AMD to introduce its Radeon R9 300-series lineup at Computex

[wand3r3r]

What if the delay is to get 20 nm parts out? That'd be a huge feat!

I doubt it anyway, but AMD said they'll have 20nm parts out in H1 but that could be anything and not necessarily the high end GPUs.

Anyways, interesting speculation.

 

[Noctifer616]

What if the delay is to get 20 nm parts out? That'd be a huge feat!

I doubt it anyway, but AMD said they'll have 20nm parts out in H1 but that could be anything and not necessarily the high end GPUs.

Anyways, interesting speculation.

We are not going to see big dies on 20nm. 20nm has way too many problems to make big chips on it. Next step is either going to be 16nm TSMC or 14nm GC/Samsung which won't come out till next year for big dies (mobile is on a different schedule).

 

[Silverforce11]

I agree here that R9 390/390X won't be priced at 290/290X current levels but what about R9 370/370X/380/380X? If AMD wants to launch the entire series in 1-2 weeks, it's not time enough to clear R9 200 series and then you will definitely have pricing overlap with 200 series cards.

They don't need an entire new lineup, they need to plug the hole that the 970/980 is tearing into them. It's a gaping wound.

With R290/X vs 780/ti they had close performance and close power use.

With R290/X vs 970/980 its worse performance and much worse power use. They then have the balls to price R290X at ~= to 970? No wonder so few are buying it. Its turned into their CPU nightmare, Faildozer against Intel.

Thats why they need 390/X asap (or months ago) and not in June. The rest of the lineup can be staggered out when ready, while current stuff (Tahiti & Hawaii) gets a wholesale major price cut to clear it out.

 

[Ed1]

They don't need an entire new lineup, they need to plug the hole that the 970/980 is tearing into them. It's a gaping wound.

With R290/X vs 780/ti they had close performance and close power use.

With R290/X vs 970/980 its worse performance and much worse power use. They then have the balls to price R290X at ~= to 970? No wonder so few are buying it. Its turned into their CPU nightmare, Faildozer against Intel.

Thats why they need 390/X asap (or months ago) and not in June. The rest of the lineup can be staggered out when ready, while current stuff (Tahiti & Hawaii) gets a wholesale major price cut to clear it out.

The bigger issue I see if rumor lineup is right . If 390/x needs water cooling and rest below are re badges thats going to be bad for sales even if it performs better than GM200 (980ti, Titan X ) .

I just can't see them selling alot if its water cooled , only niche % will buy .
OEM will stay away IMO .

 

[Udgnim]

The bigger issue I see if rumor lineup is right . If 390/x needs water cooling and rest below are re badges thats going to be bad for sales even if it performs better than M200 .

I just can't see them selling alot if its water cooled , only niche % will buy .
OEM will stay away IMO .

my thinking is that AMD's reference design will be water cooled so reviewers won't be criticizing temps and that AIB vendors will provide their own air cooled design

 

[Silverforce11]

GPUs don't "need" water cooling. Plenty of air cooling designs already exist that can handle 450W+.

Definitely a good move to have a great reference design that will be cool, quiet & exhaust heat out a case. If people don't want that, they can grab custom air designs.

 

[Techhog]

They really should introduce two reference coolers. Actually make a semi-decent blower, and have a premium SKU with an AIO.

 

[ocre]

@ Silverforce11,

Right now AMD's market share looks very bad but this is not only because of R9 200 desktop cards. Literally 3 years of missing out ona competitve line-up of laptop GPUs has finally caught up to them. Someone who got a Kepler-based laptop in 2012 is hitting his upgrade cycle in 2015. This gamer's only reasonable choice? NV mobile cards, most likely 95%+ of the time today. Essentially AMD's laptop market share for discrete cards is approaching 0% the longer this continues. Since laptop cards are > 50% of the entire GPU market, getting nearly 0% market share in laptops means even if AMD's desktop market share were a hypothetical 50%, their overall market share would be below 25%. Since we know that AMD doesn't have 50% desktop market share, it was simply a matter of time and mathematics until their overall market share fell below 25%. In another 3 months, nearly 0% AMD-powered gaming laptops will have been sold and 960/970/980 will continue to take away desktop market share, which means AMD's overall market share should dip below 20%.

Why did I spend an entire paragraph on this? Because I keep disagreeing with you where the focus should be for AMD. It should not be performance crown but laptops. Even if R9 390X beats GM200, it will mean little if AMD doesn't have good mobile 300M parts. Alternatively, if 300M is excellent, AMD will gain WAY more market share than it ever could with performance crown 390X.

The thing is almost all major laptop design wins are locked in for 1H of 2015. The next time AMD has a shot at going into a large wave of laptops is around Skylake/Windows 10 launch. By that point, NV can easily drop prices on all GM204 parts, get 10-20% faster clocked GM204/206 parts too. AMD made a critical mistake on the desktop of releasing nothing at all for 1.5 years since 290/290X, nothing worthwhile to replace 7850/7870/7950/7970 either for nearly 3 years (!) and almost nothing new in laptops for 3 years since HD7970M and below. Even today a $600 295X2 beats 980 by miles at high end gaming but gamers don't care. Sure it's dual card vs. single card but 295X2 smokes a 980 at 1440p and 4K overall. Who is buying it?
Not many people.

Performance crown is not the primary reason ATI was so much more successful. It's because ATI had good products in many price segments that launched on time or close to NV's new cards. Think about it, what would happen to ATI's graphics card business if they used 1.5-3 year old technology against NV's latest cards? What would happen if GeForce 5, 6, 7 had 7-9 months all to itself while ATI dragged its feet?

If you are AMD and you are constantly late to market and MIA in laptops, the performance crown should be the last thing on your mind. Every day you don't have new products against NV is more market share and profit losses not only at the $500+ price level of 390X, but in all mobile and desktop price segments. That's why your idea that performance crown could somehow save AMD is wishful thinking. Even if 390X beats GM200 in all metrics, you can't bring back 9 months of being late to market where NV took market share.

I guarantee that Lisa Su will change the GPU strategy after 300 series. ATI would have never allowed 1.5 years of stagnant time between a 290X and a 390X. They would much rather launch a 25% faster card in 9 months and then again a 25% faster card in 8-9 months rather than going for 50% at once but have nothing for 1.5 years IF they were in AMD's situation. ATI's management would never coast for 1.5 years with 1.5-3 year old lineup of chips made up of 2012 Pitcairn, Tahiti and 2013 Hawaii. ATI's management knew that launch timing was critical.


In other words, AMD's entire GPU strategy is wrong because the competitor launches new GPU architectures every 2 years. You either have to adopt a similar approach, or launch newer cards way faster to keep up. Otherwise you end up with a January 31, 2012 (!) HD7950 rebadged as an R9 280 competing with a January 2015 (3 years newer!!!) 960. Do you realize how ridiculous that would sound to an ATI engineer?

As I said, it's not about the performance crown. 7970Ghz had the official performance crown from June 2012 until the Titan launched. Look at almost any major reviews and 7970Ghz beat 680 and 7970Ghz OC beat 680 OC for at least 6 months. How did that work out?

The entire strategy and execution is all wrong. Even if 390X is 15% faster than GM200, but everything else in 300 lineup is unimpressive, it won't gain AMD meaningful market share or profits. What needs to happen is a competitive line-up top-to-bottom and launched on time. The next time AMD will have a shot of actually doing that is R9 400 series.

With R9 300 series, AMD lost the momentum completely. NV could easily price 970 at $249 and 980 at $329-349 and then what? That's why being late to market is devastating for AMD as its customers don't sit 6-12 months waiting like NV's do more often.

the desktop loss has been massive this last Quarter. This played a huge role in the drop down to 24% market share.

http://jonpeddie.com/publications/add-in-board-report/

• AMD’s quarter-to-quarter total desktop AIB unit shipments decreased -16.0% .
• Nvidia’s quarter-to-quarter unit shipments increased 5.5% · Nvidia continues to hold a dominant market share position at 76.0% .

AMDs mobile losses had been going on for sometime but this past quarter they started bleeding desktop marketshare at an astonishing rate.

I can agree with you that the focus should be more on mobile though. This is an area of much more opportunity, i think.
Its just that this past quarter AMD has seen a massive drop in desktop discrete. This cannot be brushed off as insignificant. I personally agree with silverforce when he says that most of AMDs mobile discrete volume was hit hard long ago. That the units shipped were already pretty low.

The 16% loss in desktop discrete played a very large role in this final drop toe 24% marketshare. That is really one of the largest tragedies for Q4

 

[soccerballtux]

what a poorly managed company

 

[Silverforce11]

We've reached a milestone.

Matrox is now officially 0.00%!

 

[raghu78]

They don't need an entire new lineup, they need to plug the hole that the 970/980 is tearing into them. It's a gaping wound.

With R290/X vs 780/ti they had close performance and close power use.

With R290/X vs 970/980 its worse performance and much worse power use. They then have the balls to price R290X at ~= to 970? No wonder so few are buying it. Its turned into their CPU nightmare, Faildozer against Intel.

Thats why they need 390/X asap (or months ago) and not in June. The rest of the lineup can be staggered out when ready, while current stuff (Tahiti & Hawaii) gets a wholesale major price cut to clear it out.

I think you are exaggerating. The GTX 970 does not look great against R9 290X. I basically disagree that the GTX 970 is a R9 290X equivalent, especially in the most demanding games when the VRAM usage gets upto 4 GB. You don't buy a USD 350 graphics card to play at settings which are suitable to the graphics card and don't expose its anomalies :whiste:

The GTX 970 stands thoroughly exposed with its VRAM limitation at both 1080p and 1440p.

http://www.pcgameshardware.de/Total-War-Attila-PC-259548/Specials/Test-Benchmarks-1151602/

the frametimes are horrible. the min fps is much lower than R9 290/R9 290X and even lower than R9 280X and all of this is at 1080p.

http://www.techspot.com/review/962-evolve-benchmarks/page4.html

"Disappointingly, the GTX 970 averaged just 44fps, being only marginally faster than the old HD 7970 GHz at 41fps. We believe this massive reduction in performance is due to the GTX 970's partitioned memory configuration."

http://www.sweclockers.com/artikel/20031-snabbtest-grafikprestanda-i-evolve/3

https://translate.google.com/translate?sl=auto&tl=en&js=y&prev=_t&hl=en&ie=UTF-8&u=http%3A%2F%2Fwww.computerbase.de%2F2015-02%2Fevolve-benchmarks-14-grafikkarten-im-vergleich%2F%23diagramm-frametimes-2560-1600&edit-text=

btw thats a Gameworks title. so no excuses for getting hammered. The latest CRYNEGINE seems to love GCN (Ryse, Evolve) and so if thats a trend its not so great for Maxwell when next gen GCN 2.0 products launch.

In fact I believe the US tech press is blameworthy for pushing this GTX 970 memory partitioning issue under the carpet or rug.

In a couple of weeks all these spoon fed sites will be raving about the incredible perf of the Titan-X. It looks as if GM200 might not beat the fp64 perf of GK110 so I am going to look at the actual fp64 perf and review conclusions of Titan-X from the so called objective press. The last bit of justification for that ridiculous price seems to have gone out of the window. :biggrin:

 

[Silverforce11]

R290X is only worth it over a 970 at the same price if we're talking about CF vs SLI. That's where the performance gap grows in favor of the R290X as well as the extra vram. In single card setups, they are quite close. We also have to factor in 970s OC to 1.5ghz easily without being a major power hog.

These are obvious reasons (besides NV features/ecosystem for more loyal users) why the R290X needs to be cheaper than 970 else its not a good buy for most gamers.

 

[3DVagabond]

R290X is only worth it over a 970 at the same price if we're talking about CF vs SLI. That's where the performance gap grows in favor of the R290X as well as the extra vram. In single card setups, they are quite close. We also have to factor in 970s OC to 1.5ghz easily without being a major power hog.

These are obvious reasons (besides NV features/ecosystem for more loyal users) why the R290X needs to be cheaper than 970 else its not a good buy for most gamers.

Have you not noticed the trend in recent games where GCN is performing better relative to Maxwell and Kepler? I think when you consider both of their trade offs, weaknesses and strengths, the 290X is actually a superior card. Now, that doesn't mean that a majority of people will feel that way, I know. Just like most people for the same money would have taken the 680 over the 7970. I believe that over time Tahiti has proven to be the stronger design though. IMO so will Hawaii.

 

[Silverforce11]

I appreciate perf/$ and with my electricity rates at 31 cents per kWh and rising (damn solar panel schemes), excessive power use directly relate to final price over time.

75-100W extra total system power is not a little amount. This is why R290X has to be ~$50 cheaper than 970 for someone like me to consider it over a 970.

Power use didn't matter much when I bought R290s compared to 780/ti since it was close enough (the 780/ti was also much more expensive here). But compared to 970s, AMD's old-tech lineup needs major price cuts.

They also need new-tech to fight Maxwell. This is why any delays is going to be very costly for AMD.

 

[raghu78]

R290X is only worth it over a 970 at the same price if we're talking about CF vs SLI. That's where the performance gap grows in favor of the R290X as well as the extra vram. In single card setups, they are quite close. We also have to factor in 970s OC to 1.5ghz easily without being a major power hog.

These are obvious reasons (besides NV features/ecosystem for more loyal users) why the R290X needs to be cheaper than 970 else its not a good buy for most gamers.

Sorry but thats completely wrong. GTX 970 is showing problems in single GPU in games like Rome Total War Atilla and Evolve. R9 290X is hammering the GTX 970 in Gameworks titles like Evolve. The GTX 970 is not a R9 290X equal according to recent reviews especially at 1440p

http://www.sweclockers.com/recension/19925-geforce-gtx-960-fran-asus-gigabyte-och-msi/17#pagehead

http://ht4u.net/reviews/2015/nvidia_geforce_gtx_960_msi_gtx_960_gaming_2g_im_test/index40.php

http://gamegpu.ru/test-video-cards/igry-2014-goda-protiv-sovremennykh-videokart.html

Even a 1.5 Ghz GTX 970 OC cannot beat a R9 290X at 1100 Mhz. btw no amount of OC can help you when your card is having horrible frametimes and stuttering due to the memory partition :biggrin:

the GTX 970 in hindsight is a bad design. the memory partition creates more problems than the extra VRAM solves. Nvidia were better off having 3 SKUs -

GTX 980, 64 ROPs , 4 GB - USD 549
GTX 970, 64 ROPs, 4GB - USD 399
GTX 960 Ti, 48 ROPs, 3 GB - USD 249

They would have still sold very well. When the R9 390X, R9 390, R9 380X and R9 380 launch the GTX 970 will be EOL and replaced with a SKU like I mentioned. Nvidia product stack will undergo some price compression. the GM200 SKUs will fill out USD 500 - 650. the GM204 SKUs will fill USD 250 - USD 400.

I appreciate perf/$ and with my electricity rates at 31 cents per kWh and rising (damn solar panel schemes), excessive power use directly relate to final price over time.

75-100W extra total system power is not a little amount. This is why R290X has to be ~$50 cheaper than 970 for someone like me to consider it over a 970.

Power use didn't matter much when I bought R290s compared to 780/ti since it was close enough (the 780/ti was also much more expensive here). But compared to 970s, AMD's old-tech lineup needs major price cuts.

They also need new-tech to fight Maxwell. This is why any delays is going to be very costly for AMD.

At current prices the R9 290X is a better card than GTX 970. Many people who bought GTX 970 are paying the price with poor performance. So imo its one of the most shortsighted cards ever designed which has shown up its limitations very early in its lifecycle.

 

[3DVagabond]

I appreciate perf/$ and with my electricity rates at 31 cents per kWh and rising (damn solar panel schemes), excessive power use directly relate to final price over time.

75-100W extra total system power is not a little amount. This is why R290X has to be ~$50 cheaper than 970 for someone like me to consider it over a 970.

Power use didn't matter much when I bought R290s compared to 780/ti since it was close enough (the 780/ti was also much more expensive here). But compared to 970s, AMD's old-tech lineup needs major price cuts.

They also need new-tech to fight Maxwell. This is why any delays is going to be very costly for AMD.

Yeap. Heard all the same arguments when the 680 was released too. I understand that. The 970 is even worse though because it's a broken design.

 

[Silverforce11]

It was never that bad with the 7970 vs 680, power use wasn't a blow out (until the 7970Ghz). They are actually quite close.

Lets face it, with Maxwell 970/980 NV really did major damage to AMD. It wasn't anywhere as bad prior. The 7950/70 held its own just fine and in fact now its clear its a better design than GK104. Same for R290/X vs 780/ti.

Except NV has moved on.

 

[raghu78]

Yeap. Heard all the same arguments when the 680 was released too. I understand that. The 970 is even worse though because it's a broken design.

yeah. more importantly it did not take long for the design flaw or feature should we say to get exposed. :biggrin:

 

[raghu78]

It was never that bad with the 7970 vs 680, power use wasn't a blow out (until the 7970Ghz). They are actually quite close.

Lets face it, with Maxwell 970/980 NV really did major damage to AMD. It wasn't anywhere as bad prior. The 7950/70 held its own just fine and in fact now its clear its a better design than GK104. Same for R290/X vs 780/ti.

Except NV has moved on.

You are spot on. The damage to AMD now is even worse than the HD 2900XT days. But do not forget what happened with the HD 4870 and HD 5870 fightback and the last time when AMD got first to a memory tech. The last time AMD got first to a memory tech they hammered Nvidia, gained market share and finally became the largest GPU vendor. It took more than 2 years for Nvidia with GTX 460 and GTX 580/GTX 570/GTX 560 to return the market leader position.

http://www.dailytech.com/GPU+Market+Shows+Impressive+Gains/article13294.htm
http://www.xbitlabs.com/news/graphi..._on_Discrete_GPU_Market_Mercury_Research.html

Nvidia's first FINFET chips are Maxwell shrinks and not Pascal. Pascal with HBM is not due till Q4 2016 or Q1 2017. AMD will have HBM based flagship GPUs from June 2015.

The advantage with HBM based GPUs will be more felt in notebooks. Thats where AMD have lagged Nvidia badly over the last 3 years. By mid next year an AMD 14nm FINFET flagship GPU with 2nd gen HBM will be available with 8 / 16 GB capacities. Nvidia's road till Pascal launches is going to be difficult.

The bigger threat to Nvidia comes from Zen based APUs with competitive x86 CPU performance against Intel and HBM as system memory. a Zen APU with a single HBM stack (2nd gen) with 8GB capacity and 256 GB/s bandwidth is basically a gaming PC on an interposer. No need for RAM slots, no need for southbridge. The form factor and cost/power benefits will be compelling from a simplified motherboard. Thats when he real damage will start happening to Nvidia as the APU's graphics performance will be on par with AMD discrete GPUs with HBM of similar area (roughly 120-130 sq mm).

 

[StereoPixel]

By mid next year an AMD 14nm FINFET flagship GPU with 2nd gen HBM will be available with 8 / 16 GB capacities. Nvidia's road till Pascal launches is going to be difficult.

I think 28nm Fiji/Bermuda-based Radeon and FirePro will have a 8GB and ~12-16GB HBM.

 

[Enigmoid]

People need to realize that if the 780 Ti and 780 are doing a lot better than the 970 its not vram. A 3 GB card is going to perform far worse than a 3.5 GB card when Vram limited.

 

[Attic]

No doubt the 390/x will be very worthwhile parts. But it's too long since the 290x to wait to deliver a new high end single gpu part IMO given the reality of the environment.


The 290x debuted back in Oct 2013.

The 780ti came out afterwards to hit back in Nov 2013.

Then nVidia followed up with the 970/980 in Dec 2014.


Now it's March 2015 and we are still waiting for a successor to the 290x.

Yes the 970 is basically a dud due to nVidia going nVidia on it's consumers, but the deception game they play'd worked and they still hit a time schedule on new releases and updates that squarely punch AMD in the nose. The performance/power improvements alone though, from the 970/980 vs previous high end cards, still for the most part mark a new era and it's one AMD is too late to.


I guess that AMD miscalculated the impact of the 295x2 as a card to stave off nVidia competition.


With all that, I expect the 390 and 390x when available to immediately make the 970/980 look weak by comparison. AMD is going to deliver big and swing hard IMO.

 

[Techhog]

People need to realize that if the 780 Ti and 780 are doing a lot better than the 970 its not vram. A 3 GB card is going to perform far worse than a 3.5 GB card when Vram limited.

That depends. Some games are better optimized to not push as much into VRAM when there's less of it, so on the 780 it's trying to make the best of the 3GB, while on the 970 it's trying to use the full 4GB. If the texture streaming is somewhat affected by the available VRAM, and it's trying to stream into the slow portion because the game doesn't "know" that it's slower than simply streaming less into memory, this seems like what would happen. Besides, what else could be causing the gap in the first place?

 

[ocre]

People need to realize that if the 780 Ti and 780 are doing a lot better than the 970 its not vram. A 3 GB card is going to perform far worse than a 3.5 GB card when Vram limited.

i dont see how anyone can disagree with this.

 

[StinkyPinky]

If it wasn't for this memory issue, I would have bought the 970 to replace my 280X. Now I'm not sure what to do. 290x?