facebook rss twitter

AMD Navi GPU reference found in latest Linux driver

by Mark Tyson on 19 December 2017, 10:01

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qadoy6

Add to My Vault: x

AMD’s Navi GPU sis already being referenced in AMD’s Linux drivers. As VideoCardz reports, AMD Vega was codenamed GFX9 and the current Linux driver set for Radeon cards includes a string mentioning a ‘gfx10’ chip of a ‘super secret’ nature, as reproduced below.

  • new_chip.gfx10.mmSUPER_SECRET.enable [0: 0]

The architecture name would usually be written in the place of ‘super secret’ above. While the string has just become a hot topic it has been present in the AMD Linux drivers as far back as July this year. Thus, this isn’t an indication of Navi being on the precipice of launch.

Looking over previous AMD graphics architecture roadmaps, Navi will follow Vega when AMD’s foundry partner gets its 7nm process fit for mass production. Likely partner GlobalFoundries won’t be ready to make 7nm chips until late 2018 or 2019, it is estimated. Earlier this month GloFo shared some specifics about its planned 7nm node.

Back in 2016, AMD published the GPU architecture roadmap above, at its Capsaicin event. It indicates that Navi will deliver ‘scalability’ (multi-chip design?) and ‘nexgen memory’ (GDDR6 or HBM3?). On the chart Navi looks like it offers double the perf/watt of Polaris. The same chart shows that Navi is due in 2018 but timescales might have slipped by now, if not due to AMD’s side, then due to the partner foundry.

HEXUS Forums :: 6 Comments

Login with Forum Account

Don't have an account? Register today!
Posted by darcotech - Tue 19 Dec 2017 11:48
If you look Radeon RX 570 Cf results ( in average, slightly better FPS numbers than GTX 1070) and the same for RX 580 CF/ GTX 1080, one has to wonder what would have happened if AMD simply doubled Polaris CUs and made a chip with that design?

Once on the chip, it should have better numbers than CF can do, so it would be better than 1070/1080 for sure, even tickling 1080ti. The only downside would be TDP, but then Vega is not so much better in that regard.

What do you guys think?
Posted by chinf - Tue 19 Dec 2017 12:01
Shame there's no fact checking going on here - the article cited is either someone being sloppy or disingenuous in search of clicks and ad revenue.

Here's Phoronix's explanation of the patch which does not even relate to the Linux drivers, and is just a sample text.
Posted by Biscuit - Tue 19 Dec 2017 13:27
darcotech
If you look Radeon RX 570 Cf results ( in average, slightly better FPS numbers than GTX 1070) and the same for RX 580 CF/ GTX 1080, one has to wonder what would have happened if AMD simply doubled Polaris CUs and made a chip with that design?

Once on the chip, it should have better numbers than CF can do, so it would be better than 1070/1080 for sure, even tickling 1080ti. The only downside would be TDP, but then Vega is not so much better in that regard.

What do you guys think?

Production of a chip that size means that there is less dies per square mm on the wafer, plus it can often result in bad production yields. This results in the chip being a very risky proposition for AMD; if things dont go really well it becomes difficult to actually make any money from.

Have a look at how things went for Nvidia when they pushed the boat with Fermi….
Posted by Xlucine - Tue 19 Dec 2017 20:14
If you can cheaply stitch together GPU dies like epyc you've essentially solved GPU's for a generation or two, since you can go slower&wider until you massively exceed the reticule size (then laugh at the competition trying to make a monolithic chip to compete). Multi-chip cards does fit with the “scalability” term that's been staring us in the face for months, but I'll believe it when I see it

darcotech
If you look Radeon RX 570 Cf results ( in average, slightly better FPS numbers than GTX 1070) and the same for RX 580 CF/ GTX 1080, one has to wonder what would have happened if AMD simply doubled Polaris CUs and made a chip with that design?

Once on the chip, it should have better numbers than CF can do, so it would be better than 1070/1080 for sure, even tickling 1080ti. The only downside would be TDP, but then Vega is not so much better in that regard.

What do you guys think?

They essentially did for V64, only there were a couple of improvements available for the core design (polaris -> vega) and they tried to avoid the cost/power of the required 512-bit memory bus with HBM2. It is better than the 1080, and takes twice as much power than the 570
Posted by Dragonstongue - Wed 20 Dec 2017 07:03
would not be the first time AMD did exactly that with many gpu from prior generations while keeping power and temperature mostly in check such as 4870x2, 5970, 6990, 7990, Radeon Pro Duo.

I would imagine the one solid thing (if they could leverage it) the HBM memory, if they took a page from Nvidia book from a GPU many years ago, their GX2 cards, AMD could conceivably use a full length card for the “bottom” and use a shorter length for the “top” of the stack/sandwich so both gpu the memory etc goes to where it is needed and with modern design especially with HBM am sure the cooling solution could be very well done am very sure if they took the time to “tune” the voltages and maybe gave it more clock steps (so it can scale much more easily from the lowest clocks to the highest clocks for performance/power etc) it would be a killer card.

HBM at least offers what previous memory did not, that is it is situated in a much smaller area instead of spread all over the place, seems like an X2 style card makes absolutely a great deal of sense (not counting cost for them to build or the buyer) I think for sure cooling has to be optimal with the cooling fins/rads not impeding airflow, and the voltage very much have to be “tuned” instead of seemingly making the end user worry about doing that (RX/Vega all benefited a great deal by reducing the base voltage it seemed to maintain higher clocks while keeping power usage down go figure)

Anyways, will be interesting the new year and into 2019 am sure. and as far as “super secret leak” as many other sites namely the scourge of BS leak site WFCCTECH, unless it comes directly from AMD mouth, there is no point hogging up internet with FUD..not hard to “claim” that something was in Linux drivers when pretty much anyone can photoshop things especially as Linux is very open source with many different versions of it, so IMO, becomes a very unsubstantiated claim, as coming from Linux themselves, they only know of Navi by name, though it really has not had even preliminary driver support from them or AMD as of yet.

I personally want to see the 2018 refresh for Ryzen (12nm) the “missing” Vega 11 cards (rumored Vega 32/28) as for the Ryzen refresh, again, too many places like the WFCC putzes, a refresh is NOT a new generation or what they are saying as Ryzen 2, Ryzen+ etc, AMD has said all along R2/Pinnacle Ridge is and always was to be 7nm, the “refresh” seemed like a last minute option cause it seems like 12nm was not part of anyone's plans, I suppose TSMC/GF/Samsung all seemed to “figure it out” at the last minute, so they now offer that option..might be no real cost increase (possible slight reduction vs 16nm or 14nm) increased performance/lower voltage.

Anyways….am done…Merry Christmas to you all, and a Happy New Year when that time comes, stay safe ^.^

SEE NEWER »