[WCCF] AMD Radeon R9 390X Pictured

[EightySix Four]

So my follow up question is: Since I'm not engineer and don't understand that much the technical shit about AA and such, but since AMD is bringing so much more bandwidth, is there possibility that it can be now utilize to make much more efficient AA? Kinda like what microsoft tried with the eDRAM on xbox360 and when people were saying that in theory they can bring free AA. But in the end the eDRAM was big enough or something like that.

AMD's argument is that due to the way memory bandwidth scaled with the amount of memory, until now there was no reason to optimize the amount of memory in use, because there was always more memory than necessary. Now, with HBM, there is a ton of bandwidth to go around but memory will be at a higher premium. This means the driver developers can focus on techniques to minimize memory usage (and accelerate moving things in and out of that high bandwidth memory) instead of it not being a focus.

They claim that this means 4gb will be enough and no one here (or anywhere except AMD, NVidia, or Intel's GPU departments) could probably successfully argue anything different until we see the results of the optimizations they are claiming exist.

Think of it like horsepower in the 60s and 70s, tons of power and it used tons of fuel. In the 80s the fuel supply went down and the power was curtailed back while efficiency was prioritized. Now we often see cars with better fuel economy than the 80s AND more power than they had in the 60s and 70s. AMD is claiming they can skip the middle part and go straight to the more efficient now.

 

[Creig]

If AMD pulled off an instant PCB swap for their partners, then a $400 launch price seems very possible/likely for a 8GB card with a respun Hawaii 'XTX' chip.

I agree. AIBs could possibly just tweak existing 290X PCBs and they're ready to go. It might even be possible to reuse the same 290X coolers if they can use higher density memory with the same number of chips. No retooling necessary.

 

[Udgnim]

can definitely see 4GB with multi GPU being a concern

for single GPU, we'll see whether GPU power or VRAM buffer becomes an issue first

 

[Despoiler]

This might be for gaming, but I'm fairly certain it won't apply to scientific/professional workloads.

If AMD is ignoring these HIGH value/margin markets, by having 4GB max with a new generation near monopoly in DP compute, then they are fools.

I don't believe they are fools.

I don't think they would be announcing HPC cards at E3. We are speaking to gaming cards for the moment.

 

[tential]

can definitely see 4GB with multi GPU being a concern

for single GPU, we'll see whether GPU power or VRAM buffer becomes an issue first

I have avoided saying anything about 4gb hbm being a limitation because no one in the entire world except for amd engineers / Product testers know at this point. Drivers and bios probably aren't finalized either so no one knows. Am I worried? Yes. But will I call it a limitation? Not until I see crossfire benchmarks. I don't care about specs. It could be 2gb hbm. If it performs, doesn't drop frames, doesn't stutter, etc. I don't care.

 

[Despoiler]

One thing is for certain, AMD marketing needs to do one heck of a job to combat the 12 > 6 > 4 crowd out there. The low information forums are chock full of the bigger numbers are better crowd. They are already blindly making up their minds and don't seem to care that analyzing benchmarks is the only way to come to a rational conclusion.

 

[swilli89]

One thing is for certain, AMD marketing needs to do one heck of a job to combat the 12 > 6 > 4 crowd out there. The low information forums are chock full of the bigger numbers are better crowd. They are already blindly making up their minds and don't seem to care that analyzing benchmarks is the only way to come to a rational conclusion.

http://forums.anandtech.com/showthread.php?t=1804008

We know where a lot of "that crowd" come from. There were viral posters a decade ago on these very forums. Luckily the mods did an excellent job routing them out.

http://forums.guru3d.com/showthread.php?t=357822

Once you see something in an industry where a company is exposed, and there is no legislative or punitive damages, you should bet the farm that the activity continues and even increases.

Ignore the haters! I have a feeling they have incredible hardware but that their drivers just aren't ready.

 

[96Firebird]

So you think the people concerned about a top-teir GPU only having 4GB of VRAM are viral posters?

 

[Stuka87]

Maybe then, but now games are starting to use more than 4GB of VRAM, especially at 4k. I guess we'll see in a couple of weeks.

Outside of things AMD can do on the driver side, this comes down to the game developer.

If we look at the Witcher 3, Ultra 1080P only uses 1.2-1.5GB of RAM. Even 4K gets nowhere close to 4GB. While another game like GTA 5 uses a crap load of VRAM. And its not that GTA 5 looks significantly better, its that the game is far less optimized for memory usage.

But even in these cases there may be things the GPU maker can do at the driver level to address this.

We just need to wait and see.

 

[tential]

So you think the people concerned about a top-teir GPU only having 4GB of VRAM are viral posters?

Nope. He's just saying that nvidia has used shady practices before to discredit their competitors. Another reason I just love nvidia from a business standpoint. Business wise it's clear nvidia is far better run than amd. From a consumer perspective though, I prefer amd since they're either too stupid to market well or firmly believe people actually will judge their products based on merit LOL.

I don't think they need viral shills. Their fanboys will do what their viral shills already would do for the memory. But I wouldn't put it past them to leak fiji being worse than the 980ti. Nvidia will stop at almost nothing to make more money. I wish they had offices near me and hired me to work there haha.

 

[swilli89]

So you think the people concerned about a top-teir GPU only having 4GB of VRAM are viral posters?

No it was rather tongue in cheek. I myself have a concern about the vram capacity so I by no means was discounting this concern as not realistic.

 

[jamesgalb]

So you think the people concerned about a top-teir GPU only having 4GB of VRAM are viral posters?

I would say that anyone more than 'concerned' would seem odd... Some people seem to have gone the way of "Rumors confirm it, NVidia won, go buy your 980Tis now before they sell out!", and that's just weird to me.

gddr5 vs ddr3 was not that long ago, so it would take a rather challenged 'enthusiast' to make definitive claims based on more gddr5 vs less hbm.

If there were ever a time to fire up the 'peer influence' brigade, it would be when you tossed out a product right before the oppositions flagship and reviews hit the world.

 

[SPBHM]

can definitely see 4GB with multi GPU being a concern

for single GPU, we'll see whether GPU power or VRAM buffer becomes an issue first

maybe, but I think the 680 also looked to have enough VRAM when it was new, but it proved to be a disadvantage to the 7970 later

if you look at how the amount of RAM progressed during the last 5 years, you would expect more than 4GB for the new card.

AMD
5870 1GB
6970 2GB
7970 3GB
290X 4GB

specially if Fiji is not that special in performance or price compared to the 980 Ti the 4GB will probably be a significant disadvantage when trying to compete for sales

 

[jackstar7]

Also, if DX12 does flip things on their head for Crossfire/SLI memory sharing, then things will change drastically and suddenly 4GB per GPU will potentially be plenty up through 4K resolutions.

 

[swilli89]

maybe, but I think the 680 also looked to have enough VRAM when it was new, but it proved to be a disadvantage to the 7970 later

if you look at how the amount of RAM progressed during the last 5 years, you would expect more than 4GB for the new card.

AMD
5870 1GB
6970 2GB
7970 3GB
290X 4GB

specially if Fiji is not that special in performance or price compared to the 980 Ti the 4GB will probably be a significant disadvantage when trying to compete for sales

But remember the AMD engineer earlier that was quoted as saying they were only increasing capacity as a side result of increasing the memory controller width to increase bandwidth. Of course AMD knew spec-racing fanboys would like to brag about their VRAM size to girls at the bar but it wasn't a design choice to first and foremost increase the capacity, merely a consequence.

 

[maddie]

Then they'll be saving any high memory capacity chips strictly for the HPC market

Only if they are supply constrained. Even then it would make sense to have the early reviews test an 8GB card to let us know the potential.

In fact, if 4GB can be made to run as well as 8GB in gaming usage, they should absolutely have 8GB in early testing, if only to show no/little difference.

At the very least we will know 8GB exists and will come.

Remove ALL naysayers from having anything to argue.

 

[sontin]

He is saying this because they can only offer 4GB. What do you expect them to do otherwise? :|

They are selling 3GB and 4GB cards right now. So why havent they started much earlier to optimize their software stack for better memory utilization?!

Simple: Marketing.

 

[jamesgalb]

Also, if DX12 does flip things on their head for Crossfire/SLI memory sharing, then things will change drastically and suddenly 4GB per GPU will potentially be plenty up through 4K resolutions.

I'm most interested in iGPs there. They could take over AA and texture rendering and potentially a lot of the memory needs.

This has my saving to upgrade to Skylake with an Iris Pro and DDR4

 

[swilli89]

He is saying this because they can only offer 4GB. What do you expect them to do otherwise? :|

They are selling 3GB and 4GB cards right now. So why havent they started much earlier to optimize their software stack for better memory utilization?!

Simple: Marketing.

But that's not what they did. They probably were aware of capacity constraints well before we even knew Fiji would feature HBM. They developed Tonga to do more with less.

AMD 100% for sure started to optimize how they use VRAM capacity the moment they knew they wouldn't just be using 8 Gigs of GDDR5. Just because the engineer was just now asked about it close to release doesn't mean his statement coincides with their behind the scenes work.

 

[thilanliyan]

http://forums.anandtech.com/showthread.php?t=1804008

We know where a lot of "that crowd" come from. There were viral posters a decade ago on these very forums. Luckily the mods did an excellent job routing them out.

Ahhh...the good old days!!
It's pretty tame here now!

 

[maddie]

He is saying this because they can only offer 4GB. What do you expect them to do otherwise? :|

They are selling 3GB and 4GB cards right now. So why havent they started much earlier to optimize their software stack for better memory utilization?!

Simple: Marketing.

So tell us again how you know this as a FACT?

 

[Despoiler]

He is saying this because they can only offer 4GB. What do you expect them to do otherwise? :|

They are selling 3GB and 4GB cards right now. So why havent they started much earlier to optimize their software stack for better memory utilization?!

Simple: Marketing.

Or not

http://arstechnica.co.uk/informatio...hbm-why-amds-high-bandwidth-memory-matters/2/

AMD's CTO, Joe Macri

You're not limited in this world to any number of stacks, but from a capacity point of view, this generation-one HBM, each DRAM is a two-gigabit DRAM, so yeah, if you have four stacks you're limited to four gigabytes. You could build things with more stacks, you could build things with less stacks. Capacity of the frame buffer is just one of our concerns. There are many things you can do to utilise that capacity better. So if you have four stacks you're limited to four [gigabytes], but we don't really view that as a performance limitation from an AMD perspective

If you actually look at frame buffers and how efficient they are and how efficient the drivers are at managing capacities across the resolutions, you'll find that there's a lot that can be done. We do not see 4GB as a limitation that would cause performance bottlenecks. We just need to do a better job managing the capacities. We were getting free capacity, because with [GDDR5] in order to get more bandwidth we needed to make the memory system wider, so the capacities were increasing. As engineers, we always focus on where the bottleneck is. If you're getting capacity, you don't put as much effort into better utilising that capacity. 4GB is more than sufficient. We've had to go do a little bit of investment in order to better utilise the frame buffer, but we're not really seeing a frame buffer capacity [problem]. You'll be blown away by how much [capacity] is wasted.

 

[sontin]

From Hardware.fr:
http://www.hardware.fr/news/14241/computex-amd-fiji-ses-4-go-hbm-photo.html

Fiji is limited to 4GB. They didnt care for more with HBM1.

But that's not what they did. They probably were aware of capacity constraints well before we even knew Fiji would feature HBM. They developed Tonga to do more with less.

What has Tonga do with AMD's statements about the memory size?
Tonga uses a color compression from the memory chips to the GPU. It doesnt help to optimize the amount of data.

AMD 100% for sure started to optimize how they use VRAM capacity the moment they knew they wouldn't just be using 8 Gigs of GDDR5. Just because the engineer was just now asked about it close to release doesn't mean his statement coincides with their behind the scenes work.

They have been selling cards with less or the same amount of memory since years. They sold a $1500 graphics card with 4GB last year and didnt bother to look after the memory utilization... It's marketing. The competition sells card with more memory.

 

[Enigmoid]

Also, if DX12 does flip things on their head for Crossfire/SLI memory sharing, then things will change drastically and suddenly 4GB per GPU will potentially be plenty up through 4K resolutions.

Doesn't change a thing as it has to go through the vastly slower PCIe interface.

He is saying this because they can only offer 4GB. What do you expect them to do otherwise? :|

They are selling 3GB and 4GB cards right now. So why havent they started much earlier to optimize their software stack for better memory utilization?!

Simple: Marketing.

This.

 

[thilanliyan]

They are selling 3GB and 4GB cards right now. So why havent they started much earlier to optimize their software stack for better memory utilization?!

They said it was cheaper to just slap more memory on, rather than optimize.