Possible Radeon 390X / 390 and 380X Spec / Benchmark (do not hotlink images!!!!!!)

[ubersonic]

Maybe because it's stacked 4GB, the 4GB actually means it's 4GB squared, so equal to 16GB of GDDR5! LMAO.

 

[Silent_Scone]

Because they work differently?

They most certainly likely do at certain sub levels. But buffer management is very much controlled via middleware ergo DirectX. The same amount of memory will be going in and out to a greater part. I am very eager to hear the spin on why 4GB HBM is now apparently ample.

 

[ubersonic]

I think the issue is everyone is thinking that as you need >4GB GDDR5 for 4K you will need >4GB HBM. This is probably not the case when you take into account the width of the i/o bus and latency between GPU and memory being much improved on HBM vs the older GDDR5

It doesn't matter how quick you are at filling a pint glass, it doesn't matter how wide the mouth of the glass is, you're still not getting >1 pint into it...

 

[D.P.]

I think the main problem that caused the 'raging' apart from misleading spec, less ROPs, weird mem config etc was stuttering once games reach that 3.5GB and had to go use the slower memory. There's tons of videos / user feedback on the internetz.

The thing is it would be trivial to limit the amount of emory in firmware to 3.5GB and ignore the slightly slower part. The end result would be even worse because the data would have to be pulled from the far slower system memory rather than the far faster 0.5GB of slower GPU memory.

 

[drunkenmaster]

Memory isn't a concern, people like to have more than they need but don't care if that option wasn't available. If no cards had more than 4GB no one would care that 1 game in a thousand might need more than 4GB, the second a 8GB card became available those who must have the latest and greatest thing suddenly saw 4GB as last year even though no games actively changed over night just because more memory became available.

There is likely to be a 8GB option, at launch or not too long after, I don't care. I'll buy a 4GB version and save the money(if I buy one) because I buy based on value rather than e-peen or the inability to see beyond consumerist hype that fanboys get into.

I've got a 1080p screen, I'm looking to get a good 1440p screen, 4k at 60hz doesn't interest me in the slightest. If 999/1000 games don't run faster at 4k with 8GB over 4GB, how many are a problem at 1440p? Will I eventually get a 4k/144hz screen, sure, but they won't be available at a reasonable price and in reasonable quality for another 2 years so why should the card I buy this month(potentially) matter, I won't have that potentially new card in two years so who the hell cares.

More to the point, I'd prefer to spend less money on what I actually need, rather than spend obscenely more money on something I won't ever use. Today I'd prefer to buy a 4GB at a good value price than a 12GB card with a horrific mark up just to be able to say I have more memory yet in no way ever come close to using the majority of it.

As for the cards themselves, the only card that has ever been interesting is the one that offers more performance than the last gen. The 5770 that matched the 4870 wasn't interesting to most people, the 5870 however was, etc, etc. I care about Fury, I'd hope AMD didn't do rebrands but I can understand why.

Normally speaking you haven't been waiting on a new process for 4 years. Spending big on a top to bottom architecture shift and tape out at the very last gasp of an old process just before the new process is simply not very financially sensible. Spend several hundred million on an entire new range of cores which will be obsolete in 8-12 months when 16nm cards are out which will ALSO cost several hundred million or just do the one new core with higher performance and save the money and invest the time in making the next generation better.

I can see reasons to do both but ultimately a top to bottom refresh this close to the next process doesn't make a whole lot of sense. 18 months ago, sure, today, not so much. The return on investment is different. 16nm products due in 9-12 months then you get 9 months of sales to profit from while you've spent millions in R&D and taping out. Spending the same money for the first 16nm cards which can have more engineering time spent on them instead and they'll be around for 18-24 months more or less with a higher return on investment.

As with the 4770 before it, Tonga to a lesser degree, bringing a new card to test out new technologies before implementing it more widely is a sound development move.

 

[D.P.]

I think the issue is everyone is thinking that as you need >4GB GDDR5 for 4K you will need >4GB HBM. This is probably not the case when you take into account the width of the i/o bus and latency between GPU and memory being much improved on HBM vs the older GDDR5

Irrelevant 4GB or texture, geometry and buffer space is 4GB, however fast you move that data about.

 

[Silent_Scone]

So you're not going 4K so it's not a problem for anyone, righto got that one nailed quickly.

Priceless. Genuinely.


1440p VRAM limitations

Lords of the Fallen

Dying Light

Shadow of Mordor (is a problem with 4GB in tandem with the Ultra texture pack if installed correctly)

Assassins Creed Unity (any form of AA above FXAA will incur stutter)

Wolfenstein New Order



Those are the ones I've played and can name off the top of my head. Not sure how many games it will take before people stop flagging something as poorly optimised.

 

[Mtom]

Someone told me the Gaming day will be more about software stuff than hardware...just like the big mantle presentation when we waited for the 290X....
If true the news saying 3XX on 18th - Fury on 24th will be right...but hopefully we will have some info on the cards..or a ninja release.

 

[Kaapstad]

snip

How I choose to game is my choice

How you choose to game is your choice

What we both need to remember is it would be wrong to try and expect or impose a hardware setup on someone else. It is up to people how they game.

The one thing that is true is that 4gb of memory is not enough for all setups and resolutions, once you accept this there is no point in arguing about the ifs and buts.

 

[shankly1985]

Not long now,, amd are hinting at some big surprises.. Am ready to be blown away..

 

[SimBy]

Not long now,, amd are hinting at some big surprises.. Am ready to be blown away..

Spoiler

 

[Apache14]

It doesn't matter how quick you are at filling a pint glass, it doesn't matter how wide the mouth of the glass is, you're still not getting >1 pint into it...

If you where filling the glass at the same rate as you where drinking, you would have infinite beer :|

Irrelevant 4GB or texture, geometry and buffer space is 4GB, however fast you move that data about.

Thats really not true at all, take the 970 and its 3.5GB "full speed" issue.

If you where using <=3.5GB of vram the gameplay was buttery smooth, then when you when above that mark things slowed down. That wasn't to do with not having enough vram that was indeed due to how fast you can fill / empty it.

Now if you can fill / empty the buffer quick enough when you need to swap in / out the data needed after exceeding the maximum vram you have you could still have a high FPS

 

[Ayahuasca]

Not long now,, amd are hinting at some big surprises.. Am ready to be blown away..

The hype is real.

I'm hoping Advanced Malfunctioning Devices can blow the nVidiots away (including myself). Another noticeable performance leap above the TX/980ti will be awesome.

 

[Orangey]

Has anyone dumped the BIOS from the Grenada cards?

 

[Brumboy]

Any idea what time the Fury reveal will be?

 

[Silent_Scone]

If you where filling the glass at the same rate as you where drinking, you would have infinite beer :|



Thats really not true at all, take the 970 and its 3.5GB "full speed" issue.

If you where using <=3.5GB of vram the gameplay was buttery smooth, then when you when above that mark things slowed down. That wasn't to do with not having enough vram that was indeed due to how fast you can fill / empty it.

Now if you can fill / empty the buffer quick enough when you need to swap in / out the data needed after exceeding the maximum vram you have you could still have a high FPS

That's because the 970 had a performance penalty by design by the 2 way l2 cache. The bandwidth was effectively slower than a great deal of things, and in no way does that have any bearing on memory consumption. Think you're a little confused.

A performance penalty from bandwidth alone is different from one that is incurred from transition buffering.

 

[mmj_uk]

I think the issue is everyone is thinking that as you need >4GB GDDR5 for 4K you will need >4GB HBM.

HBM isn't magical, 4GB is 4GB, what you are suggesting is like claiming that a 128GB SSD can store more than a 128GB mechanical HD just because it's different.

*Maybe* they will have a new texture compression method but you will likely get 10% saving at most, if it was possbile to squash down textures 50%-100% without impacting image quality it would have been done a long time ago when memory was at a premium.

This is probably not the case when you take into account the width of the i/o bus and latency between GPU and memory being much improved on HBM vs the older GDDR5

I'm sorry to say but this just shows that you don't know what you're talking about, the reason performance plummets when you run out of local memory (VRAM) is because the GPU has to fetch data from system memory, to do that it has to go over the PCI-E bus which is where the bottleneck comes in.

PCI-E 3.0 16x bus = 16GB/s
PCI-E 2.0 16x bus = 8GB/s
System memory = 30GB/s-60GB/s in todays systems
Fury HBM = 512GB/s

It is FAR faster to store everything in local memory (VRAM) and this is the reason why NVidia felt having that extra 0.5GB on GTX970 was a good idea, as handicapped as that 0.5GB was is by VRAM standards it's still miles better than having to go to system memory.

 

[ubersonic]

Not long now,, amd are hinting at some big surprises.. Am ready to be blown away..

Spoiler

Sorry for the obvious paint, can't find my PSP CD to shop it properly

 

[nashathedog]

If you where filling the glass at the same rate as you where drinking, you would have infinite beer :|



Thats really not true at all, take the 970 and its 3.5GB "full speed" issue.

If you where using <=3.5GB of vram the gameplay was buttery smooth, then when you when above that mark things slowed down. That wasn't to do with not having enough vram that was indeed due to how fast you can fill / empty it.

Now if you can fill / empty the buffer quick enough when you need to swap in / out the data needed after exceeding the maximum vram you have you could still have a high FPS

That scenario does not compare to the HBM bandwidth one.

 

[Boomstick777]

Can you explain this? as far as I know 1MB is 1MB in size regardless of whether it's FPM, EDO, SDR, DDR or Rambus...

Faster 4GB is still 4GB.

I've never used HBM or even seen it action. I'm holding judgement until we actually see the cards. HBM and GDDR5 are not the same things, they may not work in the same way. I will wait until we get more info before deciding if it's a good or bad card. I had fun with my 4GB GDDR5 980, will probably enjoy using AMD's 4GB HBM Fury X to.

1 day to go until we find out. Let's try and hold it together guys.