Not wanting to pull people away from Guru3D, but the extremetech page below has some info on the tonga shaders GCN 1.2. (edit: there are pics on the site as well ) Hilberts Tonga review doesn't have this info, or i'd have linked to it. http://www.extremetech.com/gaming/189034-amd-radeon-r9-285-review-the-gcn-1-2-torpedo-that-takes-out-nvidias-gtx-760 Excerpt :

What’s new about the R9 285? First, the R9 285 supports TrueAudio, AMD’s superior Crossfire scaling solution, XDMA, and the modified GCN front end that AMD introduced with the Hawaii family. That means eight Asynchronous Command Engines (up from two) and support for four primitives per clock cycle (everything but the R9 290 cards top out at two primitives per clock). AMD claims this will give the R9 285 a huge tessellation performance advantage. Overall, this sure looks like GCN 1.2 (though AMD reminded us that such nomenclature was created by the press; it’s not an official designation). The R9 285’s new features are as follows: Support for 16-bit floating point and integer values. This is advantageous in low power settings, where 16-bit precision is more than good enough. Up until now, AMD simply used 32-bit values for all calculations — switching to 16-bit in specific contexts will save power and internal bandwidth without noticeably degrading image quality. AMD also notes that this new Tonga GPU supports sharing data between SIMD lanes and has an improved task scheduler. Color compression. Frame buffer data is now stored in a lossless compressed format and the GPU can read and write compressed data. This leads to substantial bandwidth efficiency gains and allows AMD to shrink the memory bus from 384 bits to 256 bits. Such efficiency gains would normally result in a smaller die, but that’s not the case here. The R9 285 has a 359mm2 die and 5 billion transistors, compared to 352mm2 and 4.3 billion transistors for the old Radeon HD 7970/R9 280X. AMD has confirmed to us that the new chip has 32 CUs total (with 28 functional), implying that AMD has the option to bring a new version of the core out if it chooses to do so — but even the additional CUs don’t explain why a GPU with a smaller memory bus is both larger and more dense than its predecessor. New video decode block. The R9 285 also includes a new video decoder block for full hardware decode of H.264 4K streams. H.264 base, main, and high profile up to 5.2 are all supported. That means the new block can handle decoding 4096×2304 streams at 60 fps. There’s also a new fixed function video transcoder unit (VCE) that supports full hardware encoding to H.264, but AMD didn’t provide additional details on features or capabilities that distinguish this new unit from previous generation hardware.

Not wanting to pull people away from Guru3D, but the extremetech page below has some info on the tonga shaders GCN 1.2. (edit: there are pics on the site as well ) Hilberts Tonga review doesn't have this info, or i'd have linked to it. http://www.extremetech.com/gaming/189034-amd-radeon-r9-285-review-the-gcn-1-2-torpedo-that-takes-out-nvidias-gtx-760 Excerpt :

http://images.anandtech.com/doci/8460/Tessmark_575px.png I don't know about huge, but it gives some tessellation performance. (The 980 is up at 232FPS for comparison) But yeah I'm expecting the 3xx series will all be based on Tonga despite rumors. Makes no sense for it not to be.

So they compete with 980 TI with liquid cooling? Isn't that a bit desperate?

No there is definitely going to be a version without liquid cooling. There are also versions of the 980 ti with liquid cooling as you very well know and that does not mean the architecture is inefficient. According to confirmed reports (wccftech) Fury is going to sport 4096 stream processors and 128 rops, which makes it much more efficient than the 290x and probably even more efficient than maxwell.

Do you think there is some horrific difference between 250 and 300 Watts? Or do you think people who use water cooling for CPU or even GPU are desperate?

Didn't see any reports saying that Fiji is 300W, but I may have missed something. Source?

Didn't see any reports saying that Fiji is 300W, but I may have missed something. Source?

What kind of cooling capacity you expect from one 12cm radiator it has? Look here: http://www.guru3d.com/articles_pages/cpu_cooler_roundup_2013_q1_review,11.html Even Corsair H100i which has 24cm radiator does not keep exactly well once power consumption of chip goes over 350W which may be what CPU there eats. One 12cm radiator will simply not do job well if card goes over 300W, and people will not buy card with special cooling solution which goes to 90C.

http://images.anandtech.com/doci/8460/Tessmark_575px.png I don't know about huge, but it gives some tessellation performance. (The 980 is up at 232FPS for comparison) But yeah I'm expecting the 3xx series will all be based on Tonga despite rumors. Makes no sense for it not to be.

Considering the Tonga chip has quite a few CU's less than the 290, it looks quite good. Also found this from gibbo over on Overclockers.co.uk : http://forums.overclockers.co.uk/showpost.php?p=26883790&postcount=196

AMD Technical have explained why this is:- To provide improved geometry performance on the order of 2-4x the tessellation throughput we did a number of things Compared to R9 280 Series: • R9 285 can execute 4 primitives per clock vs. the R9 280 series which can only execute 2 primitives per clock (Quad Prim vs. Dual Prim) Compared to the tessellation unit in the R9 290 Series we’ve improved performance through the following • Larger parameter cache, backed by L2 read/write cache • Improved work distribution between geometry front-end units • Improved vertex re-use for better performance with many small triangles In summary it was done through improved HW and as we do with each generation we are constantly improving our HW as the demands of new games continue to grow. I hope this answers your question.....

What kind of cooling capacity you expect from one 12cm radiator it has? Look here: http://www.guru3d.com/articles_pages/cpu_cooler_roundup_2013_q1_review,11.html Even Corsair H100i which has 24cm radiator does not keep exactly well once power consumption of chip goes over 350W which may be what CPU there eats. One 12cm radiator will simply not do job well if card goes over 300W, and people will not buy card with special cooling solution which goes to 90C.

Isn't the Radeon R295x2 cooled with a Hybrid 120mm single radiator ( the Power stage/ram is aircooled, but the two GPU's are watercooled iirc) ?

Amd is quite late to the party. I have my 980s for over 7 months now. And yet they release products that are to compete with the old nvidia hardware for a better price.... well. And if they can only come to the range of a 980ti with HBM I don't think their architecture is any good...

For me (not saying you're wrong at all, just my opinion), I think Nvidia showed up quite early. I think they were a little hurried because of the R3xx (can charge a premium now before the competition arrived), which I believe helped AMD plan the roadmap we see from them now. AMD seems more concerned with W10 & DX12 launch which is what I'm more concerned about.

Isn't the Radeon R295x2 cooled with a Hybrid 120mm single radiator ( the Power stage/ram is aircooled, but the two GPU's are watercooled iirc) ?

It draws up to around 500W, in Hilbert's review it has quite good temperatures. Other sites are not as kind in their reviews. In Hilbert's tests GPU goes up to 68°C and PCB above VRMs to 71°C. On other site I saw 88°C for GPU and over 100°C on VRMs. Which I would say is quite more accurate. And here I can't trust Hilbert's review, not because he cheats or anything. He reports what he measures, but there are many variables in testing, like type or load, duration, airflow. And if HH's review was taken as correct then this one 12cm rad keeps 500W GPUs at 68°C. And it raises question about power consumption of i7-3770k which ends up running 75°C with 24cm H100i. Guru3D: i7-3770k@4.6Hz

http://images.anandtech.com/doci/8460/Tessmark_575px.png I don't know about huge, but it gives some tessellation performance. (The 980 is up at 232FPS for comparison) But yeah I'm expecting the 3xx series will all be based on Tonga despite rumors. Makes no sense for it not to be.

Being ~70% faster is really huge from an otherwise weaker card. If they scale the chip surpass the 290(X), that could equate to something at least 20% faster than the 980 in tessellation. The 980 is 52% faster than the 290X in that bench.

It draws up to around 500W, in Hilbert's review it has quite good temperatures. Other sites are not as kind in their reviews. In Hilbert's tests GPU goes up to 68°C and PCB above VRMs to 71°C. On other site I saw 88°C for GPU and over 100°C on VRMs. Which I would say is quite more accurate. And here I can't trust Hilbert's review, not because he cheats or anything. He reports what he measures, but there are many variables in testing, like type or load, duration, airflow. And if HH's review was taken as correct then this one 12cm rad keeps 500W GPUs at 68°C. And it raises question about power consumption of i7-3770k which ends up running 75°C with 24cm H100i. Guru3D: i7-3770k@4.6Hz

It would depend on fan speed, a LOT. Also, not all rads are the same thickness, or material, or number of fins, or build quality. Why would you trust this other site over Guru3D ? They could have left auto fan on for all we know.

Being ~70% faster is really huge from an otherwise weaker card. If they scale the chip surpass the 290(X), that could equate to something at least 20% faster than the 980 in tessellation. The 980 is 52% faster than the 290X in that bench.

Me too. I would be happy to just very conservatively double the graphs R285 scores, which would be about just above the 980, possibly on par with the Ti.

And if HH's review was taken as correct then this one 12cm rad keeps 500W GPUs at 68°C. And it raises question about power consumption of i7-3770k which ends up running 75°C with 24cm H100i. Guru3D: i7-3770k@4.6Hz

It would depend on fan speed, a LOT. Also, not all rads are the same thickness, or material, or number of fins, or build quality. Why would you trust this other site over Guru3D ? They could have left auto fan on for all we know.

fyi: If you dared to click Hilbert's source and looked, then you would know that mentioned 24cm rad with H100i ended up as 2nd best cooling solution (since it was running at max). So, come on, use your imagination and draw your own conclusion about power consumption and heat generated by i7-3770k in scenario where 12cm rad keeps 500W chips 7°C cooler than 24cm rad keeps this i7-3770k. Tell us what that i7-3770k actually eats and makes as heat. Or use what's left from your common sense and realize that 12cm rad will not keep 500W thing at 68°C.

fyi: If you dared to click Hilbert's source and looked, then you would know that mentioned 24cm rad with H100i ended up as 2nd best cooling solution (since it was running at max). So, come on, use your imagination and draw your own conclusion about power consumption and heat generated by i7-3770k in scenario where 12cm rad keeps 500W chips 7°C cooler than 24cm rad keeps this i7-3770k. Tell us what that i7-3770k actually eats and makes as heat. Or use what's left from your common sense and realize that 12cm rad will not keep 500W thing at 68°C.

I clicked, I looked. I use a H100i on my unlocked i5. I'm under 60°C with just the CPU burn. More if i have GPU and CPU Burn at the same time. Its also in a case. Hilberts is open desk. It depends on the type and build of radiator, and the fans, and the pump, and the ambient temp, and even the luck of the draw with the CPU!. Show me the link to the other site, and i'll check that link out too. 😉 My 290 has a 240 setup, and overclocked it goes from 32 to 42°C full load. I could have easily used a 120mm. There are 120mm radiators rated for well over 500W, not that they are the ones being used here. They are just admitting that not everyone can fit a 240mm radiator in their case, whereas just about anyone can fit a 120mm. Also, the air is supposed to come IN from the back of the case to the R295x2 Rad, so fresh air only, not warm case air, which some sites may get wrong. edit - I must admit I wouldn't buy a H100i again. I would just build a custom loop and be done with it. The pump isn't great, and no way to tell if the coolant level is good or not. One site that did a test (Bit-Tech I think) found only a half of a mug of coolant. The radiator is not as good quality as a good separate one would be like. Right now, I would go with a 120mm for the CPU, and a 240mm for the GPU, 35 and 55mm thick, one loop.

@Evildead666: My bad for not putting all sources immediately as I see them. Here is TPU review 70°C GPU, 105°C VRMs PCB: TPU - r9-295x2 Here is Ocers who got it bit higher: OC.C - r9-295x2 They have measured dT to room temperature 50°C, can't find that test with over 80°C on GPU. So there is possibility I misread temp from VRMs area somehow since r9-295x2 uses temperature targets and if it can't keep to it by cooling, it starts to throttle. But it can be unlocked and some people OCed it, so that one may not be stock in the end, making it invalid source anyway. My cooling is rated 300W and it keeps 205W GPU under 61°C in burn tests (57°C max in Witcher 3 - GPU loaded to 100%), my guess is that it would do around 70~75°C on r9-290x. And I think AMD has to keep Fiji under 70°C for both water and air cooling without making a lot of noise anyway, to be attractive. And that does not leave much room for power consumption over 300W. +if it is around 300, and can catch up to 980Ti/T-X, then reasonable price will make those 50W meaningless for most. But if it runs 350W then it is already 100W difference and it will feel same way as 290x vs 970. Maybe even worse, because 290x is old thing vs 970 new generation. And now it is new vs new. Btw, your dT is like 18~20°C to ambient, even with 24cm rad it is way too good. That idle dT is 8~10°C above room, which matches mine. Water cooling usually do not match or win in idle temperatures over air as they show teeth under load. Do you have some kind of control over pump/fans on rad?

What kind of cooling capacity you expect from one 12cm radiator it has? Look here: http://www.guru3d.com/articles_pages/cpu_cooler_roundup_2013_q1_review,11.html Even Corsair H100i which has 24cm radiator does not keep exactly well once power consumption of chip goes over 350W which may be what CPU there eats. One 12cm radiator will simply not do job well if card goes over 300W, and people will not buy card with special cooling solution which goes to 90C.

You cannot compare like that, a 4770k has a die size of 177mm2 while a Titan X has a massive 601mm2 die. That's more than 3 times the size of the CPU. It's quite easy to tell that the GPU will be far easier to cool down since the surface area is far greater. My friend, a 3770k @ 4.6GHz could no way in 7 hells dish out 350W TDP. Hell, I've never had a mobo that was able to feed the CPU with more than 300W. My 2600k @ 4.9GHz doesn't reach 200W. I know this because I had my power limit set to 300W on my mobo and I just got curious to see what the actual consumption would be. I dumbed it down to 200 and it still went on perfectly stable. It failed to boot on 150W, but it did boot on 175W. So 150-175 for a 4.9GHz 2600k, my assumption is around 130W for a 4.6GHz 3770k, highly depends on the chip because voltage plays a big role. But having such a small surface area makes it heat up like crazy. And please note that my test was about actual power consumption the motherboard was sending into the cpu, not TDP which is going to be lower than the actual consumption. Albeit just slightly lower. But for the sake of the argument, if that GPU would dish out 300W TDP why would AMD go underwater? It's completely unnecessary. By the sound of it Fury X might just as well dish out far more than 300W. Theoretically the card could draw up to 375W, 2 8-pin connectors offer 300W and the PCI-Express interface offers 75W. I honestly wouldn't be surprised if it would be around 350W. I really don't think they'd just throw water at it because why not. My 2 cents. Of course, this is purely hypothetical because we know nothing about the card at hand. We can only extrapolate.

It draws up to around 500W, in Hilbert's review it has quite good temperatures. Other sites are not as kind in their reviews. In Hilbert's tests GPU goes up to 68°C and PCB above VRMs to 71°C. On other site I saw 88°C for GPU and over 100°C on VRMs. Which I would say is quite more accurate. And here I can't trust Hilbert's review, not because he cheats or anything. He reports what he measures, but there are many variables in testing, like type or load, duration, airflow. And if HH's review was taken as correct then this one 12cm rad keeps 500W GPUs at 68°C. And it raises question about power consumption of i7-3770k which ends up running 75°C with 24cm H100i. Guru3D: i7-3770k@4.6Hz

Hey fox my experience tells me that the HH review is more closely aligned to real use temps.. ive witnessed very similar temps with one fan on the 295 rad. With 2 fans in push pull its usually 62 deg.. ive also got an h100i and good air flow.

@Evildead666: My bad for not putting all sources immediately as I see them. Here is TPU review 70°C GPU, 105°C VRMs PCB: TPU - r9-295x2 Here is Ocers who got it bit higher: OC.C - r9-295x2 They have measured dT to room temperature 50°C, can't find that test with over 80°C on GPU. So there is possibility I misread temp from VRMs area somehow since r9-295x2 uses temperature targets and if it can't keep to it by cooling, it starts to throttle. EDIT - Cheers, gonna have a read tomorrow. 😉 But it can be unlocked and some people OCed it, so that one may not be stock in the end, making it invalid source anyway. My cooling is rated 300W and it keeps 205W GPU under 61°C in burn tests (57°C max in Witcher 3 - GPU loaded to 100%), my guess is that it would do around 70~75°C on r9-290x. And I think AMD has to keep Fiji under 70°C for both water and air cooling without making a lot of noise anyway, to be attractive. And that does not leave much room for power consumption over 300W. +if it is around 300, and can catch up to 980Ti/T-X, then reasonable price will make those 50W meaningless for most. But if it runs 350W then it is already 100W difference and it will feel same way as 290x vs 970. Maybe even worse, because 290x is old thing vs 970 new generation. And now it is new vs new. Btw, your dT is like 18~20°C to ambient, even with 24cm rad it is way too good. That idle dT is 8~10°C above room, which matches mine. Water cooling usually do not match or win in idle temperatures over air as they show teeth under load. Do you have some kind of control over pump/fans on rad?

For the GPU, the 240 is on the front of the case, with 4 Corsair SP120 in push pull, at 12V. 50mm thick, with a laing DDC, also at 12V. I never mentioned ambient, which is about 16°C (unheated corridor). The radiator is well overkill for a 290 with a mild OC. No fan control, 12V all the time.. All the watercooling and Fans have a separate PSU. Thats why i said it depends on the radiator, and the fanspeed, even the ambient. It was a planned build of a whole system, rather than upgrading every year. Wait a couple of years, maybe 3, and do another full system. If i can hold out that long. 🙂 Its only a few days left, and i'm sure Hilbert has been playing with his Fiji for a bit now, and is dying to show us the specs, fps and wattage 🙂

You cannot compare like that, a 4770k has a die size of 177mm2 while a Titan X has a massive 601mm2 die. That's more than 3 times the size of the CPU. It's quite easy to tell that the GPU will be far easier to cool down since the surface area is far greater. My friend, a 3770k @ 4.6GHz could no way in 7 hells dish out 350W TDP. Hell, I've never had a mobo that was able to feed the CPU with more than 300W. My 2600k @ 4.9GHz doesn't reach 200W. I know this because I had my power limit set to 300W on my mobo and I just got curious to see what the actual consumption would be. I dumbed it down to 200 and it still went on perfectly stable. It failed to boot on 150W, but it did boot on 175W. So 150-175 for a 4.9GHz 2600k, my assumption is around 130W for a 4.6GHz 3770k, highly depends on the chip because voltage plays a big role. But having such a small surface area makes it heat up like crazy. And please note that my test was about actual power consumption the motherboard was sending into the cpu, not TDP which is going to be lower than the actual consumption. Albeit just slightly lower. But for the sake of the argument, if that GPU would dish out 300W TDP why would AMD go underwater? It's completely unnecessary. By the sound of it Fury X might just as well dish out far more than 300W. Theoretically the card could draw up to 375W, 2 8-pin connectors offer 300W and the PCI-Express interface offers 75W. I honestly wouldn't be surprised if it would be around 350W. I really don't think they'd just throw water at it because why not. My 2 cents. Of course, this is purely hypothetical because we know nothing about the card at hand. We can only extrapolate.

I think they're throwing water at a flagship card, because its going to be expensive anyway, and in low volume, maybe very low volume. May as well give it a good cooling system right off the bat. Lets face it, people who are likely to buy this are either going to leave it stock and enjoy the quietness, or overclock the boll*cks off it. Also, the renders I saw all had 8+6 pin headers. Did I miss a Pic ?

For the GPU, the 240 is on the front of the case, with 4 Corsair SP120 in push pull, at 12V. 50mm thick, with a laing DDC, also at 12V. I never mentioned ambient, which is about 16°C (unheated corridor). The radiator is well overkill for a 290 with a mild OC. No fan control, 12V all the time.. All the watercooling and Fans have a separate PSU. Thats why i said it depends on the radiator, and the fanspeed, even the ambient. It was a planned build of a whole system, rather than upgrading every year. Wait a couple of years, maybe 3, and do another full system. If i can hold out that long. 🙂 Its only a few days left, and i'm sure Hilbert has been playing with his Fiji for a bit now, and is dying to show us the specs, fps and wattage 🙂

That ambient 16°C explains a lot, thanks. Had bit trouble fitting it into water vs air behavior. And you are right that 295x2 has 5~6 times higher area to transfer heat into cooling solution than that i7. With Fiji it will be bit less but still... from photo I took 5x7mm for HBM and chip measured 24x20mm that's only 480mm^2 in comparison to 438mm^2 Hawaii. (AMD fitted 3.1b transistor Carrizo in same area as 2.4b transistor Kaveri and still managed to bump GPU clock by +-150MHz. So 28nm still improves.) ...ouch, just realized something... There was not even guess where it is built. Is it TSMC or GloFo?