[WCCF] AMD Radeon R9 390X Pictured

[LTC8K6]

What are your guy's opinion on how well the water cooled 390x might handle overclocking?

The water cooled 390X has not been confirmed.

 

[lilltesaito]

Could you guys point me to where you got that confirmation that HBM1 only supports 4GB total addressable space?

It was from a guy that knows a guy in the know that said it was only 4GB. So no one can show you this.

 

[LTC8K6]

It was from a guy that knows a guy in the know that said it was only 4GB. So no one can show you this.

I don't recall anyone saying that the 4gb limit was confirmed, or that the new cards were using HBM1 and not HBM2.

Also, several articles suggested ways of having more than 4gb of VRAM even with a 4gb HBM limit, such as a hybrid setup.

 

[96Firebird]

The latest rumor came from Joel Hruska, an author over at Extreme Tech:

I have shown those slides to a contact in a position to know what AMD is launching this quarter. They have confirmed that Fiji tops out at 4GB, not 8.

Slides are easy to fake, and a huge number of false ones have been making the rounds lately.

His reputation is on the line come June, when the cards finally launch.

Source

 

[twjr]

The latest rumor came from Joel Hruska, an author over at Extreme Tech:



His reputation is on the line come June, when the cards finally launch.

Source

In what was is his reputation on the line? He (Joel) never claimed anything as it all comes from a 'contact in the position to know'. At worst he stands to lose some trust from his source if he is wrong. But by using an unnamed source he can make claims like more or less anyone on this forum.

 

[DrMrLordX]

A few thoughts:

1). Perhaps AMD is expecting that a majority of future games will be DX12 titles (or Vulkan titles) that allow multigpu. One of the features I've heard associated with multigpu is the ability to make full use of all VRAM on all cards . . . in other words, 2x 390x Xfire would provide 8Gb usable VRAM, even if each card only had 4Gb VRAM.

2). If any of you are put off by small cards, then maybe I should dig up my old Ensoniq Soundscape Elite:

Now THAT'S a sound card! If I had ISA slots, I'd still use it . . . eh well, maybe not.

 

[Riceninja]

@RS i mostly agree with everything you've said, but two points i want to reiterate:

1. i firmly believe 4gb will -not- be enough for 1440p 2xmsaa/dsr 4k/modded textures for the games coming in the next 16 months. regular 1080p/1440p for sure, but no one is buying a $500+ card for only that.

2. nvidia will latch onto the 4gb vs 980ti 6gb and drive that talking point hard. amd can't afford to have any sign of "first glance" weakness because nvidia is a well executed marketing machine targeted at the vast majority who aren't cruising these forums daily.

 

[RussianSensation]

What are your guy's opinion on how well the water cooled 390x might handle overclocking?

Welcome to AT!

Assuming Fiji is a large die 500mm2 chip with densely packed transistors and is already clocked at 1.05Ghz (rumors), I don't see it overclocking more than 13-15% on average on air, 15-20% on water.

Hawaii is a larger/more complex chip, made on 28nm node similar to Tahiti. On average, regardless of the cooling system used, overclocking on Hawaii XT is worse than on Tahiti XT.

R9 290X
Air = 1142mhz
Water = 1185mhz
Cascade = 1306 mhz
LN2 = 1427mhz
http://hwbot.org/hardware/videocard/radeon_r9_290x/

HD7970
Air = 1199mhz
Water = 1318mhz
Cascade = 1452mhz
LN2 = 1530mhz
http://hwbot.org/hardware/videocard/radeon_hd_7970/

All things being equal, I would give the ease of overclocking and overclocking potential edge to GM200 based on intuition/best educated guess considering GM200 (Titan X) overclocking is a known quantity - it overclocks very well with minimal voltage increase and scales well.
http://www.anandtech.com/show/9059/the-nvidia-geforce-gtx-titan-x-review/17

That was one of my point points that I think even if R9 390X can tie the TitanX, ultimately, NV can just release a 1250-1300mhz+ GM200 6GB MSI Lightning/EVGA Classified edition card.

His reputation is on the line come June, when the cards finally launch.

Source

No, he actually presented his view in a very professional manner without stating with 100% certainty that he knows Fiji is 4GB. He says based on everything he has heard and knows so far it appears to be but it's not a 100% confirmation. Let's say he is probably 99% confident.

"I have reason to believe that Fiji is a 4GB card, just as I had reason to believe that Skybridge was canceled last week. I am not claiming that it is physically and utterly impossible for anyone to build an HBM product with >4GB of memory, but based on my understanding of where the technology was when AMD decided to bring Fiji to market, that was one of the limitations the company accepted. Since no other GPU today for the mass market carries more than 4GB of RAM, it's a minor concession.

I have communicated, specifically, how you can determine that Fiji is a 4GB product based on publicly available data. That's not "sinking AMD." That's communicating the most accurate information I have access to.

Could AMD surprise launch an 8GB Fiji? I suppose they could. But if they *do* launch an 8GB Fiji I'll be extremely surprised for specific reasons I'm not going to disclose to you."

If Fiji is 4GB, his reputation increases. If Fiji is 8GB, his reputation is not going to be negatively affected imo because he never claimed that 4GB is the only possibility for Fiji XT for the entire generation.

 

[Silverforce11]

Don't assume GloFo's 28nm behave the same as TSMC. All predictions on OC-ability is pointless.

 

[raghu78]

Don't assume GloFo's 28nm behave the same as TSMC. All predictions on OC-ability is pointless.

well said. :thumbsup: GF 28SHP is a custom 28nm process built specifically for AMD's high performance requirements - APUs like Kaveri which need 4 Ghz clocks. This is the highest performing foundry 28nm process without a doubt.

http://www.tomshardware.com/reviews/fx-7600p-kaveri-apu,3842.html

http://www.anandtech.com/show/7974/...hitecture-a10-micro-6700t-performance-preview

"AMD claims a 19% reduction in core leakage/static current for Puma+ compared to Jaguar at 1.2V, and a 38% reduction for the GPU. The drop in leakage directly contributes to a substantially lower power profile for Beema and Mullins."

Beema / Mullins was built at GF 28SHP compared to Kabini / Temash which was built at TSMC 28HP.

A 38% lower GPU leakage is a huge improvement and this will matter when you want to push for higher freq. In terms of transistor density GF 28SHP is also more dense than TSMC 28HP as its a gate first process compared to gate last for TSMC 28nm. Gate first designs have a 10 - 20% higher density and thus lower area. see slide 7

http://globalfoundries.com/docs/def...undries-stmicro-28nm-mobile-apps.pdf?sfvrsn=2

So even without high density packing AMD can cram 8 billion transistors in roughly 520 - 550 sq mm due to a denser GF 28SHP process. Another fact is that the GPU is more uniform with reduced I/O size and complexity as the signals to memory (HBM) will not have to travel on the PCB but on a silicon interposer. Remember that I/O can not be packed as dense as logic. The reduce GPU die complexity was also mentioned in the leaked slides for R9 390X.

http://videocardz.com/55146/amd-radeon-r9-390x-possible-specifications-and-performance-leaked

So again as silverforce said predictions are of no use when there are so many changes happening.

 

[dacostafilipe]

"I have reason to believe that Fiji is a 4GB card, just as I had reason to believe that Skybridge was canceled last week. I am not claiming that it is physically and utterly impossible for anyone to build an HBM product with >4GB of memory, but based on my understanding of where the technology was when AMD decided to bring Fiji to market, that was one of the limitations the company accepted. Since no other GPU today for the mass market carries more than 4GB of RAM, it's a minor concession.

So, no confirmation just a gut feeling?

The only limitation is the I/O pins on the GPU, but with a 500mm2 die there's space for a lot of bumps.

Also, with the relative low frequency inside the stacks (1Ghz) I could see a high-speed ring/hybrid bus happening.

 

[Glo.]

A few thoughts:

1). Perhaps AMD is expecting that a majority of future games will be DX12 titles (or Vulkan titles) that allow multigpu. One of the features I've heard associated with multigpu is the ability to make full use of all VRAM on all cards . . . in other words, 2x 390x Xfire would provide 8Gb usable VRAM, even if each card only had 4Gb VRAM.

That is the functional base of Mantle, that has been used by developers of Vulkan and DX12, and Metal. It also allowes to use system memory for GPU purpose as well.

You should know know what that means . RAM for big Data, HBM for data that requires gigantic bandwith and low latency. Also helpful for systems with unified memory, eg: consoles, CPUs with integrated Graphics, smartphones...

Sorry for off-top. Back to topic, please .

 

[railven]

A few thoughts:

1). Perhaps AMD is expecting that a majority of future games will be DX12 titles (or Vulkan titles) that allow multigpu. One of the features I've heard associated with multigpu is the ability to make full use of all VRAM on all cards . . . in other words, 2x 390x Xfire would provide 8Gb usable VRAM, even if each card only had 4Gb VRAM.

2). If any of you are put off by small cards, then maybe I should dig up my old Ensoniq Soundscape Elite:

Now THAT'S a sound card! If I had ISA slots, I'd still use it . . . eh well, maybe not.

That image. hrnnnnnnggggggg!!

On the card size/looks issue, I remember discussing this with the GF when I said I thought it was ugly. I used the metaphor of muscle cars to sports cars. I prefer the look of classic muscle cars over some of the newer stream line sports cars. It's just aesthetics and ever one has a preference.

 

[therealnickdanger]

So I have to ask... is Hawaii dead? So much talk has surrounded Tonga and Fiji that I've heard very little about Hawaii. If a repurposing of Hawaii included lower power leakage, H.265, and color compression... it could be an incredible $200-250 SKU (like it is now).

 

[Silverforce11]

If you redesign hawaii to include all those new features, it isn't hawaii anymore.

It really has no place in the next lineup, too power hungry. If downclocked, not enough performance to compete against 970/980.

 

[Jacky60]

Even if your 295X2 had 16GB of VRAM, you could never run Crysis 3 with 8xMSAA at 4K at 60 fps.

What you need for that game is GPU processing power, about 3-4X more than an R9 295X2 because even without AA, it can't even reach 40 fps.

Well I can run at an average of 60fps at 4xMSAA as the HARocp review I linked to stated. I wasn't expecting 60fps and never thought that would be possible but was hoping for 30fps not 5fps. It's a massive drop although FXAA actually looks good.
Will post pics when I eventually tidy my desk.

 

[Creig]

I don't understand the "390X needs 8GB of VRAM" crowd. If they would stop and think, they would realize that the 980 SLI systems people bought for 4K monitors only have 4GB. And I don't hear any of those owners complaining about a lack of VRAM holding them back. So a 390X with "only" 4GB of VRAM isn't going to be any worse off.

 

[pj-]

The water cooled 390X has not been confirmed.

No, but it would be awfully odd of AMD to show a render of a short videocard with a radiator looming overhead if they weren't planning on releasing a water cooled card soon.

The slide that image is used in on AMD's website is their GPU roadmap for the year, which would make the lack of a 390x WCE even odder.

 

[exar333]

I don't understand the "390X needs 8GB of VRAM" crowd. If they would stop and think, they would realize that the 980 SLI systems people bought for 4K monitors only have 4GB. And I don't hear any of those owners complaining about a lack of VRAM holding them back. So a 390X with "only" 4GB of VRAM isn't going to be any worse off.

The 980 4GB was release in 2014. It has been somewhat replaced by the Titan X (for those willing to spend the cash), and likely will be this year by a 6GB 980Ti. Will 4GB be enough in 2016? For 1080P yes, but likely will be pushing-it for 1440P and just barely enough for 4K. 16/14nm is a WAYS away, will this be enough to bridge the gap?

If I am spending $500+ on my GPU, why take the chance? If NV releases a 980Ti for $100-150 more, it would be worth the extra cash to get that 50% more 'insurance'.

Keep in mind a lot of sheeple buy based on specs. A '6GB' upgrade from a standard 4GB card on a system builder will impress many. Just saying...

Edit: I would argue the 390x doesn't HAVE to be 8GB, but the option of 8GB/4GB is important for multi-GPU users. Otherwise they will flock to NV offerings with more VRAM.

 

[Glo.]

If you redesign hawaii to include all those new features, it isn't hawaii anymore.

It really has no place in the next lineup, too power hungry. If downclocked, not enough performance to compete against 970/980.

http://forums.anandtech.com/showpost.php?p=37394580&postcount=14

No It doesn't have to be like that.

People shown that GCN is extremely flexible architecture, way more than any other that has been on sale to this day. If you will count that 250W GPU can be sqeezed to 129W(Mac Pro), and 290W can be sqeezed to 145W, while having 85% of performance, that is pretty damn impressive.

 

[exar333]

Don't assume GloFo's 28nm behave the same as TSMC. All predictions on OC-ability is pointless.

Agree wholeheartedly.

Although not exactly the same, we have seen process changes that reduce power consumption and/or leakage, but either provide the same, or even less overclockability. GloFo's 28nm should be a GREAT process for the default voltages and frequencies, but it will come down to the reviews and actual results from production samples to know the ability to push the freq.

Fingers crossed for a great OC on the 390/x line (not hoping it is bad or poor) but I will wait for reviews to know for sure.

 

[x3sphere]

I don't understand the "390X needs 8GB of VRAM" crowd. If they would stop and think, they would realize that the 980 SLI systems people bought for 4K monitors only have 4GB. And I don't hear any of those owners complaining about a lack of VRAM holding them back. So a 390X with "only" 4GB of VRAM isn't going to be any worse off.

Already happened in GTA V. 4xMSAA with everything maxed at 1080P goes over 4GB

 

[Udgnim]

A few thoughts:

1). Perhaps AMD is expecting that a majority of future games will be DX12 titles (or Vulkan titles) that allow multigpu. One of the features I've heard associated with multigpu is the ability to make full use of all VRAM on all cards . . . in other words, 2x 390x Xfire would provide 8Gb usable VRAM, even if each card only had 4Gb VRAM.

2). If any of you are put off by small cards, then maybe I should dig up my old Ensoniq Soundscape Elite:

Now THAT'S a sound card! If I had ISA slots, I'd still use it . . . eh well, maybe not.

I think DX12 / Vulcan's ability to be able to make full use of VRAM is overblown because it requires developer support to implement

I don't believe a significant # of developers will devote resources to coding for multi-gpu in this manner

 

[thetuna]

I think DX12 / Vulcan's ability to be able to make full use of VRAM is overblown because it requires developer support to implement

I don't believe a significant # of developers will devote resources to coding for multi-gpu in this manner

I could see major engines devoting the time.
You don't have to remake the multi-gpu wheel for each game.

 

[LTC8K6]

No, but it would be awfully odd of AMD to show a render of a short videocard with a radiator looming overhead if they weren't planning on releasing a water cooled card soon.

The slide that image is used in on AMD's website is their GPU roadmap for the year, which would make the lack of a 390x WCE even odder.

Oh I saw the pic. But I don't know what card it is.

It could be the 390X, but maybe it isn't?