[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

[ShintaiDK]

Considering the latest info and the rebrand up to the 370 series. Maybe Fiji is the 380 series and 2xFiji the 390 series.

DX12 looks to improve multi GPU scaling quite well. So its not exactly silly to bet on something like Tonga x2 with HBM for the long run.

 

[3DVagabond]

Well, let's see: combine a proper CLC-ready full-card cooling plate, with a full CLC solution, and what do you think it might cost?

For $30 you get a skimpy bracket from NZXT that comes with a sleeve-bearing fan. And then you have to add a CLC cooler of various cost, which leaves you $70 to beat EVGA. Granted, you can, but factor in the difference of the plate, and a fully-warrantied setup, and that's totally worth it IMHO. You figure you'll pay $50-60 on the CLC itself.

I replaced the sleeve-bearing fan with a 92mm Noctua on my NZXT G10/Corsair H75 setup, and I also replaced the two junk stock Corsair fans with EK Vardar F4-120s. Thankfully the 290X Lightning came with an awesome base-plate that remained behind after the stock heatsink was removed, that plate definitely helps keep the other components cooler than what the basic G10 would accomplish, as it is literally just a fan pointing at the components.
(FWIW, the Lightning cooler is awesome, but I only had a 3-slot space between GPUs and as a 3-slot cooler in Crossfire, it was getting choked, big time. Top card got the CLC/AIO cooler, bottom card remains stock - wish I had space to put a CLC on the bottom card too though, that would make my system much more silent during gaming loads)

What you or I pay for something has nothing to do with what EVGA would pay.

 

[Techhog]

Has there actually been any proof of new dies other than Fiji existing? Plenty of proof about Fiji, starting from that 550mm^2 leak last year but I've yet to see anything else that would imply a new die. Anything remotely reliable that is.

Also the stuff about Fiji being a lower end GPU with HBM (4GB) and bermuda being a higher end HBM one with (8GB), the theory that raghu at least bases his speculations on, seems to be most likely false.

http://cdn.videocardz.com/1/2015/04/Hot-Chips-Symposium-Fiji-2.5-HBM.png

^ If AMD had a higher end GPU than fiji and fiji was going to be used for the 380X instead of the 390X they wouldn't be holding a presentation about fiji in august. They'd be doing a bermuda presentation.

And the rest of the lower end leaks for mobile and for desktop have showed reasonable proof for rebrands. Even the leaked XFX double dissipation card had a pcb that was identical to hawaii PCBs from everything we could actually see on the card.

Personally I think that AMD's architecture improvements over tonga are limited to nonexistent and as such it didn't make any sense to design any more GPUs which would be only slight improvements over the old stuff, but expensive to design.

We'll likely see Fiji with almost identical GCN to tonga that gets its extra performance from the power savings from HBM, added die size and added space on the die caused by the change in memory controllers.

This is a stopgap generation anyway. The real deal will be when 16nm rolls out, both manufacturers are on HBM and we'll see which architecture actually is the superior one.

That was the 390 (seriously, are you guys blind?), so either what you said is irrelevant, the 390 is a rebrand, or the leak was fake.

 

[Techhog]

Considering the latest info and the rebrand up to the 370 series. Maybe Fiji is the 380 series and 2xFiji the 390 series.

DX12 looks to improve multi GPU scaling quite well. So its not exactly silly to bet on something like Tonga x2 with HBM for the long run.

No, it's definitely silly for anyone with common sense and the capability to get over extreme bias.

 

[destrekor]

What you or I pay for something has nothing to do with what EVGA would pay.

You can buy it from them direct (which is what we pay), or you can buy it on a card, marked up by the same value (which is what we pay).

A fully-warrantied and whole card mount should basically cost exactly what they are asking. And considering they are selling at retail, the price difference for it already assembled also makes sense.

I highly doubt you'd see such an offering for less.

 

[Cloudfire777]

Thanks to you all regarding the air cooling vs hybrid cooling. Sounds like hybrid may be the way to go. BUT I still have a problem with that I need another exhaust in my chassis for the radiator, and then one for circulation for the whole case. Both are pointed out toward the room.
With air cooling I only need one for the circulation in the chassis, the other fan`s are inside mounted on the CPU/GPU and with a proper case that is isolated, I still wonder if this is more quiet? I understand what you are saying regarding hybrids not going in crazy fanspeed mode when under load while air fan`s does. I need to read about this some more

I remember LinusTechTips silent build video though and he used Rock Pro 3 without fan`s and used the circulation fan on the case to blow over the Rock Pro. Neat

 

[Cloudfire777]

CodeXL 1.7

So, basically, new lineup is... only Fiji as new card?

Similar architecture to Tonga, just bigger. And with HBM.

Yep, that seems accurate. Fiji is still the same GCN 1.3 as Tonga I think, just scaled up. The process seems to be the key here. Still hopeful about 20nm. If it isnt, then its one heck of a 28nm process from GloFo. Almost strange if Nvidia isnt using it if it give so much benefit power wise.
Great find btw

Remember Kyle Bennett said "400 series will be the real new architecture"? Pretty much in line with VR-Zone leaks that said Tonga and Hawaii will be renamed to a pirate theme and released as 300 cards, and the chart above and your screenshot

 

[Head1985]

10 months late with rebrands...

 

[Glo.]

IMO, were looking at 20 nm cards.

There is one problem with that list. There is no Tonga. I don't believe, that we will see no Tonga card in upcoming lineup of GPUs.

Someone asked if waiting for selling out cards making sense if they are rembranded. It makes a lot of sense, if they are made on 20 nm, give more performance than last gen, and using less power at the same time. Cut down Fiji is perfectly in line with GCN core count and first benchmarks on ChipHell, that shown 20% increase in performance, and around 200W power sonsumption.

P.S. IMO we will see full Hawaii with 3072 GCN cores, rather than 2816.

 

[Head1985]

I dont believe it will be on 20nm
Btw full hawaii is already in 290x

 

[Cloudfire777]

OMG I wish I could show you guys this information :/

Lets say there may be one or two more cards with HBM. And holy crap the efficiency! AMD have done some amazing work

Yeah Im signing off before revealing more

 

[BaronVonMehl]

OMG I wish I could show you guys this information :/

Lets say there may be one or two more cards with HBM. And holy crap the efficiency! AMD have done some amazing work

Yeah Im signing off before revealing more

Tommyknockers is going to get you, when this all is just pure imagination

 

[Kenmitch]

OMG I wish I could show you guys this information :/

Lets say there may be one or two more cards with HBM. And holy crap the efficiency! AMD have done some amazing work

Yeah Im signing off before revealing more

AMD pretty much clamped down on the heavy leakers a while ago it looks like. I guess I can see why your alleged source is somewhat worried about spreading the word.

Most of the current rumors are just rumors and a lot of them are most likely spread by the AMD haters in the world. FUD campaign casts a lot of doubt on those waiting for their next GPU purchase.

I have zero love for my GTX 970 and am eagerly waiting to see what AMD brings to the table.

 

[msi2]

IMO, were looking at 20 nm cards.

There is one problem with that list. There is no Tonga. I don't believe, that we will see no Tonga card in upcoming lineup of GPUs.

Someone asked if waiting for selling out cards making sense if they are rembranded. It makes a lot of sense, if they are made on 20 nm, give more performance than last gen, and using less power at the same time. Cut down Fiji is perfectly in line with GCN core count and first benchmarks on ChipHell, that shown 20% increase in performance, and around 200W power sonsumption.

P.S. IMO we will see full Hawaii with 3072 GCN cores, rather than 2816.

Fiji is Tonga... except twice.

 

[Cloudfire777]

AMD pretty much clamped down on the heavy leakers a while ago it looks like. I guess I can see why your alleged source is somewhat worried about spreading the word.

Most of the current rumors are just rumors and a lot of them are most likely spread by the AMD haters in the world. FUD campaign casts a lot of doubt on those waiting for their next GPU purchase.

I have zero love for my GTX 970 and am eagerly waiting to see what AMD brings to the table.

I was told to not show it to anyone. Screw these connections. Whats the purpose of getting hold of information then? I want to share it but can`t. I always post my leaks :/

Very specific information about price, performance, TDP and process of a bunch of R9 300 cards. Lets just say you are gonna laugh when you see the efficiency over top end Kepler cards. I think one with HBM (not 390X/390) is beating Maxwell in efficiency too.

Looking at GTX 960, and the performance over that card, yep, thats better efficiency.

Ok, seriously this will get me in trouble. Ive said way too much. Gone for good now

 

[Head1985]

Who cares about kepler?NV already have here 8 months maxwell...
If 3xx cant beat maxwell it is fail.

 

[JDG1980]

OMG I wish I could show you guys this information :/

Lets say there may be one or two more cards with HBM. And holy crap the efficiency! AMD have done some amazing work

Yeah Im signing off before revealing more

I hope these alleged leaks are right. We need some real competitiveness back in the GPU market.

It's worth pointing out that Lisa Su did publicly state that there are "some very good products in the pipeline" for 2015 - and the context was specifically about discrete graphics. And we also officially know that AMD is going to do something with 20nm in 2H 2015. So it's possible that 20nm GPUs are indeed coming, and that the rumors to the contrary are FUD. On the other hand, AMD has disappointed often enough in the past that it's hard to get our hopes up too high.

 

[JDG1980]

Why do they need to make major improvements in architecture beyond Tonga? Tonga's architecture more than doubles their geometry performance, increases memory bandwidth efficiency 40%, increases pixel fill-rate efficiency 70%. At that point, all they really need to do is grow their die size to increase shaders and textures and use HBM1 to lower power usage. Why is that not a good solution? In Tonga's case, the GPU is basically shader and texture limited and because it only has 32 ROPs, it had no chance of ever beating a 290. On paper it looks like Tonga's 384-bit controller is 20% larger in size than R9 290X's 512-bit because it's the same controller used in Tahiti - highly inefficient old gen one. Most games are mostly shader and texture limited, not tessellation limited. That's why Tonga is such a poorly executed product but I believe it was just a card made specifically to test out the underlying architectural changes R9 390 will have. I still think AMD will make other improvements that we haven't seen in Tonga.

Do we know that Tonga actually has a 384-bit controller? I've heard these rumors, but it seems odd that AMD would put this in and then not use it anywhere. The R9 M295X (full Tonga with 2048 SPs in Retina iMac) still has a 256-bit memory bus, just like the desktop R9 285 and FirePro W7100. If they did so, it seems like Tonga was a very rushed design, and could probably be substantially improved both in terms of die space and power usage with some relatively minor tweaking.

The most compelling argument I've heard about the mystery of Tonga was from a user named Casecutter on a TPU comment thread. He thinks (and I agree) that AMD just purchases Tonga wafers as needed to meet Apple's orders for the R9 M295X. The desktop R9 285 is basically just a dumping ground for trash silicon from these wafers.

Casecutter's hypothesis explains a lot: why there's no fully-enabled Tonga on the desktop, why the R9 285's power efficiency is such a disappointment, and why a product like Tonga was even created by AMD in the first place. If we interpret Tonga as an Apple custom SKU, it makes a lot more sense. (We know for a fact that the Retina iMac has a custom TCON for 5K support; Apple openly said as much in their marketing slides.) Keep in mind that the R9 M295X only uses about 125W maximum, despite having a clock not that much lower than R9 285 (850 MHz for the former, 918 MHz for the latter). The RAM is downclocked only a tiny bit on the mobile part, from 1375 to 1362 MHz. And this is despite the fact that the mobile part is a fully-enabled Tonga with 2048 GCN shaders, compared to the R9 285's 1792 shaders. The FirePro W7100 is also cut down in terms of shaders, but despite having 8GB of VRAM, it has a TDP of only 150W. I suppose this could be achieved by power limit throttling, but it also fits a scenario of top-bin parts going to Apple, then the next-best going in FirePro W7100, and the leakiest trash being reserved for the R9 285.

 

[Silverforce11]

So in this thread, we've gone from.

1. Hot and super power hungry, R290X made bigger, potentially 400W requiring water cooling.

2. It's a dual-GPU, Tonga x 2 (this is the worse rumor, Tonga is trash, multiplying trash by 2 = double trash). Hot, power hungry and slower than R295X2.

3. Only 20% faster than 980..

4. 65% faster than R290X at same power usage.

5. It's super fast and crazy efficient, even more than Maxwell!

What's funny is this supposed launch in June, and this delay is to build volume.. AIBs would have already had cards for a long time yet we have such conflicting leaks. AMD is doing a bang good job of misinformation.

 

[96Firebird]

Sales on Hawaii are getting better and better, fire sale to clear inventory for 300 series?

$200 R9 290

 

[jpiniero]

Do we know that Tonga actually has a 384-bit controller? I've heard these rumors, but it seems odd that AMD would put this in and then not use it anywhere. The R9 M295X (full Tonga with 2048 SPs in Retina iMac) still has a 256-bit memory bus, just like the desktop R9 285 and FirePro W7100. If they did so, it seems like Tonga was a very rushed design, and could probably be substantially improved both in terms of die space and power usage with some relatively minor tweaking.

You have to remember that both AMD and nVidia planned on having 20 nm GPUs by now... both rushed something out.

 

[Cookie Monster]

I was told to not show it to anyone. Screw these connections. Whats the purpose of getting hold of information then? I want to share it but can`t. I always post my leaks :/

Very specific information about price, performance, TDP and process of a bunch of R9 300 cards. Lets just say you are gonna laugh when you see the efficiency over top end Kepler cards. I think one with HBM (not 390X/390) is beating Maxwell in efficiency too.

Looking at GTX 960, and the performance over that card, yep, thats better efficiency.

Ok, seriously this will get me in trouble. Ive said way too much. Gone for good now

So HBM is also available for not just the top end but high/mid end. I think the benefits of lower power consumption due to HBM is really paying off here Its an interesting matchup because even if they are still on GCN1.2 (or an updated 1.3 from Tonga) the power savings alone from the memory side must be huge (and maybe the rumours of 20nm is infact correct(?) adding to more power savings?). Whatever they've done, looks like they've done their own maxwell in terms of power efficiency which is good news.

Not going to tell us performance numbers? My guess is similar performance.. better perf/w due to lower power consumption? Or better performance and similiar power consumption which would be indeed huge for them.

 

[destrekor]

You have to remember that both AMD and nVidia planned on having 20 nm GPUs by now... both rushed something out.

I wouldn't say they rushed anything out.

They planned on having 20nm, but what they ended up doing is having a clear understanding that would not be happening in nearly enough time, so that adapted architectural changes designed for 20nm back into the 28nm process, and it looks like AMD, from an early point, understood it's best bet was to contract with GloFlo for their 28nm SHP (this, at least, seems most likely).

This isn't a stop-gap, hurry up and do something, this was planned, both parties simply adapted and remained on the current processes, and put certain new features they intended for this generation into the development plans for the next generation. And for the next generation, at the timeline they had already envisioned for said next generation, they'll be jumping from 20nm and adopting 16nm/14nm. (that is likely the plan... as we've seen, that is subject to change of course)

 

[ocre]

Unless we accept that all of a sudden, after so many really awesome architectures, AMD GPU division just turned incompetent............
The 300 series is gonna be pretty darn good.

I have always had high hopes for fiji. But i am always afraid to let rumor hype take over. I think it is always better to just wait and see. Pre hype can make a great architecture meh if the bar is set to high.

I have high hopes for AMDs 300 series. It should be really really good. AMD has always had a respectable architecture in my book. I remember the 8800gtx and really starting to worry about AMD. But then the 4870 and 4890 came out. They were really really great designs. Designs that forced nvidia to drop prices over night. I had a gtx 260 in my main rig back then. But my next purchase was the 5850. I got that one straight away before the prices went up.

Every since then, AMD has always had respectable performance. Their 600 seires was very good. The 7000 series was toe to toe. The Ghz won that battle. Nvidia got a leg up with the gk100 but hawaii really was a great accomplishment. As far as chip design, AMD engineers knocked it out of the park. So really dumb decisions were made and it launched in a completely laughable state. Whoever decided that blower and 94c backwards boost.......they needed fired on the spot. Worse was the lack of any other option. All there offered, that stupid reference cooler.

another thing that hit hawaii was the mining craze. I know many people dont agree, but i always believed there was some sort of supply constraint at play as well. It took far too long for supply to level out. These GPUs were selling at about a grand completely keeping hawaii out of the hands of many many gamers.

But hawaii was a great chip. It was able to hang well with the gk110. It took a while for it to shine but it wasnt the chips fault.

So it is hard to believe that all of a sudden, AMD GPU engineers are incompetent. If we look at time frames, Nvidia was able to get a leg up by launching the gk110. AMD eventually responded and hawaii is really a great chip. Its rough take off wasnt the chips fault, the design is really really good. Nvidia was slightly ahead and have been able to capitalize that by launching the gm204 and now the gm200.

But it is really hard for me to accept that AMD, all of a sudden, cant design. History has shown that AMD has been going toe to toe for the many many years now. Fiji should be no different. It should be pretty freaking great.

With the extra time, AMD might surpase the gm200. But even if they dont, there shouldnt be a huge gap. AMD should be able to beat, catch, and/or almost catch maxwell on every metric. History has shown this.

I hate to get too excited before launches and i really do hate pre hype. But i do have high hopes for fiji and little reason to think it wont be impressive.

 

[destrekor]

edit: nevermind, I actually already replied to the same post I just replied. Wow, I'm too young for my memory to be failing! lol