD/flash options through the generation alongside HDD increases, and/or be user upgradeable for power users, and should cost reduce well (SSDs gaining the benefit of process shrinks as SOc and HBM would). If I was in charge, I'd be seriously considering that option. In fact I'm finding it hard to seriously justify another design without an SSD. Gobs of RAM and a slow ass HDD will be a terrible user experience.
In fact, that's my prediction. Or rather vision, which no-one's going to follow because they don't listen to me (although Nintendo got close with my dockable console vision) - A semi-modular console with a base unit, HDD, SSD, and stacked RAM, maybe GDDR6 or such. Optional optical drive can be attached on the top in something infinitely more classy than Sega MegaCD... Going one step further, the SSD is replaced with two flash ports (XQD, CFast, whatever) under a cover at the front. Base unit is thus compact, only housing HDD and mobo. Optional ODD stacks on top for squarer console than usual - slim drive needn't add too much - so kinda Apple TV/Amazon Fire TV. Defers cost of movie playback to specific users, dropping entry level price. Power users can be wooed with the possibility of adding dual flash carts at whatever speed for faster operation, and replacing HDD with larger/faster drive. And for a final foray into dream-land, base units can be stacked to work in parallel replacing the need for mid-gen refreshes. As the generation progresses and the base unit gets cheaper, more users will buy additional stacks. A new console generation won't be needed for 15 years, simply because the average stack size of 24 units by then will be too top-heavy and liable to fall over.
It was late so I should have probably worded myself a bit better. I'm guessing a Switch successor would show up around 2021 or 2022. Where do you guys think Nvidia's mobile technology will be around that time?
I can't wait to see what Nvidia's mobile SoC successor to Xavier is, on 7nm or 7nm+ and post-Volta GPU architecture. Whatever that is, the Switch successor will probably use a downclocked version of around the time frame you mentioned.
If the Switch sells in confidence inspiring numbers, the volume is there to make a special version for Nintendo.I'm kind of curious though, since it seems most of the successors to the Switch's SoC are going to be for smart cars. I wonder if the Switch will be enough of a success to get Nvidia to put the time and effort into a new gaming focused SoC.
The fact that the thing can run Doom and Wolfenstein 2 (albeit with lower textures, resolution and half the frame rate) is impressive. It's kind of exciting to think of what the next Switch will be able to do.
16 of those 16Gb GDDR6 chips would give 32GB with a total bandwidth of 1024GB/s. Perfect harmony. Next gen redeemed. A world without sin.The fastest and lowest-power DRAM for next generation, graphics-intensive applications. It processes images and video at 16Gbps with 64GB/s data I/O bandwidth, which is equivalent to transferring approximately 12 full-HD DVDs (5GB equivalent) per second. The new DRAM can operate at 1.35 volts, offering further advantages over today’s graphics memory that uses 1.5V at only 8Gbps
16 of those 16Gb GDDR6 chips would give 32GB with a total bandwidth of 1024GB/s. Perfect harmony. Next gen redeemed. A world without sin.
512-bit bus?16 of those 16Gb GDDR6 chips would give 32GB with a total bandwidth of 1024GB/s. Perfect harmony. Next gen redeemed. A world without sin.
It'd make some sense to have a beefier CPU at the start of a gen if we're going with rolling generations from now on where they can just have a fatter GPU in the mid-gen refresh while sticking somewhat closely to the original CPU. The games would just scale the sampling amount automagically.Another thing that's becoming clear to me is that with the diminishing returns of resolution increases, I'd rather have a higher ration of cpu to gpu than they had with the ps4/xbox one era. There should be enough cpu that 60fps in open world games like GTA, Assassin's Creed is not an issue. Here they're talking about having potential cpu related frame drops in a small-scale linear game.