[Rumor (Various)] AMD R7/9 3xx / Fiji / Fury

[.vodka]

A marketing hypefest is already better than the rumors we've been speculating on for the past months or even year. At least there's useful information in these slides on launch day.

 

[ocre]

Strange, you had very little "hope" to begin with.





Fortunately things will unravel soon and rumors will make room for numbers. Then we won't have to worry about strayed hopes and furious punches... or will we?

You seem to have me mixed up with someone else.

I have always had high hopes got fury. But, dont take my word on it. There is always my post history, you will have a hard time finding me say anything negative about it.

I have only very recently conceded to the idea that fiji will be at or slightly over titan x performance. But i am not totally convinced.

I always say that having great expectations and hyping something way high before launch is not a good idea but i never once dogged on fiji.

I have always had high hopes for fiji and the 300 series. The better AMD does in this generation, the better for everyone

 

[RussianSensation]

This is dragging on way too long. AMD really needs to crap or get off the pot. A June release is much later than expected anyway, and almost everyone was expecting a Computex reveal. Then AMD specifically said the reveal would be at E3. If they now say "just kidding, we're not going to show you anything but a few worthless marketing slides" then a lot of people are going to decide it just isn't worth the wait. The GTX 980 Ti is already out there and selling like hotcakes. AMD doesn't have the luxury of time. The longer they stretch this out, the less confidence I have in the performance of their product. These repeated delays smell like they're trying to patch things up at the last moment. According to Hardware Canucks, "Drivers are still in their pre-beta form and clock speeds are undergoing final tuning." At this late date, that's incredibly alarming. It basically sounds like Fiji is a disaster and they're desperately trying to whip it into competitive shape.

If someone waited this long, they weren't satisfied with what GTX900 or R9 200 series had to offer from some angle. You continue to make it sound like Fiji is 9 months late or something when NV just launched 980Ti June 1st, that's 15 days ago. From a marketing perspective, 980 sounded nice but the gamers not in a crazy rush to upgrade stayed far away from it knowing what's coming down in the form of Fiji and GM200.

The current state of affairs is there are no NV/AMD GPUs worth buying between a $330 970 and a $650 980Ti. Looking more closely, a $650 only gets one a reference 980TI - a major compromise for overclocking as it runs hot and loud. I would essentially classify a reference 980Ti in many ways similar to as a reference 290X - unless you are gaming with headphones, you have no choice but to pay yet another $40-50 for an after-market 980TI (or go water). Alternatively, you are leaving 20%+ OC headroom on the table or have a jet engine in your rig. Then, we have $690 Gigabyte Gaming 980Ti (OOS) and $700 EVGA SC+ 980Ti and $770 EVGA 980Ti Hybrid (OOS). NV clearly came unprepared and this is a rushed 980Ti launch when after-market 980Ti options are slim pickings and the cards are supply constrained 15 days after launch.

Ok so between sub-$330 290/290X/970 and $690 980Ti, NV/AMD have a gigantic $300-350 gap in the market. A lot of gamers who buy GPUs at BestBuy don't overclock.

If R9 390X is just 5% faster than a 290X, it ends up at 76% at 1440P vs. 980's 80%, for supposedly $110 less. A lot of gamers who were scared off by hot and loud 290 will suddenly notice a card 4-5% slower than a 980 which costs $110 less. That's a ton of free games or a 250GB SSD or moving from an i5 to an i7. But it's highly unlikely AMD will have just a $389 390X and a $649-799 Fury line-up. That's where AMD can really drop the hammer with a $499-549 Fiji PRO and $599 Fiji XT air. I am just using these as hypothetical examples but the point is I never recall any GPU generation where there every card between $330 and $650 was a waste of $$$ and I think AMD will realize this and capitalize on this point. For that reason, a lot of gamers who want to spend $400-550 have no choice but to wait for AMD because NV has nothing worthwhile in the $330<X<$650 space for anyone who pays attention to benchmarks of a 970/290X vs. a 980.

 

[Silverforce11]

The better AMD does in this generation, the better for everyone

Amen.

Doesn't matter if you're an NV fan or AMD fan, competition = gamers win. No competition = companies win.

 

[3DVagabond]

This is dragging on way too long. AMD really needs to crap or get off the pot. A June release is much later than expected anyway, and almost everyone was expecting a Computex reveal. Then AMD specifically said the reveal would be at E3. If they now say "just kidding, we're not going to show you anything but a few worthless marketing slides" then a lot of people are going to decide it just isn't worth the wait. The GTX 980 Ti is already out there and selling like hotcakes. AMD doesn't have the luxury of time. The longer they stretch this out, the less confidence I have in the performance of their product. These repeated delays smell like they're trying to patch things up at the last moment. According to Hardware Canucks, "Drivers are still in their pre-beta form and clock speeds are undergoing final tuning." At this late date, that's incredibly alarming. It basically sounds like Fiji is a disaster and they're desperately trying to whip it into competitive shape.

Actually it's been Q2 since sometime last year. Many just didn't believe it.

 

[Silverforce11]

This is dragging on way too long. AMD really needs to crap or get off the pot. A June release is much later than expected anyway, and almost everyone was expecting a Computex reveal. Then AMD specifically said the reveal would be at E3.

There has been no delays. AMD has never said they would launch sooner than June 2015. For June itself, they never said they would launch at Computex, but at a special event in E3. This they confirmed many months ago.

In fact, what they and their partners have said since last year, is they want to clear inventory before they launch in June 2015.

Wishing or expecting it to come sooner != delays.

 

[looncraz]

A lot of people who had NV and wanted to upgrade but were on the fence about which camp to go with also bought NV. AMD literally skipped an upgrade cycle, or at least have come into the middle of one.

The market does NOT have an "upgrade cycle."

While enthusiasts will watch the GPU releases most people do not. They go to a geeky friend they know with a dollar amount and the geeky friend finds them the proper upgrade.

Or, they go to Best Buy and find a card in their price range, maybe doing a quick google search on their phone to compare one or more options (or asking a no-nothing Best Buy sales rep).

Other times, they save up money and do more research - most people aren't fan boys, most people don't peruse forums when choosing their cards. Most people just read a couple reviews and try to find charts comparing the two, then buy what they can afford that seems like it would give them the least grief.

AMD isn't off the minds of most people - it wasn't too long ago when they were the go-to for excellent, yet still not stupidly overpriced, video cards. Most people who just swapped over to nVidia will swap back to AMD with their next upgrade if their priorities lean that way.

 

[96Firebird]

1. The fact that you bought a 770 during R9 280X generation shows you don't care about price/performance at all. 280X was $299 vs. $380 for 770 2GB (obsolete) and $450 against 770 4GB. Also, the fact that you waited that long to buy a 770 and skipped 1Ghz 7970 cards that were $300 8 months before 770 even released shows you had no interest in any AMD card.

2. 970 was only cheaper than a 290 for 1 month at best. You aren't discussing how an after-market 290 was $350-375 5 months before 970 launched or how 290 was $400 10 months before 970 launched. Again, there was plenty of time to sell a 770 and buy a 290 but you didn't do that either. Instead you waited 10 months to get a card 5% faster. Again, shows you had no interest in buying an AMD card. You aren't fooling anyone trying to spin things as if your 970 purchase was objective. This entire forum already knows you only buy NV cards and the fact that you owned GeForce 5 and 7 and Fermi puts you in a very special group of GPU owners, which I won't name as I'll get an infraction.

Time to eat your words!

You say the 770 2GB was $380 when I was looking? This thread seems to disagree with you... I paid $290 after selling the games that came with my 770, while the 280X was also hovering around $300. You can also see that I was considering older 79xx series cards, just missed out on some of the better deals for them. If you didn't notice, I wanted to upgrade for BF4, which had come out just a month before I started looking at a getting a new card. So that makes everything in point 1 completely wrong.

Shortly after buying my 770, I purchased a 290. Dell never shipped it, so I never got to play with one. I wanted to try out Mantle and do some Windows 7 --> Windows 8.1 benchmarking between the two. That was around the time the mining boom hit and the prices skyrocketed.

As for point 2, guess when I sold my 770? The night the 970 was announced, because I knew the resale value would tank as soon as the 970 was released at an MSRP of $330. Now I was cardless. So my options were a $330 SC 970, a $340 290, or a $450 290X. At launch and at my resolution, the 970 was 20% faster than the 290, and 14% faster than the 290X. Which card is the logical one to choose? I'm sure for you it would be the 290(X), but any sane person can see the 970 was the clear winner. :thumbsup:

Also, I owned an FX5200 because the computer my parents bought came with it. I've never owned a Geforce 7 series, not sure where you pulled that from. I did own an 8800GT 512MB (for the first computer I ever built myself), but lets be real, who didn't own one of these awesome, legendary cards?! You're right, I did own a Fermi. A GTX 460 1GB that overclocked like crazy, another awesome, legendary card.

Now that we have my card purchasing history out of the way, is your foot in your mouth yet? Anyways, is there a reason you're so concerned with my card history? Should we go into yours? Nah, I really don't care.

I hope people who read this can clearly see how you make stuff up out of thin air, just to try to prove a point that is clearly wrong. I've seen you do it time and time again, and I continue to call you out on it.

If you're done attacking me, we can go back to discussing the upcoming Fiji cards, since the 300-series looks like nothing new.

 

[RussianSensation]

Amen.

Doesn't matter if you're an NV fan or AMD fan, competition = gamers win. No competition = companies win.

Fair competition cannot happen as long as there is GW. Even if Fiji = 980Ti in hardware terms, NV's marketing will ensure sites use GW titles in major reviews (despite same sites not using Dirt Showdown since they felt it was unfair to NV). The more GW titles are added to the review, the more the artificial GW advantage will change objective summaries of reviews.

http://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_G1_Gaming/20.html

or

http://techreport.com/review/28356/nvidia-geforce-gtx-980-ti-graphics-card-reviewed/4

As long as these marketing games are used in reviews and their scores are tallied towards a conclusion, the end result cannot be an objective summary table.

GW accomplishes 2 things in 1 - undermines performance on single AMD cards and most likely makes it difficult for AMD's CF since the developer basically neglected AMD's cards throughout development because NV is their key partner.

GW in play, 980 is only 9% behind a 295X2 and 290X can't even beat a 970 at 1440P. The Witcher 3, Project CARS, AC Unity, the entire test suite is turning into GW. Soon Batman AK will be added too and more brand agnostic older games will be dropped. It's turning into "Review of AMD cards in NV-sponsored/source code developed games." Once UE4 games come out, might as well start using Next gen AMD cards vs. older gen NV cards as PhysX is probably going to underline a lot of UE4 titles.

GW + after-market 980Ti vs. reference 290X = 61% faster than a reference 290X = the most perfect situation for NV.

 

[LTC8K6]

Tonga was their chance for a new tech. All they had to do was improve on it. Imagine if we had a 3072 Shader, 384-bit 6GB Tonga as the 390X? More shaders, even if it stayed 32 ROPs it'd be better than Hawaii 64 ROPs and use less power, and 384-bit with color compression should equal 512-bit Hawaii with less power. Even if it were 2816 shaders, it'd still be an improvement over 290X. Just give us 2 new cards (2560 and 3072 Tonga) along with the 285 and Tonga XT:

1792 Tonga to fight the 960 as it does now
2048 to fight the 960Ti and/or the gap between 960 and 970
2560 and 3072 to battle vs 970 and 980

Was it just not worth it to invent new cards since they were so late and 970 already sold so much?

Instead, they won't even give us the full 2048 Tonga.

I say we will get full Tonga, because I demand it.

 

[Silverforce11]

@RS
NV isn't going to play nice with GameWorks. They aren't going to make their libraries open source anytime soon.

Everytime a GW title bombs on AMD, most gamers (from steam forum, neogaf, gaming forums) don't care that its NV sponsor, they blame AMD.

It's up to AMD to dig themselves out of the hole. They need to allocate more resources to driver optimization/dev relations. In cross-platform games, the bulk of the sales are for consoles, thus, the PC port (AMD GE program) needs to have close source AMD features that run much better on GCN**. Being a nice guy when NV is on the offensive with GW is going to be the death of AMD.

**An incentive for gamers to go with AMD GPUs. Currently GW is a massive incentive to go with NV and an added effect of an anti-AMD "drivers suck in new games!"

 

[gamervivek]

285 was like 6% faster overall than 280. A tongaish Hawaii would be on par with gtx980 in sweclockers recent review of 980Ti if you believe that GCN1.1 did nothing over GCN1.0

AMD would do far better with more games like Hitman Absolution and reviewers like TechSpot,

http://www.techspot.com/articles-info/977/bench/Hitman.png

 

[tential]

Lol at buying video cards from Dell in 2014... I love how I saw the thread title and that's the first post I read too. Just lol....

 

[96Firebird]

Lol at buying video cards from Dell in 2014... I love how I saw the thread title and that's the first post I read too. Just lol....

Did you miss the part where I purchased it for $460, and then the price shot up to $579? Do you recall the prices during the mining craze?

Just lol at your reading comprehension.

 

[casiofx]

I wonder why people keep saying increasing memory bandwidth for Hawaii don't give increased FPS.

I only tried 2 benchmarks that I have and both net me 3 to 4 percent increased FPS consistently. I put the core clock of my R9 290 at 1000Mhz and compare the score for 1250Mhz memory speed vs 1500Mhz.

I tested on Unigine Heaven at 1080p extreme preset and Sleeping Dogs 1080p with Extreme AA. Normal AA on sleeping dogs didn't even tax the GPU so there's no gain in FPS.

Going to run 3DMark later to see the effects. I have other game built in benchmarks but I had uninstalled the games (Tomb Raider, Arkham Origins etc)

 

[JDG1980]

GW in play, 980 is only 9% behind a 295X2 and 290X can't even beat a 970 at 1440P. The Witcher 3, Project CARS, AC Unity, the entire test suite is turning into GW. Soon Batman AK will be added too and more brand agnostic older games will be dropped. It's turning into "Review of AMD cards in NV-sponsored/source code developed games." Once UE4 games come out, might as well start using Next gen AMD cards vs. older gen NV cards as PhysX is probably going to underline a lot of UE4 titles.

The problem with this argument is that these are among the most popular games of the year. They're AAA blockbuster titles that people want to play. It would be dishonest for review sites to pretend that they don't exist.

If AMD wants to sue Nvidia for anti-competitive practices, I could get behind that. But bitching about it in public will accomplish nothing except making AMD looks like a bunch of excuse-making whiners.

For what it's worth, this is why AMD is strongly behind DX12: a low-level interface puts more of the onus on game developers, where it belongs, instead of on the GPU vendors. Right now we've gotten into an absurd situation where it is somehow considered the responsibility of AMD and Nvidia to fix crappy, unoptimized game code, not to mention creating custom multi-GPU profiles for every title. From a software engineering perspective, this is incredibly sloppy and hackish. From a business perspective, it tilts the playing field against AMD because of their smaller and more overloaded R&D budget. With any luck, DX12 will create a more sane situation where game performance is the responsibility of the game developers (imagine that!) and AMD and Nvidia can write clean, generic driver code that works on everything without needing patches every two weeks.

 

[JDG1980]

Did you miss the part where I purchased it for $460, and then the price shot up to $579? Do you recall the prices during the mining craze?

Just lol at your reading comprehension.

Yeah, I remember when Dell had a Tahiti or Hawaii card in stock at MSRP and everyone was swarming over it as if it was some kind of insane deal.

Crazy times.

 

[xthetenth]

"Kepler effect"

Kepler Syndrome.

 

[flopper]

For what it's worth, this is why AMD is strongly behind DX12: a low-level interface puts more of the onus on game developers, where it belongs, instead of on the GPU vendors.

Most have no idea how much DX12 will change the market.
developers will move to it faster than any other DX version and games will come faster with engines created for it. AMD bet on it and Fury seems to be designed for it.

 

[Silverforce11]

I wonder why people keep saying increasing memory bandwidth for Hawaii don't give increased FPS.

vram OC on Hawaii boost performance in synthetics, rarely in games. I think only one major game responds to vram OC and its Crysis 3, IIRC.

It's not bandwidth bottlenecked.

 

[Silverforce11]

The problem with this argument is that these are among the most popular games of the year. They're AAA blockbuster titles that people want to play. It would be dishonest for review sites to pretend that they don't exist.

If AMD wants to sue Nvidia for anti-competitive practices, I could get behind that. But bitching about it in public will accomplish nothing except making AMD looks like a bunch of excuse-making whiners.

For what it's worth, this is why AMD is strongly behind DX12: a low-level interface puts more of the onus on game developers, where it belongs, instead of on the GPU vendors. Right now we've gotten into an absurd situation where it is somehow considered the responsibility of AMD and Nvidia to fix crappy, unoptimized game code, not to mention creating custom multi-GPU profiles for every title. From a software engineering perspective, this is incredibly sloppy and hackish. From a business perspective, it tilts the playing field against AMD because of their smaller and more overloaded R&D budget. With any luck, DX12 will create a more sane situation where game performance is the responsibility of the game developers (imagine that!) and AMD and Nvidia can write clean, generic driver code that works on everything without needing patches every two weeks.

You know that's not going to change with DX12, lazy or untalented devs will still be lazy & untalented. These guys rely on engines and 3rd party addons. As long as NV provides close source feature addons, it will be a detriment for AMD.

So AMD needs to play ball, play dirty, play the same game NV is playing. Invest more in dev support, with closed source features. Designed for DX12 and run best on GCN.

 

[Alatar]

Fair competition cannot happen as long as there is GW. Even if Fiji = 980Ti in hardware terms, NV's marketing will ensure sites use GW titles in major reviews (despite same sites not using Dirt Showdown since they felt it was unfair to NV). The more GW titles are added to the review, the more the artificial GW advantage will change objective summaries of reviews.

http://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_G1_Gaming/20.html

or

http://techreport.com/review/28356/nvidia-geforce-gtx-980-ti-graphics-card-reviewed/4

As long as these marketing games are used in reviews and their scores are tallied towards a conclusion, the end result cannot be an objective summary table.

GW accomplishes 2 things in 1 - undermines performance on single AMD cards and most likely makes it difficult for AMD's CF since the developer basically neglected AMD's cards throughout development because NV is their key partner.

GW in play, 980 is only 9% behind a 295X2 and 290X can't even beat a 970 at 1440P. The Witcher 3, Project CARS, AC Unity, the entire test suite is turning into GW. Soon Batman AK will be added too and more brand agnostic older games will be dropped. It's turning into "Review of AMD cards in NV-sponsored/source code developed games." Once UE4 games come out, might as well start using Next gen AMD cards vs. older gen NV cards as PhysX is probably going to underline a lot of UE4 titles.

GW + after-market 980Ti vs. reference 290X = 61% faster than a reference 290X = the most perfect situation for NV.

For some reason I get the feeling that you considered the extremely Gaming evolved title slanted review lineups brand agnostic.

Looking at TPU's lineup I'd say that they're just picking popular and new games. If I was a hardware reviewer I'd do it too. I definitely would not omit some titles because they might be sponsored by one company or the other. That wouldn't actually be serving the actual consumer. You'd just be trying to skew the results from reality into something more synthetic.

Oh and I would like to know how pCARS is a gameworks title. I mean it's fun spreading FUD about gameworks since all the ignorant reddit/wccf/linustechtips crowd believes everything but honestly I prefer the facts. What gameworks features does pCARS include?

And how would witcher 3 be biased as a game choice when TPU even went as far to disable hairworks entirely?

And what harmful gameworks features does unity have? Tesselation lol? If you look at TPUs Unity performance numbers the AMD cards are doing just fine. They're slightly below average but this is normal for a sponsored title. Also happens to NV cards in AMD titles. I don't really remember seeing people complain about sleeping dogs or hitman absolution being used as review games even though they were clearly outliers for poor nvidia performance in games.

But what would you propose? TPU seems to be going as far as to disable some gameworks features not to skew results. Should they just drop NV sponsored titles entirely (since even the non gameworks pCARS isn't okay in your opinion)? What would you pick instead? Shouldn't they just pick the latest big titles to inform the consumer about how their future GPU will perform in the latest titles?

Or are we just playing the GW boogeyman to spread fud?

 

[Enigmoid]

The ironic thing is if you removed GM200 the chart is identical to the 980 launch. Little bit of kepler sickness showing but 970 = 290X.

 

[JDG1980]

The current state of affairs is there are no NV/AMD GPUs worth buying between a $330 970 and a $650 980Ti.

This is true, though I would modify it to specify a reference GTX 970 (which is far superior to most aftermarket versions) at about $350.

GTX 980 should really be about $399. It still has a big pricing gap between it and the GTX 970, which only made sense when the GTX 980 was the flagship halo card (from September 2014 until the release of the Titan X in March 2015).

Looking more closely, a $650 only gets one a reference 980TI - a major compromise for overclocking as it runs hot and loud. I would essentially classify a reference 980Ti in many ways similar to as a reference 290X - unless you are gaming with headphones, you have no choice but to pay yet another $40-50 for an after-market 980TI (or go water).

It's true that a reference 980 Ti is going to fall behind an aftermarket model when it comes to overclocking, but not everyone overclocks. For stock speeds, the 980 Ti stock cooler is a decent choice - nowhere near as terrible as the Hawaii blowers. TPU measured the 980 Ti stock cooler at 42 dbA, whereas the R9 290X in Uber mode hit 50 dbA. This may sound like a small difference, but remember that the decibel scale is logarithmic; a 10 dbA difference means the sound is twice as loud.

Consider, too, that the R9 295 X2 closed loop cooler (which I believe you've praised in the past) measured 41 dbA - almost exactly the same as the 980 Ti stock blower.

If R9 390X is just 5% faster than a 290X, it ends up at 76% at 1440P vs. 980's 80%, for supposedly $110 less. A lot of gamers who were scared off by hot and loud 290 will suddenly notice a card 4-5% slower than a 980 which costs $110 less.

I think the 390X will do a bit better than that; with a 5% increase in core clock and a 20% increase in memory bandwidth, 10% overall performance improvement over the 290X seems likely. This is about what the 270X got over the 7870 from an equivalent boost in core clock (exact same before-and-after speeds, in fact) and a slightly smaller boost in RAM bandwidth. I'm assuming that the 8GB won't do anything to improve performance in a single GPU configuration; even in Crossfire with maxed-out settings, differences are marginal at best.

The big problem is that the 390/390X will still fall flat on their faces in power efficiency. Sure, the AIB coolers can dissipate the heat, but it still has to go somewhere, and the energy still has to come from somewhere. In my part of Florida, we've got a high of 98 degrees scheduled later this week, and my apartment's AC can't keep up. Having a 300W+ space heater in my rig does not sound appealing.

That's where AMD can really drop the hammer with a $499-549 Fiji PRO and $599 Fiji XT air.

To which Nvidia can respond by dropping the price of the GTX 980 to $399, as suggested above. And if they really want to get aggressive, imagine what a price cut on the GTX 980 Ti to $549 would do. Remember, they sold the GK110-based GTX 780 at $499 when they needed to in order to compete with Hawaii. GM200 doesn't have that much bigger a die than GK110 (601 sq.mm. vs 561 sq.mm.), and both cards have similar PCBs with the same cooler, and 12 GDDR5 chips on a 384-bit bus. The 980 Ti achieves its 6GB by using denser chips, but even so, the increase in BoM is probably fairly marginal. At the same time, Nvidia could release a "GTX 980 Ti Black Edition" that has a fully enabled GK200 chip (basically a Titan X with 6GB of RAM and allowing aftermarket designs) for $749 to get some more sales on the high end.

This competition may indeed be good for consumers. But I don't think it is going to help AMD's bottom line much.

 

[chimaxi83]

Wow I wish this Fury thread had Fury info instead of back and forth about Nvidia pricing, GameWorks, and Furmark.