GPU vendors concerned with declining AMD demand

[robb213]

I really don't see a problem. As said, what helps Nvidia is their new product release. At the same time, it may negatively affect them since many are unhappy over the 970. When AMD comes out with their's, things will change around again, especially if they remain with HBM over GDDR5. To me, it's just the usual cat and mouse game.

I wish Nvidia would continue to work on things that are actually nice like 3D Vision (which is neglected now), rather than things like GameWorks which imo look equal to anything already used in games.

[shaithis]

People were quite happy buying Nvidia cards when they were less efficient and its funny how most systems are lucky to see 400w even at the wall.

There was a big difference then.....in fact a huge difference.

Basically you had to choose between top-end performance at reduced efficiency or noticeably reduced performance at a better efficiency. That was also at a time when 1600p and multi-monitor setups were starting to penetrate the gaming market.

Now, you are looking at AMD cards with the same or less performance and reduced performance, the trade-off now is merely the cost of the card and any value added extras, like the never settle bundles.

It's quite a different situation.

We saw people switching when there was good reason to: GeForce 5 and GeForce 400....and then even the terribly designed GTX480 "only" drew 250w max.

I feel that AMD have put too much resource into the APU. Yes, it's a good idea, yes, it brought money in (and I do love my little AM1 system)...but has it been worth it?
Yes, it gave them the console contracts but I do feel (at least from looking at the recent profits) that they are not getting back what they put in.....now trade that against their wallowing in the desktop and server CPU markets and falling further behind nVidia and we can only guess at how much the never settle bundles are eating into their profits.......I do wonder if they made the right decisions.

[watercooled]

The opposite has been true just as many times if not more, especially if you don't just compare flagships where Nvidia have typically made huge-die halo products which AMD didn't release a competitor to.

One example, the 480 was barely faster than the 5870 but used massively more power, 122W vs 233W, that's far more in percentage terms than even Maxwell vs Hawaii.

And what about the 5970? In another class of performance vs 480 and guess what? Still used far less power at 178W. Same for 5970 vs the 580.

6970, similar performance to 570 yet less power.

4870, less power than 280 yet similar or faster performance.

The purchase price and bundles can be significant trade-offs, especially if they nullify the potential electricity cost saving as they often do for regular PC users/gamers.

[DanceswithUnix]

The interesting bit for me is the die size of the 960.

GTX 960 = 227 mm^2 (from http://en.wikipedia.org/wiki/GeForce...89xx.29_series)
R9 285 = 359 mm^2 (from http://en.wikipedia.org/wiki/AMD_Rad...#Chipset_table)

That is a massive difference for parts that game at the same sort of frame rates. The raw GFLOP rates for the 285 are much higher, where has all that computing power gone, are the drivers throwing it away?

[watercooled]

The 285 is a part-disabled Tonga die, we haven't seen a fully-enabled part on desktop yet. Maybe those dies are still being eaten up by the Retina iMacs.

[shaithis]

The 285 is a part-disabled Tonga die, we haven't seen a fully-enabled part on desktop yet. Maybe those dies are still being eaten up by the Retina iMacs.

My understanding is that the Mac parts aren't really any different to their PC counterparts....and are still labelled (and marketed) with the same part numbers.

If the 285 is a part-disabled die you have to ask why.....my guess, the same reason the GTX480 was. Why would you be holding back now when you got another generation to release soon?

[watercooled]

The 285 is part disabled (1792 shaders), it's been known since release. The M295X is also Tonga and is fully-enabled (2048 shaders), albeit clocked lower for the mobile market.

The reason for no desktop release is unknown, as I said one would be if Apple are eating up large amounts of fully-enabled dies for the M295X (which probably carries a higher margin than a theoretical 285X).

[DanceswithUnix]

So fully enabled the 285 might go from about as quick to about 15% faster? It does that with a die that is 60% bigger, that hurts.

Anyone seen any opencl benchmarks? Have only seen them under Linux, where the broken AMD drivers make the results meaningless.

[watercooled]

The bus on Tonga is twice as wide as GM106 which adds to die size, while the 960 goes for more expensive, higher-clocked memory. I've no idea how it evens out, but there's a trade-off in there.

The 285 is also clocked a fair bit lower than Tahiti; I'd be very surprised if a '285X' wouldn't match Tahiti clocks and outperform the 280X.

[Jimbo75]

There was a big difference then.....in fact a huge difference.

Basically you had to choose between top-end performance at reduced efficiency or noticeably reduced performance at a better efficiency. That was also at a time when 1600p and multi-monitor setups were starting to penetrate the gaming market.

Now, you are looking at AMD cards with the same or less performance and reduced performance, the trade-off now is merely the cost of the card and any value added extras, like the never settle bundles.

It's quite a different situation.

We saw people switching when there was good reason to: GeForce 5 and GeForce 400....and then even the terribly designed GTX480 "only" drew 250w max.

The 480 drew way over 250W, it's still the most power-hungry single GPU card ever built.

The press fawning over Maxwell is ridiculous. I don't remember anyone going nuts over a card like say, the 5770? It was as fast as the 4870 at half the price, half the die size and power. It absolutely mullered the 550 Ti in perf/Watt.

Now check out that piece of garbage that is the GTX 960. That card should be getting slaughtered as an obvious cash-grab but instead nobody mentioned the overclocked review cards. Why is Nvidia getting away with this with every release - because they've done it so often it's just accepted now? How can a card that is basically the same performance overclocked as the 760 get away with the 960 name? AMD get slaughtered for changing their entire naming scheme and rebranding old as new - Nvidia gets away with simply adding 2 digits to the beginning of their same-performance card and nobody in the press bats an eyelid.

Do you remember how the press and Nvidia fanboys lost their minds over the 560 Ti? A card that was the same performance, same power draw, same die size of the 5870 16 months later? Guess what card sold the most out of those two? AMD started competing against it with the superior 6950 - at lower prices more often than not - and the 560 Ti *still* sold more.

What about the 6850 vs the GTX 460? Around the same performance level, one of these cards was small die and lower power - the other outsold it by around 4-1. Any guesses which was which?

Then Nvidia starts releasing £1K cards like Titan - which has been beaten how many times by new "28nm" Nvidia cards since release? How many "fastest cards" has Nvidia had on 28nm anyway? 7?

GTX 680
GTX 690
Titan
GTX 780 Ti
Titan Black
Titan Z
GTX 980

7 "top performance" cards on the 28nm generation? And the fanboys and press continue to lap every one of them up!

How much faster is the 980 compared to the 680? According to techpowerup, 56% faster.

http://tpucdn.com/reviews/NVIDIA/GeF...rfrel_1920.gif

I know people that bought the 680, bought Titan or a 780 and also bought a 980! Are these people just dumb or what? Laying out £500+ or so every year for 30% faster cards? Then again you can't expect people to know what Nvidia is up to - however the tech press does know and should be pointing these things out.

The press let Nvidia off the hook during the Fermi days (apart from the 480 which rightly got slaughtered) and they've given absolutely nothing to AMD. All we hear about is how inefficient Hawaii is, how Never Settle is of dubious value and how all that matters is perf/Watt now that Maxwell is out.

[CAT-THE-FIFTH]

There was a big difference then.....in fact a huge difference.

Basically you had to choose between top-end performance at reduced efficiency or noticeably reduced performance at a better efficiency. That was also at a time when 1600p and multi-monitor setups were starting to penetrate the gaming market.

Now, you are looking at AMD cards with the same or less performance and reduced performance, the trade-off now is merely the cost of the card and any value added extras, like the never settle bundles.

It's quite a different situation.

We saw people switching when there was good reason to: GeForce 5 and GeForce 400....and then even the terribly designed GTX480 "only" drew 250w max.

I feel that AMD have put too much resource into the APU. Yes, it's a good idea, yes, it brought money in (and I do love my little AM1 system)...but has it been worth it?
Yes, it gave them the console contracts but I do feel (at least from looking at the recent profits) that they are not getting back what they put in.....now trade that against their wallowing in the desktop and server CPU markets and falling further behind nVidia and we can only guess at how much the never settle bundles are eating into their profits.......I do wonder if they made the right decisions.

The GTX460 consumed far more power than an HD6850 and yet was no faster. An HD6850 over clocked to nearly 1GHZ consumed slightly less power than a GTX460.

[shaithis]

The 480 drew way over 250W, it's still the most power-hungry single GPU card ever built.

It didn't and the 290x draws noticeably more....I am all for bashing the 400 series, it was a design that was built assuming there wouldn't be any manufacturing issues, which was a major oversight and cost them dearly.....but the 480 had a TDP of 250 and total power draw of ~450 in a high end system. The 290x peaks at what?

The press fawning over Maxwell is ridiculous. I don't remember anyone going nuts over a card like say, the 5770? It was as fast as the 4870 at half the price, half the die size and power. It absolutely mullered the 550 Ti in perf/Watt.

It was hard to "fawn" over something that wasn't really high-end. It was a great solid, filler card and they sold a lot of them.

Now check out that piece of garbage that is the GTX 960. That card should be getting slaughtered as an obvious cash-grab but instead nobody mentioned the overclocked review cards.

It was the same with the GTX680, a mid-range card masquerading as a high-end. Now ask yourself, how they got away with it....

Why is Nvidia getting away with this with every release - because they've done it so often it's just accepted now?

Ah, so you haven't worked it out.....it's 2 fold....firstly, the competition isn't up to snuff, secondly they have created brand power (just like how intel managed to keep selling leaf blowers - most people do not read forums and tech sites!), so people who aren't in the know will buy what they "think" is the best. While this is frustrating for techs, the great unwashed don't know better.

How can a card that is basically the same performance overclocked as the 760 get away with the 960 name? AMD get slaughtered for changing their entire naming scheme and rebranding old as new - Nvidia gets away with simply adding 2 digits to the beginning of their same-performance card and nobody in the press bats an eyelid.

Both have been hammered in the press for it....but it keeps happening. Seems par for the course now.

Do you remember how the press and Nvidia fanboys lost their minds over the 560 Ti?

Nope, not really, although at the time I was running CFX 5870s......which after a year I sold because I got sick of the bug that went on for 3 years where your mouse cursor corrupted while playing Blizzard titles....but I digress.....why did the 560Ti sell more? Probably as I mentioned before...brand power. You also have the "extras" to take into account. Stereoscopic film playback built into the drivers for free, Physx (if you play any of the titles that use it) and CUDA was starting to raise it's head around then as well.

What about the 6850 vs the GTX 460? Around the same performance level, one of these cards was small die and lower power - the other outsold it by around 4-1. Any guesses which was which?

Ditto above. Starting to get the picture yet?

Then Nvidia starts releasing £1K cards like Titan - which has been beaten how many times by new "28nm" Nvidia cards since release? How many "fastest cards" has Nvidia had on 28nm anyway? 7?

And we all laughed when they released it and we saw the price.....but they still sold them.....Brand power maybe?

I know people that bought the 680, bought Titan or a 780 and also bought a 980! Are these people just dumb or what? Laying out £500+ or so every year for 30% faster cards? Then again you can't expect people to know what Nvidia is up to - however the tech press does know and should be pointing these things out.

You tell me if they are dumb, you know them! If people want to lay out £500 for a 30% improvement who are you or me to tell them otherwise? Do the tech press know exactly what nVidia are up to? How long did it take them to twig about the 970?

The press let Nvidia off the hook during the Fermi days (apart from the 480 which rightly got slaughtered) and they've given absolutely nothing to AMD. All we hear about is how inefficient Hawaii is, how Never Settle is of dubious value and how all that matters is perf/Watt now that Maxwell is out.

But the 400 was the only real thing about the Fermi days to give them a slaughtering about.....I spent over a year with a GTX580 in my main rig and a 7970 in the front room rig. After 6 months I would have happily swapped the 7970 for a nVidia card. I was getting most frustrated with no physx in Batman, having to use crappy 3D playback software that was a premium purchase, incorrect sound channels being sent over HDMI and that was all after the 2 days I spent flashing BIOSes to get the card to give a display in the first place!
Now, I persevered with those issues and overcame them, would Joe Public have? Would that effect his next purchase do you think?

[kalniel]

It didn't and the 290x draws noticeably more....I am all for bashing the 400 series, it was a design that was built assuming there wouldn't be any manufacturing issues, which was a major oversight and cost them dearly.....but the 480 had a TDP of 250 and total power draw of ~450 in a high end system. The 290x peaks at what?

About 375W total system for a high end rig, at least 50W less than the gtx480.

[shaithis]

About 375W total system for a high end rig, at least 50W less than the gtx480.

Yup with a more modern and efficient CPU, mobo, RAM and PSU.

290x TDP = 290w

[kalniel]

Yup with a more modern and efficient CPU, mobo, RAM and PSU.

Ok in the tests (anandtech) I was looking at - the system accounts for ~45W difference with the same GPU. The gtx 480 still comes out drawing the same or more power when this is accounted for.

290x TDP = 290w

You know what TDP is right? And that you can't compare nVidia's TDP values to AMD or Intels?

[CAT-THE-FIFTH]

I had a look at some TPU charts. According to them the HD6970 was generally faster overall than the GTX570 and that was with 2GB of power hungry GDDR5 instead of 1.28GB with the GTX570. The same goes with the TPU charts with the GTX470 and the HD5870. Plus TPU tends to be more favourable to Nvidia anyway. So for 99% of the market Nvidia had worse power consumption,worse/no better performance and cards which produced more heat too.

Lets look at two popular games of the time,ie,Skyrim and BF3:

http://techreport.com/review/22048/t...us-in-skyrim/6
http://techreport.com/review/21982/t...-battlefield-3

They are both much of a sameness even with frametimes.

Yet where was all the whinging then? Nobody appeared to have cared so much that the even cards under the GTX480 and GTX580 were all consuming more power and that the GTX480 and GTX580 cost much more than any single GPU AMD/ATI card at the time. OFC when AMD had the fastest cards with the HD7970 and HD7970GE and the R9 295X2,its not important anymore since they CONSUME MOAR POWER and are TOO EXPENSIVE. Yet when Nvidia had the GTX480 or GTX580 it didn't matter they consumed MOAR POWER but THEY WERE THE FASTEST AT ALL COSTS! N-V-I-D-I-A RULZ!!

The same with features - C-U-D-A! P-H-Y-S-X! All I-M-P-O-R-T-A-NT! Yet,any AMD festures are not important like their generally better OpenCL performance along the whole range which is actually useful too.

Sounds pretty much like Apple vs Google/Android. You hear all the same arguments.

The worst thing it was nowhere as bad like a decade ago. Things have got worse since the Nvidia Focus Group and their successes and now AMD getting some of the same people to do their PR.

The funny thing is that some of my mates who were Nvidia die-hards actually tried some of the ATI/AMD cards and now have realised all this perceived superiority of each brand(at least under Windows) is a load of bull**** and just get the best deal they can,or if they need something specific to a brand.

Had dozens of each companies cards in the last 12 years.

I will still chose by price/performance.

I must have been lucky as I have only really been affected by minor niggles for the most part.

Edit!!

Anyway,thats me done here.