[WCCF] AMD Radeon R9 390X Pictured

[happy medium]

I think you're confused

me?

 

[tg2708]

I think you're confused, Nvidia released GM107 (750 Ti and 750) in 2014, along with GM204. But why are we ending in 2014? From the beginning of 2014 until now, they have also released GM206 and GM200. Just wondering what has been going on all this time...

This. AMD new cards might be a winner but with nvidia releasing newer cards at a faster rate AMD's market share gain might be minimal at best. AMD to me seems to be waiting for nvidia to react then counter by delaying new cards which leaves the latter in the right position to gobble up consumers and money.

 

[JDG1980]

Excuse me but what has nvidias team been working on then? Still using Kepler for mid range on down from 2013? It's ok for nvidia to rebrand 680 to 770 or to not release a 950 ti or 960 non ti? Also keep in mind 960 ti didn't show up til MONTHS after big maxwell. Why must amd present a full stack when nvidia just released a single new chip in all of 2014?

Nvidia released four new chips in 2014: GM107, GM108, GM204, and GK210. And the GM206 and GM200 were released in the first few months of 2015.

People wouldn't be complaining if AMD had released a viable new GCN lineup over the past 12-18 months. But they haven't. The only new GPU that AMD has released in all that time is Tonga. What we've seen of it (which may not be what Tonga is fully capable of) is not particularly impressive. It comes off as a testbed or pipe cleaner for new technologies, not a desirable part in and of itself.

The fact is that, despite trickling the parts out one at a time, Nvidia now has a complete top-to-bottom lineup of Maxwell. Everything down to the $99 GTX 750 has the current architecture. The only Kepler/Fermi rebrands are in the ultra-low-end crap, and no one cares about that. In contrast, AMD is still selling 2012 GCN 1.0 parts for a significant portion of their lineup (265, 270, 270X, 280, 280X). They're going to keep doing this in the OEM market (even in $3000 Apple laptops!). If they do it in the AIB market, it amounts to a total surrender on the innovation front, and a desire to milk existing designs for every possible penny. Consumers should not reward such behavior.

 

[Red Hawk]

I'm really baffled why AMD hasn't released a fully enabled 4 GB version of Tonga. With better tessellation capabilities, color compression, more asynchronous compute engines, it would be in a much better position to compete with Maxwell than Tahiti. Surely some Tonga chips must come out fully functional? Or are they working with a broken process? What gives?

Really, I'm further baffled why there isn't a even a 4 GB version of the R9 285 as is. Nvidia has 4 GB versions of the 960. The 960, which has a 128 bit wide memory bus. Surely 4 GB would let Tonga stretch its legs and at least supplant the 280 (non-X)?

 

[Tuna-Fish]

I'm really baffled why AMD hasn't released a fully enabled 4 GB version of Tonga. With better tessellation capabilities, color compression, more asynchronous compute engines, it would be in a much better position to compete with Maxwell than Tahiti. Surely some Tonga chips must come out fully functional? Or are they working with a broken process? What gives?

Apple pays better than AIBs? Full Tongas currently go into Radeon R9 M295Xs (4GB), which are found in top model iMacs. The new retina iMacs were a surprise hit, beating sales predictions and still selling strong 7 months later, significantly raising the ASP of the Mac division.

We don't have solid numbers on how many retina iMacs are actually specced out with the better GPU, but I would not be surprised if the reason we don't have full Tongas simply that early on, Apple absorbed enough of them to not leave supply for an AIB product.

I very much expect a full Tonga in the 300 lineup.

 

[Azix]

AMD can sell GCN as long as its relevant.

I'm not sure one can say AMD has been behind on releases. It's just perception. In 2014 they put out the 280, 285, 295x2, 265, 250x

2013 they had 290 290x 260x etc.

Now they are going to launch a new stack again. Nvidia launched the 970 980 last year so people think amd is behind but nvidia is not launching anything this year to match the new lineup. So will people say nvidia is behind next year? Nope.

 

[LTC8K6]

Full Tonga is kind of late, now.

It should have been in PC's as the 285X, battling with NV a couple months after the 285 was launched, imo.

But I will still be glad to see it. ::biggrin:

 

[JDG1980]

AMD can sell GCN as long as its relevant.

I'm not sure one can say AMD has been behind on releases. It's just perception. In 2014 they put out the 280, 285, 295x2, 265, 250x

That's one new chip, one dual-GPU card with two chips from the previous year, and a bunch of rebrands.

 

[96Firebird]

me?

No, I was responding to swilli just didn't quote his post. When I started typing my post, yours wasn't there...

AMD can sell GCN as long as its relevant.

I'm not sure one can say AMD has been behind on releases. It's just perception. In 2014 they put out the 280, 285, 295x2, 265, 250x

2013 they had 290 290x 260x etc.

Now they are going to launch a new stack again. Nvidia launched the 970 980 last year so people think amd is behind but nvidia is not launching anything this year to match the new lineup. So will people say nvidia is behind next year? Nope.

Sure, they can keep selling old chips if they want. But it doesn't seem to be working well for them marketshare-wise. I think a new stack of new GPUs would sell well for them, and that is what I hope to see, but rumors are suggesting otherwise.

The only new AMD GPU to come out in 2014 was Tonga (285). The 295X2 may have required some man-hours to get the cooler worked out... The rest of the cards you mentioned were old GPUs with tweaked clocks, aka rebrands. These required little work from AMD to release.

 

[swilli89]

No, I was responding to swilli just didn't quote his post. When I started typing my post, yours wasn't there...



Sure, they can keep selling old chips if they want. But it doesn't seem to be working well for them marketshare-wise. I think a new stack of new GPUs would sell well for them, and that is what I hope to see, but rumors are suggesting otherwise.

The only new AMD GPU to come out in 2014 was Tonga (285). The 295X2 may have required some man-hours to get the cooler worked out... The rest of the cards you mentioned were old GPUs with tweaked clocks, aka rebrands. These required little work from AMD to release.

Yet nvidia has one compelling price/performance card and that is the 970. The 960 is pretty bad at its price point. Who cares if its new silicon? It could be a vacuum tube processor for all i care. I am not a business analyst, I am a gamer that wants max frames per second for my dollar.

 

[JDG1980]

Full Tonga is kind of late, now.

It should have been in PC's as the 285X, battling with NV a couple months after the 285 was launched, imo.

But I will still be glad to see it. ::biggrin:

I think this was due to channel inventory. After the cryptocoin bubble burst (around the end of Q1 2014), AMD was left with a huge oversupply of Tahitis, which to make matters worse had to compete with tons of cheap, used mining cards. This meant that AIBs weren't going to want to take Tonga when they already had plenty of inventory of Tahiti on hand that wasn't selling. I think the only reason the R9 285 was released is that AMD needed something to do with the lower binned chips from wafers ordered for Apple. My best guess (and this is an educated guess, not inside info) is that wafers are ordered by AMD on an as-needed basis, with Apple getting the best chips, mid-range chips going into the FirePro W7100 and Dell's Alienware, and trash silicon going into the R9 285 cards.

Perhaps a better question is why Tonga is made at TSMC rather than GloFo. It would have made more sense to do the latter, since the process is said to be lower-leakage and better optimized for GPUs, and because AMD needs to buy products from GloFo as part of the WSA anyway. The only thing I can think of is that AMD was originally planning to fab Tonga on 20nm TSMC, and this turned out to be unworkable and/or economically unviable. Delay wasn't an option because of the Apple contract, so AMD had to rush to get Tonga out the door, and the only way to do that was a fairly direct port to TSMC 28nm. This would actually explain quite a lot about why Tonga fell short of expectations.

 

[JDG1980]

Yet nvidia has one compelling price/performance card and that is the 970. The 960 is pretty bad at its price point. Who cares if its new silicon? It could be a vacuum tube processor for all i care. I am not a business analyst, I am a gamer that wants max frames per second for my dollar.

This assumes you care about perf/$ in gaming, and nothing else. If that's the case, then yes, a Hawaii card is a great deal (Newegg currently has a Sapphire R9 290 for $269.99 with $20 MIR plus DiRT Rally and an 8GB DDR3 DIMM thrown in).

If you care at all about perf/watt, then Nvidia stomps all over AMD; nothing comes close to Maxwell. If you want the best performance no matter what, then Nvidia currently has the upper hand with Titan X (though Fiji may change that).

If you just care about gaming (and assuming you don't play all GameWorks titles) then the GTX 960 is probably a poor buy. However, it is the best HTPC card on the market at this time: low power usage, semi-fanless, HDMI 2.0 support, hardware HEVC decoding, and plenty of shader power for madVR. As I don't play AAA games but do use madVR, I'm seriously considering a GTX 960 for my next build. That said, I'll probably go with AMD if they bring something competitive to the table. Tonga ported to GloFo with lower power consumption (<150W) and updated to include HEVC decoding would probably win me over.

 

[96Firebird]

Yet nvidia has one compelling price/performance card and that is the 970. The 960 is pretty bad at its price point. Who cares if its new silicon? It could be a vacuum tube processor for all i care. I am not a business analyst, I am a gamer that wants max frames per second for my dollar.

I think you missed the gist of my post. I am wondering what the AMD engineers are working on, since it isn't showing much on the GPU side. Maybe the hardware engineers are working on bettering the APUs, maybe porting to GloFo is labor intensive? Maybe Fiji and HBM took a lot of their resources? If so, I hope it pays off for them with experience when it comes time to bring the rest of their lineup with HBM.

 

[swilli89]

I think you missed the gist of my post. I am wondering what the AMD engineers are working on, since it isn't showing much on the GPU side. Maybe the hardware engineers are working on bettering the APUs, maybe porting to GloFo is labor intensive? Maybe Fiji and HBM took a lot of their resources? If so, I hope it pays off for them with experience when it comes time to bring the rest of their lineup with HBM.

Maybe I did, sorry. I think their engineering resources are stretched even thinner than we had previously thought. In that one comment where AMD mentioned that they had never even assigned a single engineer to memory allocation or what have you. To me this is a sign they could be lacking manpower. I know a quality engineer is going to cost at least $100,000 a year so it would be typical upper management idiocracy to cut a handful of engineers to save a million in a fiscal year..

Also, I'm not sure if they are interchangeable but I have a suspicion that their majority of researchers have been working on Zen. If they came out with a truly robust core it could right the ship for the next 5 years at least.

 

[Bobisuruncle54]

From what I've been reading here and on other forums, the lists below seem to be the best and worst case scenarios that people have been talking about. One thing that stands out is the question of how AMD will structure their line up with regards to pricing given that Fiji looks like it will have a lot more SPs than the next chip down which is likely to be a tweaked Hawaii design so the performance gap could be substantial. Either it could mean AMD is going to enter with aggressive pricing (compared to Titan but still higher than the overpriced 980 (relative to the 970)) or present it as a much higher tiered product with pricing much similar to Nvidia's top end.

The lack of leaks and information about the entire lineup is interesting and the OEM 300 series has caused some controversy, however it's possible that these cards could be named the 400 series and as a result would not have appeared in any drivers currently out. At the moment it seems anything is possible but one thing is clear: AMD have to bring a noticeable performance benefit to consumers in every tier along with updated features, better power consumption and better coolers right out of the gate to change the perception of their brand. Whether this will translate into sales isn't a certainty, but a surprise like the 4800 series or the 9700 is needed at this point to give them a significant boost.

This might not be the best time to do it with regards to fabrication shrink schedules but it is for them as a business as they need a meaningful response to current consumer demands to boost their market share.

Caveat: I live in the UK so I'm going off their prices at the moment, these scenarios might be a little bit different everywhere else, but the overall message probably still holds true.

Best case scenario - GloFo 28nm production

390X/Fury Maxx - Fiji XT

• £550 (Water cooled edition)
• 4096 Shaders
• 1050Mhz
• 8GB HBM
• Titan rivalling performance

390/Fury Pro - Fiji Pro

• £450
• ~3584 Shaders?
• 1000Mhz
• 6GB HBM
• 980Ti rivalling performance

380X - Hawaii XT

• £370
• 2816 Shaders (updated GCN architecture)
• 1100Mhz (due to better GloFo process)
• 8GB GDDR5
• 980 rivalling performance

380 - Hawaii Pro

• £300
• 2560 Shaders (updated GCN architecture)
• 1050Mhz
• 8GB GDDR5
• 970+ performance

370X - Tonga XT

• £170
• 2048 Shaders (updated GCN architecture)
• 1100Mhz
• 4/8GB GDDR5
• 960+ performance

370 - Tonga Pro

• £140
• 1792 Shaders (updated GCN architecture)
• 1000Mhz
• 4GB GDDR5
• 960 performance

Worst case scenario - TSMC 28nm production

390X/Fury Maxx - Fiji XT

• £650+ (Water cooled edition)
• 4096 Shaders
• 1050Mhz
• 4GB HBM
• 980Ti rivalling performance

390/Fury Pro - Fiji Pro

• £550+
• ~3584 Shaders?
• 1000Mhz
• 4GB HBM
• 980+ performance

380X - Hawaii XT (rebranded 290X)

• £350

380 - Hawaii Pro (rebranded 290)

• £280

370X - Tonga XT ("full" Tonga)

• £200

370 - Tonga Pro (rebranded 285)

• £160

It's amazing just how little we know for a fact at the moment, but AMD must surely know that they can't make any assumptions and need to be aggressive to build back up.

 

[PG]

Everything down to the $99 GTX 750 has the current architecture.

This isn't true. It's optimized Kepler, and is missing features compared to the later GTX 960/970/980.

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/4

"When NVIDIA introduced the Maxwell 1 architecture and the GM107 based GTX 750 series, one of the unexpected aspects of their decision was to release these parts as members of the existing 700 series rather than a newer series to communicate a difference in features. However as it turned out there really wasn’t a feature difference between it and Kepler; other than a newer NVENC block, Maxwell 1 was for all intents and purposes an optimized Kepler architecture. It was the same features built upon the efficiency improvements of the Maxwell architecture.

With that in mind, along with the hardware/architectural changes we’ve listed earlier, the other big factor that sets Maxwell 2 apart from Maxwell 1 is its feature set. In that respect Maxwell 2 is almost a half-generational update on its own, as it implements a number of new features that were not present in Maxwell 1. This means Maxwell 2 is bringing some new features that we need to cover, but it also means that the GM204 based GTX 900 series is feature differentiated from the GTX 600/700 series in a way that the earlier GTX 750 series was not."

 

[Pinstripe]

With that in mind, along with the hardware/architectural changes we’ve listed earlier, the other big factor that sets Maxwell 2 apart from Maxwell 1 is its feature set. In that respect Maxwell 2 is almost a half-generational update on its own, as it implements a number of new features that were not present in Maxwell 1. This means Maxwell 2 is bringing some new features that we need to cover, but it also means that the GM204 based GTX 900 series is feature differentiated from the GTX 600/700 series in a way that the earlier GTX 750 series was not."

I already have that hunch that in Q1 2016, Nvidia will launch the GTX 950 and 950 Ti as shrinked Maxwell (aka GP100), followed by the real Pascal (GP204/200) in Summer/Fall 2016.

 

[Silverforce11]

From what I've been reading here and on other forums, the lists below seem to be the best and worst case scenarios that people have been talking about.

Now you just have to think which scenario is more likely.

Would AMD under Lisa Su decide to go premium with Radeon Fury branding if it did not convincingly beat Titan X? That would be major flop/fail if it only rivals 980Ti performance (or worse, those fools with clickbait Fiji XT = 980 + 20%!).

Would they do such a massive publicity stunt on Time Square if they weren't confident in their ability to deliver?

Would tech journalists with first hand experience make statements like "it's a lot better than what leaks are suggesting (50% over R290X from last year via ChipHell)"?

Welcome to the HBM revolution!

 

[BHZ-GTR]

@Bobisuruncle54@

Why mention the best graphics card for Global Foundries 390X has 8 GB of memory while TSMC said in 4GB ?

Explain To Me Please ?

 

[Cloudfire777]

Prediction:

Radeon Fury 4GB HBM 4096 shaders: 100% - $800-$850 - 350W TDP
GTX Titan X 12GB GDDR5 3072 cores: 95% - $999 -250W TDP
GTX 980Ti 6GB GDDR5 2816 cores: 90% - $650 - 235W TDP

This one I`m not fully sure are coming or not:
(Fiji Pro/unknown name 4GB HBM 3500 shaders : 85% - $700 - 320W TDP)

 

[Silverforce11]

You should predict gaming power consumption (avg gaming load) because AMD TDP != NV TDP, that should be quite clear to anyone who has paid attention for the past many generations.

R290X TDP = 300W
R290X gaming avg power use = 225-250W (varies on models)

NV's TDP = their gaming avg power use, it lines up well.

 

[maddie]

From what I've been reading here and on other forums, the lists below seem to be the best and worst case scenarios that people have been talking about. One thing that stands out is the question of how AMD will structure their line up with regards to pricing given that Fiji looks like it will have a lot more SPs than the next chip down which is likely to be a tweaked Hawaii design so the performance gap could be substantial. Either it could mean AMD is going to enter with aggressive pricing (compared to Titan but still higher than the overpriced 980 (relative to the 970)) or present it as a much higher tiered product with pricing much similar to Nvidia's top end.

The lack of leaks and information about the entire lineup is interesting and the OEM 300 series has caused some controversy, however it's possible that these cards could be named the 400 series and as a result would not have appeared in any drivers currently out. At the moment it seems anything is possible but one thing is clear: AMD have to bring a noticeable performance benefit to consumers in every tier along with updated features, better power consumption and better coolers right out of the gate to change the perception of their brand. Whether this will translate into sales isn't a certainty, but a surprise like the 4800 series or the 9700 is needed at this point to give them a significant boost.

This might not be the best time to do it with regards to fabrication shrink schedules but it is for them as a business as they need a meaningful response to current consumer demands to boost their market share.

Caveat: I live in the UK so I'm going off their prices at the moment, these scenarios might be a little bit different everywhere else, but the overall message probably still holds true.

Best case scenario - GloFo 28nm production

390X/Fury Maxx - Fiji XT

• £550 (Water cooled edition)
• 4096 Shaders
• 1050Mhz
• 8GB HBM
• Titan rivalling performance

390/Fury Pro - Fiji Pro

• £450
• ~3584 Shaders?
• 1000Mhz
• 6GB HBM
• 980Ti rivalling performance

380X - Hawaii XT

• £370
• 2816 Shaders (updated GCN architecture)
• 1100Mhz (due to better GloFo process)
• 8GB GDDR5
• 980 rivalling performance

380 - Hawaii Pro

• £300
• 2560 Shaders (updated GCN architecture)
• 1050Mhz
• 8GB GDDR5
• 970+ performance

370X - Tonga XT

• £170
• 2048 Shaders (updated GCN architecture)
• 1100Mhz
• 4/8GB GDDR5
• 960+ performance

370 - Tonga Pro

• £140
• 1792 Shaders (updated GCN architecture)
• 1000Mhz
• 4GB GDDR5
• 960 performance

Worst case scenario - TSMC 28nm production

390X/Fury Maxx - Fiji XT

• £650+ (Water cooled edition)
• 4096 Shaders
• 1050Mhz
• 4GB HBM
• 980Ti rivalling performance

390/Fury Pro - Fiji Pro

• £550+
• ~3584 Shaders?
• 1000Mhz
• 4GB HBM
• 980+ performance

380X - Hawaii XT (rebranded 290X)

• £350

380 - Hawaii Pro (rebranded 290)

• £280

370X - Tonga XT ("full" Tonga)

• £200

370 - Tonga Pro (rebranded 285)

• £160

It's amazing just how little we know for a fact at the moment, but AMD must surely know that they can't make any assumptions and need to be aggressive to build back up.

Best/Worst case for whom? Best for either company is worse for us the user. That is the point the non-fanatical posters here, are trying to convey.

I tell you if your best case scenario pans out, the fanatical Nvidia supporters will criticize the low prices. I can see it now.

AMD is too cheap.

 

[happy medium]

whats a fury pro? who made this up?

 

[Tuna-Fish]

whats a fury pro? who made this up?

"PRO" is part of the AMD cut chip naming lineup. Full chip = XT, cut a little = PRO, cut even more = LE, CE. These suffixes are not found in actual product marketing names, but in the codenames of the chips.

I don't think the poster above meant that "Fury Pro" would be the actual name of the product, just that since not every Fiji will come out right, it makes sense for them to have a slightly weaker, cut down product. I agree in this assessment -- there will almost certainly be a Fiji PRO. Under which marketing name this will end up is less certain.

 

[jpiniero]

If the Fury MAXX is really 4096 shaders and 300 W TDP, it should be faster than the Titan X given that the water should keep the clocks up.