add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube

Next nvidia GPU scheduled for "a long time from now"

yawumpus
  • 22 months ago

https://www.hardocp.com/news/2018/06/04/dont_expect_new_geforce_gpu_for_long_time

It isn't entirely clear that what advantages they could produce by introducing a new Volta chip. Simply checking naive division: [chip] [process] [#Gtransistors] [area in mm**2] [Mtransitors/area] 1030 12nm 1.8 74 24.3 1050 12nm 3.3 132 25.0 [same core used for 1050ti] 1060 14nm 4.4 200 22.0 1080 14nm 7.2 314 22.3 [including 1070, 1070ti] 1080ti 14nm 12.0 471 25.4 [including Titan X, Titan XP] volta 12nm 21.1 815 25.9 [numbers from Titan V]

Don't expect to get any more transistors from Volta and TSMC's 12nm process (Volta needs more communication lines, but also the FP64 logic should be more dense than consumer FP32 logic leaving a wash [at least for numbers derived from wiki]).

Nvidia made volta for their supercomputing and AI customers, presumably for a reason. So maybe the 12nm gives sufficiently high performance to give it a run. But don't expect miracles, and don't expect a GPU any time soon. Note that TSMC just claimed to have started producing "mass volume" on its 7nm process: nvidia may be targeting that and possibly waiting to see how it comes out before putting in massive manhours on a chip [you know Moore's law is dead if a company can wait for the process to be available. This last bit is entirely speculation on my part and reading quite a bit into "a long time from now"].

Note - Nvidia stuck with 28nm before dropping to 12-14nm. Perhaps they want to hold pat until they can cut the pitch in half at 7nm (of course, they shipped a few generations at 28nm, but there has been little reason to spend the money since).

On the Vega front, rumors have it that GloFo will produce Vega20 in 7nm as a "pipecleaning" run. Presumably this means a professional Polaris job [don't expect any volume from GloFo, and I doubt they plan on wild success from HBM[2-3?] either] with a consumer [i.e. remotely sanely priced] Radeon for another year once you see the RadeonPro (don't hold your breath for GloFo). AMD is likely pleased as punch with Raven Ridge, and expect further GPUs to be built primarily for ALU integration (I'm assuming this has been true since they started making ALUs, coinciding with total concession of the high end GPU market to nvidia).

Comments

  • 22 months ago
  • 1 point

NVidia is basing the gaming GTX series off of the Volta architecture as well the tensor cores are needed for the newer graphics tech that they are working with Microsoft to integrate into the upcoming DX12 revision they are both throwing a ton of money at redesign of the API to better suit development and developers.

AMD is being held back by delays in GF 7nm ramp up because they have to completely redevelop how they scan for mistakes in the plates. Performance wise they are stuck with being limited to HBM2 bandwidth on an architecture that already needs more then what that can provide, so until a overhaul occurs to lower dependency on memory bandwidth, or HBM3 arrives things are pretty well stuck since Navi was going to reuse Vega architecture but scale it across the entire line-up.

Grim days for the graphics market for sure.

  • 22 months ago
  • 1 point

"limited to HBM2 bandwidth" is hardly a problem. Of course, I'd expect HBM[2-3] to be available (and working) before GloFo 7nm is ready to pump out a good supply of chips. Don't be too surprised if "a long time from now" nvidia supplies the next GPU with HBM (probably beating mainstream Vega), they are already using HBM2 on Volta.

The bandwidth issue does show up in Raven Ridge (which unfortunately is a fraction of the power of discrete Vega), so I suspect they will have to work on their memory efficiency (compress more textures/graphics).

  • 22 months ago
  • 1 point

"limited to HBM2 bandwidth" is hardly a problem.

The current Vega models are already limited in performance by HBM2 you gain more from memory overclocking to increase bandwidth then you gain from clock speed increases.

Don't be too surprised if "a long time from now" nvidia supplies the next GPU with HBM (probably beating mainstream Vega), they are already using HBM2 on Volta.

Unlikely HBM isn't really worth the money spent outside of very select uses.

HBM costs 5-6 times more per GB and you are limited to only using a package design like AMD is doing with Vega. Moving to GDDR6 is far more economical both in production and in the cost per GB.

The larger problem is the memory is still very early into ramp up and only from a single supplier currently, with the other FABs planning ramp up within a few months with no clear idea on mass availibility.

  • 22 months ago
  • 1 point

This makes sense, and it is actually also why I decided to stop waiting and pulled the trigger on getting a GTX 1070ti a week ago.

I was a follower of the rumors since the beginning of this year and had been holding out for next-gen, but being that there just isn't any significant challenger to Nvidia's current GPU generation, there is no reason for them to release anything new to us at the moment. Also, when next-gen launches, just imagine how many people are going to scoop up those cards quickly at prices higher than what we're seeing now for current gen.

I, myself, am a moderate-casual PC gamer. I figured picking up a 1070ti would meet my needs for the next 2-3 years. At that point, I'll upgrade again.

[comment deleted by staff]

Sort

add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube