IMHO:
* Apple A11 won't be the only 10FF chip for next year; see MTK Helio X30.
* Xavier has its reasons for being announced as early but predicted sampling in Q4 2017 means that mass production might start significantly later.
* Assume GV100 (HPC only) is truly slated for 16FF+, then chances are truly good for it to launch before next year runs out.
* Is 10FF TSMC really worth the hussle/increased costs, or will the 20SoC saga repeat itself and we'll see a direct jump for GPU chips from 16FF to 7FF?
* Yes I could imagine something like "1170/1190" in the second half of 2017, but I'd be VERY surprised if it would be anything but Pascal "reloaded" and obviously at 16FF+.
* Apple A11 won't be the only 10FF chip for next year; see MTK Helio X30.
* Xavier has its reasons for being announced as early but predicted sampling in Q4 2017 means that mass production might start significantly later.
* Assume GV100 (HPC only) is truly slated for 16FF+, then chances are truly good for it to launch before next year runs out.
* Is 10FF TSMC really worth the hussle/increased costs, or will the 20SoC saga repeat itself and we'll see a direct jump for GPU chips from 16FF to 7FF?
* Yes I could imagine something like "1170/1190" in the second half of 2017, but I'd be VERY surprised if it would be anything but Pascal "reloaded" and obviously at 16FF+.
Blazkowicz
Legend
I'm thinking Nvidia is leaving the tablet market - but makes a custom chip for Nintendo with Pascal, and the stuff for industrial connectors etc. left out.
D
Deleted member 13524
Guest
You're suggesting they haven't already?
Parker has a very odd CPU core arrangement that seems to be focused on GPGPU and driving external GPUs. Xavier seems to have a substantial portion of its area for hardware dedicated to stuff that would be useless for a tablet.
Is the Tegra K1 or even X1 still being produced? As soon as 16FF solutions catch up on GPU performance next year, why would these SoCs keep being made?
Blazkowicz
Legend
Parker bulks up some but it might conceivably used in high end tablets (like Google Pixel, and there's rumor of some Pixel tablet/laptop hybrid). You find high end Android SoC and x86 on that market.
This baby's really big : it's there not so much to drive an external GPU, but rather to replace it.
Only Tegra 4 and 4i (Tegra 1 and 2 too I assume) couldn't support external GPUs.
Kudos if you've seen it coming, but this is the one that really doesn't care about mobile. Parker doesn't really have PR targeted about something other than cars perhaps, but if it's called the "Tegra X2" it might get some limited use (with its own share of RAS and industrial features shutdown, disabled, restricted)
This baby's really big : it's there not so much to drive an external GPU, but rather to replace it.
Only Tegra 4 and 4i (Tegra 1 and 2 too I assume) couldn't support external GPUs.
Kudos if you've seen it coming, but this is the one that really doesn't care about mobile. Parker doesn't really have PR targeted about something other than cars perhaps, but if it's called the "Tegra X2" it might get some limited use (with its own share of RAS and industrial features shutdown, disabled, restricted)
D
Deleted member 13524
Guest
IMO, the fact that there's no "ninja core" or LITTLE core module speaks rather clear of how much this SoC was thought to (not) fit into a handheld.
Anandtech's Joshua Ho suggests the Denver cores could play as both big and LITTLE because they could scale sufficiently low in power consumption.. but I think that's a bit of a long shot.
Especially when nvidia themselves call Parker a:
It's "Parker - an Automotive SoC". Not "Parker as an Automotive SoC".
IMO, the fact that there's no "ninja core" or LITTLE core module speaks rather clear of how much this SoC was thought to (not) fit into a handheld.
Anandtech's Joshua Ho suggests the Denver cores could play as both big and LITTLE because they could scale sufficiently low in power consumption.. but I think that's a bit of a long shot.
Especially when nvidia themselves call Parker a:
It's "Parker - an Automotive SoC". Not "Parker as an Automotive SoC".
X1, K1 Denver have no low power core or BL either. X1 never uses both CPU banks at the same time.
D
Deleted member 13524
Guest
X1 has a Cortex A53 module (not used in Pixel C because afaik it's broken).
K1 Denver didn't have any low power cores, and that worked out so well (not) that nvidia went back to using standard ARM cores for X1.
K1 Denver didn't have any low power cores, and that worked out so well (not) that nvidia went back to using standard ARM cores for X1.
How do you know X1's A53s are broken? The X1 PR always seems to keep the banks separate. As if they are there only if you want to implement the chip in a special low power design. It is a bit curious though.
Last edited:
D
Deleted member 13524
Guest
I don't know for sure that they're broken, but there's got to be a good reason why they're never powering up in the Pixel C.
Unless there's been a firmware update that I don't know of and activated them.. do you know of any?
I've been occasionally looking at what the community is up to with Pixel C but haven't seen anything about the A53s. My impression is the A53s are just an entirely separate bank of CPUs and you can only run with A53s or A57s. Strange but yeah who knows what the plan was there.
But X1 seems well behaved in that tablet. Surprisingly. Even with no low power CPUs.
*Apparently Helio X30 is the first chip on TSMC 10nm. Quite an achievement for MTK given that they arent even shipping a 16nm chip yet. However it will be produced in relatively small quantities compared to A11. Samsung and Qualcomm will also be mass producing products on 10LPE next year.
*I am very surprised that Xavier is on 16nm. Given that it is sampling only in Q4'17, 10nm would have been mature enough. NV must have some reasons...
*Yep GV100 is basically a supercomputer only product initially
*Not as bad as 20SoC but not substantially better than 16FF either. Offers a density increase but 7nm is supposedly following within a year. Will dig up more on this
Ignoring that this is a bit off topic for a Volta GPU thread, nVIDIA has no place in tablets for quite some time. The only tablet products they had lately were low volume stunt deals with Google. The TK1 Denver powered Nexus 9 failed miserably and the Pixel C is unknown for most people. Even if Google launches the rumored Andromeda OS to try and get some of the traditional Desktop OS pie, nVIDIA's business model is not compatible with low margin products like tablets and mobile phones. They are not capable of sustaining the high revenue nVIDIA needs for dedicated R&D. Now, if Google manages to grow Andromeda into a Desktop OS that can compete with Windows (which I seriously doubt), nVIDIA could have some chance to have Tegra powering such machines as long as they are way more expensive than your average tablet.
iMacmatician
Regular
Do you know if the A10X will be on 10 nm as rumored?
No idea tbh but if the rumoured release date of March is true then I would be surprised if it is.
Frankly I don't expect anything desktop Volta before 2018. Tidbits and rumors left and right the internet point at a GDDR6-whatever necessity, which makes sense in that by N degree more powerful future Volta SKUs will also need by X degree more bandwidth, preferably with a solution that will keep power & cost within reasonable boundaries.
GDDR5x in theory could go up to 14Gbs over the current 10 Gbs in the 1080.
The question is how much do we expect performance to increase over Pascal? Given that we are staying at the same process, I'm not sure more than 12-14 Gbs GDDR5x would be needed for a GV102 (and below) product.
Personally I'm not going expecting a jump as big as we got with Pascal since we're still on the same node.
The question is how much do we expect performance to increase over Pascal? Given that we are staying at the same process, I'm not sure more than 12-14 Gbs GDDR5x would be needed for a GV102 (and below) product.
Personally I'm not going expecting a jump as big as we got with Pascal since we're still on the same node.
ImSpartacus
Regular
This Anandtech article mentions that Micron's "theoretical" limit to gddr5x is 16Gbps, but the standard uses 10-14Gbps.
JEDEC’s GDDR5X SGRAM announcement discusses data rates from 10 to 14 Gbps, but Micron believes that eventually they could be increased to 16 Gbps. It is hard to say whether commercial chips will actually hit such data rates, keeping in mind that there are new types of memory incoming. However, even a 256-bit GDDR5X memory sub-systems running at 14 Gbps could provide up to 448 GBps of memory bandwidth, just 12.5% lower compared to that of AMD’s Radeon R9 Fury X (which uses first-gen HBM).
http://www.anandtech.com/show/9883/gddr5x-standard-jedec-new-gpu-memory-14-gbps
That was confusing to me since Samsung's hot chips statements said that GDDR6 was notable because it could get to 14-16 Gbps. At first glance, you go, "that's not any higher than the long term GDDR5X goals," but maybe GDDR6 will have tweaks that make it easier to actually realize the higher end of that bandwidth spectrum? Or hell, maybe GDDR6 functionally IS long term GDDR5X...
Me neither and if the rumours that NV is skipping 10nm are true and Volta is still on 16nm..NV will have to do another Maxwell as they will be area and power limited. I'm not sure if GDDR6 is a necessity as such. They'll get a bit more out of delta compression (Diminishing returns there though) and GDDR5X will scale to at least 14 Gbps which is 40% higher than what they're shipping today. That should be enough to tide them over.
Last edited:
They are skipping 10nm, 7nm will be their next node, as for the release timeframe for Volta, it's certainly been accelerated, and since it will still be on 16nm I see no reason for them not to release it, there won't be prohibitive cost increase or anything. I imagine the chips at each tier will be larger, they each have 25% more area to go
