AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

[kalelovil]

http://videocardz.com/56205/amd-radeon-300-series-roadmap-leaked
If this slide isn't fake, two things to note:
1. 'Fiji' card availability slightly after the 3xx series.
2. No 360X, 370X or 380X cards.

 

[hoom]

Huh, since when did AMD change die names for rebrands?
Respin with minor changes yes but I don't remember that for the same chip

It seems to me like certain people are throwing 'must have more than 4GB or you are trash' lines with minimal actual evidence.
I'd like to see some proper investigation with graphs, quotes from Engine Devs & from AMD/NV tech guys etc.

 

[kalelovil]

Huh, since when did AMD change die names for rebrands?

Pitcairn->Curacao. Same chip, IIRC codename change justified by AMD as new PCB layout allowed a 17% memory upclock.

 

[sebbbi]

DirectX feature level 12.0 mandates tiled resource support (64 KB granularity virtual memory paging for textures and buffers). New consoles support it as well. Tiled resources make it easy for developers to implement fine grained JiT texture and mesh streaming (only visible parts of a texture need to be loaded to memory).

Some last gen console games (including two of our games) already implemented software virtual texturing, allowing practically unlimited amount of texture content inside a single game level, with only a constant memory footprint (less than 100 MB at 720p). 1080p virtual textured game only needs 256 MB of GPU memory for all material textures (assuming four 8bpp DXT material layers) regardless of the texture resolution or texture count/variety. Tiled resources make virtual texturing easy and efficient, meaning that we will see AAA games that do not require obscene amount of GPU memory, even at ultra settings.

As said earlier, DX 12 async copy makes streaming more efficient and reduces the latency. Volume tiled resources (12.1) allow easy implementation of space efficient voxel and distance field data structures (used commonly for global illumination, ambient occlusion and soft shadows).

DX 12 also allows aliasing the same memory region with multiple resources (when you don't need both at the same time). This is a commonly used console memory optimization trick and it is proven to produce big GPU memory savings. Now we can do the same on PC as well.

Just two generations ago NVIDIAs highly praised flagship GeForce GTX 680 was released with 2 GB of graphics memory. Nobody complained. There are still lots of players with these cards (and other one or two generation old 2 GB hardcore gamer cards). It would be a very bad business move to release a game that didn't run properly on these cards at 1080p.

I do however understand that multimonitor gaming and 4K gaming will at some point require a GPU with more than 4 GB of memory. But currently we are talking about a less than 1% niche here.

 

[RecessionCone]

My favorite thing about 4 GB of memory were the endless 970 scandal threads. Clearly, 3.5 GB of memory is fatally too small. But now we know that 4 GB of memory is comfortable and not a problem at all!

 

[silent_guy]

FWIW: I remember a whole bunch of people crying foul when the GTX 680 came out with 1GB less than the 7970. Or the GTX780 with 1GB less than the 290X.

I think the amount of RAM of a GPU is very important not for general game performance, at least medium term, but as a marketing talking point. AMD used to have the upper hand there, now it's Nvidia. It will soon change again with 8GB Hawaii but complicated for Fiji.
Last month, I was in Fry's next go the GPU section. Saw a guy with his GF select a GPU and talking to her about why he chose one over the other: it was entirely based on the amount of GB that was printed on the box. There is large world out there like this and we have no idea.

 

[Kaarlisk]

New My favorite thing about 4 GB of memory were the endless 970 scandal threads. Clearly, 3.5 GB of memory is fatally too small. But now we know that 4 GB of memory is comfortable and not a problem at all!

It's not that 3.5GB is too small. It's that the slow memory region might need game-specific driver fixes and cause slowdowns if there isn't one.

 

[Kaarlisk]

Last month, I was in Fry's next go the GPU section. Saw a guy with his GF select a GPU and talking to her about why he chose one over the other: it was entirely based on the amount of GB that was printed on the box. There is large world out there like this and we have no idea.

I see this often. And the worst part is, I even see this among people spending money hundreds of euros on video cards.

 

[CarstenS]

I've been trying to convince graphics card vendors to put an USB-3-interface on their card so they can have basically any sane amount of memory they want and keep it field upgradeable at the same time. But they wouldn't listen... *SCNR*

 

[pjbliverpool]

One additional aspect for the 4-GiB-topic: Games are largely streaming based nowadays, meaning they do not load a whole level into local memory. Graphics drivers manage this according to the amount of memory present. If they have a larger wiggle room as in Titan X or FirePro W9100, they can afford to leave used assets untouched for a longer time. If that asset is used again they do not have to reload it from main memory, thus saving time and - depending on the capabilities of the engine - achieving smoother rendering. Having non-blocking DMA-engines help in that …

The amount of memory _really_ needed at a certain point in time cannot be measured this simply with tools like GPU-z or MSI Afterburner.

Just two generations ago NVIDIAs highly praised flagship GeForce GTX 680 was released with 2 GB of graphics memory. Nobody complained. There are still lots of players with these cards (and other one or two generation old 2 GB hardcore gamer cards). It would be a very bad business move to release a game that didn't run properly on these cards at 1080p.

I was going to say something along these lines. As a peasant gamer still rocking a 2GB 670 I'd have thought 4GB would be more than enough for a good while. Even though memory tests at various review sites show games "using" more than 2GB I can play those games without any problem on my card at settings which are supposed to require more memory than I have. For example, SoM, Assassins Creed Unity and Far Cry 4 are all no problem at their second from highest texture settings despite apparently using more than 2GB in all cases. In fact if I run RTSS in any of those games I'll always see my memory usage maxed out but that doesn't have an effect on the games smoothness. And I'm talking frame delivery here rather than average frame rate. i.e. I'm not seeing any judder, hitches, pauses, whatever you want to call it that you would usually expect if you're truly running out of VRAM.

Granted I don't use the highest settings but I imagine a 3GB GPU would handle those just fine and 4GB wouldn't even break a sweat.

All that said though, I'd still prefer an 8GB GPU over a 4GB one even if I don't think it would have much practical benefit in the near to medium term. But then I like big numbers.

 

[CaptainGinger]

Maybe you should make a petition CarstenS, ToTTenTranz might sign it.

 

[silent_guy]

All that said though, I'd still prefer an 8GB GPU over a 4GB one even if I don't think it would have much practical benefit in the near to medium term. But then I like big numbers.

Exactly. And that 12GB on a Titan X is just so deliciously decadent.

 

[Ryan Smith]

Unless AMD are using it to differentiate between the 3xx and 2xx cards, at least the rebranded GCN1.0 cards that don't get a mention, it very strongly hints to the new gpus at least having the hardware scaler like the other GCN1.1 and above gpus.

More likely it just means AMD has finally gotten around to implementing shader-based VSR.

If 4GB isn't enough in those titles how is a 295x2 able to beat cards with much more memory?

http://www.anandtech.com/show/9306/the-nvidia-geforce-gtx-980-ti-review/6

Funny enough, if you check our GTA 5 results, the 295X2 takes a massive hit in minimum framerates at our highest settings. Granted, it's 4K and nothing can average 30fps, but the point being that it's possible to trip up 4GB cards if your tests are designed to do so (mine haven't specifically been designed for this). 3GB cards on GTA also get hammered at 4K, and 2GB cards get hammered at 1440p.

 

[3dilettante]

I've been trying to convince graphics card vendors to put an USB-3-interface on their card so they can have basically any sane amount of memory they want and keep it field upgradeable at the same time. But they wouldn't listen... *SCNR*

With the latest standard, could you insert the card both ways?

All that said though, I'd still prefer an 8GB GPU over a 4GB one even if I don't think it would have much practical benefit in the near to medium term. But then I like big numbers.

It depends on where the product slots into the market, but with so many signs pointing to such a lofty range, "practical" tends to be a shrinking fraction of what buyers are going by, even for a luxury item like a gaming card.
The gains are incremental, but if you want to play in the steeper portion of the pricing exponential curve, expect the buyers to be as reasonable as the price is.

 

[Deleted member 13524]

Funny enough, if you check our GTA 5 results, the 295X2 takes a massive hit in minimum framerates at our highest settings. Granted, it's 4K and nothing can average 30fps, but the point being that it's possible to trip up 4GB cards if your tests are designed to do so (mine haven't specifically been designed for this). 3GB cards on GTA also get hammered at 4K, and 2GB cards get hammered at 1440p.

Thank you for your sanity, honestly.

The gains are incremental, but if you want to play in the steeper portion of the pricing exponential curve, expect the buyers to be as reasonable as the price is.

I also think that longevity plays a really big part in halo products, and that's what the 4GB puts in question for Fiji.
Did you notice the reaction towards nVidia on reddit, tech forums and chans when the users saw their 1st-gen Titans and 780/Ti getting a performance close to the GTX 960 in Witcher 3 with Hairworks enabled?
I for one saw hundreds of posts of GK110 owners swearing not to buy a nVidia card again.
Yeah, maybe most of those guys don't even own a GK110 card. Maybe most of them will buy a nVidia card again, regardless of what the competition has in store.

But having your supposed golden customers, the ones who paid $1000 for your halo product with huge margins, badmouthing the company is definitely a position that no PR division wants to be in.

 

[CaptainGinger]

Funny enough, if you check our GTA 5 results, the 295X2 takes a massive hit in minimum framerates at our highest settings. Granted, it's 4K and nothing can average 30fps, but the point being that it's possible to trip up 4GB cards if your tests are designed to do so (mine haven't specifically been designed for this). 3GB cards on GTA also get hammered at 4K, and 2GB cards get hammered at 1440p.

How would you separate the effect of limited VRAM from general frame pacing issues? The single chip 4GB cards seem fine.

 

[Deleted member 13524]

Reading the article (or at least the aforementioned page) would've been convenient:

At 4K very high quality, 4GB cards have just enough VRAM to stay alive, with the multi-GPU R9 295X2 getting crushed due to the additional VRAM requirements of AFR pushing it over the edge. Not plotted here are the 3GB cards, which saw their framerates plummet to the low single-digits, essentially struggling to complete this benchmark. Meanwhile 1440p at high quality crushes our 2GB cards, with less VRAM than a Radeon HD 7970 falling off of the cliff.

 

[3dilettante]

I also think that longevity plays a really big part in halo products, and that's what the 4GB puts in question for Fiji.

There's a certain expectation that these products would be running at the top for at least a while.
There are currently certain targeted cases where 4GB could be a hinderance, and the rare-case argument doesn't fly when a product is targeting an uncommon buyer, where there's not much room at that end of the bell curve for average requirements.

Did you notice the reaction towards nVidia on reddit, tech forums and chans when the users saw their 1st-gen Titans and 780/Ti getting a performance close to the GTX 960 in Witcher 3 with Hairworks enabled?

That would be the case of expectations of being at the top for a reasonable time being part of the purchase of a top-priced product. People are paying serious-business money for an impractical pursuit, with an ephemeral status or benefit, which leaves few motivations that aren't close to some really basic emotional drives.

I for one saw hundreds of posts of GK110 owners swearing not to buy a nVidia card again.

If there happened to be a credible alternative. Some no doubt were burned by the other alternative's multi-GPU implementation that lost half their frame gains.

 

[CaptainGinger]

Reading the article (or at least the aforementioned page) would've been convenient:

Some useful info congratulations. So Ryan is pretty sure the single 4GB cards are OK and and it's the extra VRAM required by the Crossfire solution that makes the 295x2 tank.

I'll also point out that under this demanding test even the boutique 12GB card cannot manage 30 FPS average.

 

[Albuquerque]

I own a 3GB 7970Ghz model right now, and I run out of video memory capacity before I run out of "oomph" while playing GTAV at 1440p. And while I understand that a properly implemented DX12 game can use the newly extant technology to "page" items into VRAM from main memory more efficiently and effectively, that future is still quite far out.

Insofar as I'm concerned, 4GB should be the minimum to consider. If I'm paying for a top-end card, you're going to be hard pressed to tell me that I "only" need 4GB. The first thing I'll do with a new video card is load up every demanding game I own, including GTAV with every slider moved as far right as it will allow.

If the limitation of 4GB VRAM kicks me in the nuts, I'll be pissed and send it back and be far less likely to take that manufacturer's word for it at any point in the future. AMD better make damned sure this isn't an issue, otherwise the gaming fanbois will make it the Kilimanjaro of mole hills...