AMD quizzed over sharp drop-off in R&D expenditure

[HEXUS]

AMD shares plummeted by as much as 13 per cent yesterday.

Read more.

[mudpie]

I think it would have been better if his name was cpumaster/gpumaster.
Fury line is decent. Not extraordinary. Just decent. Why would I buy decent?

[shaithis]

About time the "professionals" asked the questions.

[semitope]

I think it would have been better if his name was cpumaster/gpumaster.
Fury line is decent. Not extraordinary. Just decent. Why would I buy decent?

There's something else to buy at $550? As far as fury X... someone else clearly is buying it.

[b1candy]

Wait, so in the last couple of weeks the shares rise 25% and now they fall 13% and it's news?
Also, AMD confirmed those Zen slides were fakes months ago at the Financial Analyst Day.

I'd post a link but I don't meet the required post limit to do so.
Do a google search of 'Zen slides faked' and you get the links to SemiAccurate and Planet3DNow.

[outwar6010]

Wait, so in the last couple of weeks the shares rise 25% and now they fall 13% and it's news?
Also, AMD confirmed those Zen slides were fakes months ago at the Financial Analyst Day.

I'd post a link but I don't meet the required post limit to do so.
Do a google search of 'Zen slides faked' and you get the links to SemiAccurate and Planet3DNow.

I remembered the sharp increase and thought this new was odd.

[Jimbo75]

So R&D is down 40% since the beginning of the decade. It's not really hard to figure out what's gone on and Papermaster just confirmed it. Su also mentioned it recently.

At 14nm stuff like the old Brazos/Kabini line is going out. Su said they're dropping out of the low-margin stuff. The low-end PC market is about to be completely eaten by Chromebooks and the like, anybody who want's to fight over that stuff can go ahead, but AMD won't be.

They've ditched much of the server stuff as well, with 1 ARM core and the next based on Zen. All the other crap, Seamicro etc has gone.

They're out of phones completely.

They no longer pay fab R&D for SOI at Globalfoundries. They no longer pay leading-edge R&D at TSMC.

R&D expenses are shared with the customer as part of the semi-custom setup.

Here's the big one though - they'll no longer be pushing the industry forward in stuff like memory technology etc. GDDR3, GDDR5, HBM etc, you can thank ATI/AMD for much of that but you won't be thanking them much more in future.

Incidentally - It wasn't so long ago that people were slating AMD for having no presence in phones and ultramobile. Well Intel continues to lose multiple-billions in that every year while Nvidia just took a $100m+ writedown on Icera. Both companies have failed utterly in ultramobile and it's cost them billions. No need for AMD to waste R&D there, that space is owned by bigger and better ARM companies.

There's your 40% reduction in R&D since 2010. Everything they have has gone into 14nm for years. Bulldozer has had buttons spent on it and even 28nm GCN was just iterated on. Christ, Fury is just GCN 1.2, that's why it has no HDMI 2.0. It's basically year-old Tonga technology.

[tygrus]

WARNING for AMD: Build it and they will come. If you stop building it, they stop coming. (re-write of classic quote).

If AMD had 14nm parts today and OEM/ODM were including them in SFF, thin&light devices (notebooks, tablets etc.) then AMD could be selling at least 20% more APU's and the share price could be triple. The longer they drag it out the more they become irrelevant and bankrupt. Sorry, but you can't keep saving money and expect the sales to increase.

[DanceswithUnix]

WARNING for AMD: Build it and they will come. If you stop building it, they stop coming. (re-write of classic quote).

If AMD had 14nm parts today and OEM/ODM were including them in SFF, thin&light devices (notebooks, tablets etc.) then AMD could be selling at least 20% more APU's and the share price could be triple. The longer they drag it out the more they become irrelevant and bankrupt. Sorry, but you can't keep saving money and expect the sales to increase.

AMD have had competitive tablet chips for years, they built them and no-one came. It really isn't that simple.

[CAT-THE-FIFTH]

AMD have had competitive tablet chips for years, they built them and no-one came. It really isn't that simple.

Well,Intel spending billions on contra-revenue probably didn't help and AMD does not have that sort of money anyway. Before that Zacate and Brazos did very well against Atom getting 10s of millions of sales.

[PuremasterZ]

AMD is doing worse because they lack cpus and their gpus are expensive but without AMD the gpu market will fall

[Jimbo75]

WARNING for AMD: Build it and they will come. If you stop building it, they stop coming. (re-write of classic quote).

If AMD had 14nm parts today and OEM/ODM were including them in SFF, thin&light devices (notebooks, tablets etc.) then AMD could be selling at least 20% more APU's and the share price could be triple. The longer they drag it out the more they become irrelevant and bankrupt. Sorry, but you can't keep saving money and expect the sales to increase.

This is true. The reason they're *really* struggling is because guys like me didn't upgrade. I used to buy a new GPU mostly every year but I'd skip the stuff that was pretty bad. So long as they're somewhat close I have no issues swapping back.

For example I switched from a Q6600 to a 940 BE. Bit of a sidegrade tbh but I was happy to switch back. Of course I switched again to SB as that was a no-brainer. Graphics wise I've gone for a couple of X1950 Pro's then a switch during the 2000 and 3000 series where Nvidia had clearly the better stuff with the 8000 series. After that though I had a couple of 4770's a couple of 4870's, 5850 and a couple of 5770's. Bought a 6850, now using a 7850 with a hefty OC on it.

Every year yeah, but nothing since the 7850 because they didn't give me a good reason to upgrade. I actually built a new PC for my girlfriend last year with a 280X though. It's almost inconceivable that I won't be upgrading with 14nm but they need to get there first. They're gonna lose half their cash before then, pretty much assured about that.

AMD's main issue is that their buyers are more informed. A lot of Nvidia buyers just buy based on the number, completely oblivious to actual performance. I've seen guys with 980's downgrade to Titan Blacks, guys "upgrading" from SLI to triple-SLI Titan X's because they get lag and stutter in VR (more cards = even more latency using AFR)...sheer madness but they just don't understand the underlying technology.

Then you get guys who "punish" Nvidia for the 970 fiasco by buying 980's....then there's the guys who buy "gaming pc's" with GT 610's in them, guess what they do when the performance is crap? Nvidia getting paid twice for one system over and over. Maybe even three times in certain cases.

[shaithis]

AMD's main issue is that their buyers are more informed. A lot of Nvidia buyers just buy based on the number, completely oblivious to actual performance. I've seen guys with 980's downgrade to Titan Blacks, guys "upgrading" from SLI to triple-SLI Titan X's because they get stutter in VR...sheer madness but they just don't understand the underlying technology.

More customers = more occurrences. Calling AMD purchasers more informed based on a small subset of data you have seen is a stretch at best.

I think you'll find AMDs problems are more likely the ones some of us have pointed out for quite some time: Drivers, Marketing and TWIMTBP.

If AMD wanted to compete, they should have kept Mantle alive and fleshed it out with a GamesWork style set of libraries. Without that (or a huge shift in the way games are developed) people will buy the hardware that their games are telling them they run best with. This is further accentuated by forums like these mentioning TWIMTBP titles and the performance impact of it on certain hardware.

[Jimbo75]

More customers = more occurrences. Calling AMD purchasers more informed based on a small subset of data you have seen is a stretch at best.

I think you'll find AMDs problems are more likely the ones some of us have pointed out for quite some time: Drivers, Marketing and TWIMTBP.

If AMD wanted to compete, they should have kept Mantle alive and fleshed it out with a GamesWork style set of libraries. Without that (or a huge shift in the way games are developed) people will buy the hardware that their games are telling them they run best with. This is further accentuated by forums like these mentioning TWIMTBP titles and the performance impact of it on certain hardware.

Yes but the lie of Nvidia's supposedly better drivers (black ops), marketing and TWIMTBP (combination of black-ops and marketing) basically comes under the same blanket as being "uninformed".

Mantle is alive and well in LiquidVR, that's why AMD has a huge VR lead. Do you see any of the usual tech press talking about it though?

Really simple tbh - if you're buying on what you're being told in the usual tech press you're not informed, you're simply being marketed into your choice. Places like Hexus long since fell to Nvidia's marketing bucks. Look at the articles here - all pro Nvidia and all negative AMD. It's like an Nvidia marketing cannon and it couldn't be more obvious. Just look at the current front page, how many Nvidia articles?

Look at this one, "Can AMD continue to devote fewer resources to R&D but succeed against the likes of Intel and Nvidia?" The article made it crystal clear than GPU was not at all affected and that it's more been about focusing rather than cutting back? Why the constant negativity and loaded questions?

[edzieba]

Sticking with GCN in the long term is Bad News for long-term VR.
GCN, like every GPU architecture so far from both AMD and Nvidia, is throughput-optimised. But good VR performance is not hinged on throughput, it's hinged on latency. To optimise for latency is going to require changes in how GPUs are arranged (and how jobs are assigned) at the functional level. This isn't something you can morph an existing architecture into.

[Jimbo75]

Sticking with GCN in the long term is Bad News for long-term VR.
GCN, like every GPU architecture so far from both AMD and Nvidia, is throughput-optimised. But good VR performance is not hinged on throughput, it's hinged on latency. To optimise for latency is going to require changes in how GPUs are arranged (and how jobs are assigned) at the functional level. This isn't something you can morph an existing architecture into.

It's fine, because LiquidVR and GCN is already capable of <10ms latency, which is basically as close to perfect as you can get. It actually makes total sense given the market share GCN products have.

Nvidia struggles to remain under 20ms consistently because they have neither the software (Mantle) or the hardware (ACE's). Pascal should solve the hardware issue but the software is just as important.



The combination of LiquidVR and GCN 1.2 for fine-grained preemption (285, 380, Fury X, Fury) is perfect VR already. Don't expect to read anything about that in the tech press, they're all waiting on Nvidia getting their act sorted out first.

If you want to get deeper into the tech I'd recommend this - http://amd-dev.wpengine.netdna-cdn.c...lications.ppsx

There's a very good reason why all the major VR release titles are being done on LiquidVR right now.