• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

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

Status
Not open for further replies.
Kaapstad said:
There won't be any pooling of memory with DX12 or anything else for a number of years to come as PC hardware is nowhere near up to the speed required across the PCI-E slots.

Even AMD are hinting that this is true as the 390X comes with 8gb which would be unneeded with pooled memory.
Click to expand...

I think that's categorically untrue given what Huddy and others have said re: DX12 / Vulkan.

Also, 8GB on 390/X has nothing to do with pooled memory being implemented (or not).
 
pmc25 said:
It wasn't implemented into the Mantle drivers.

Oh, it won't effect older games? You mean the ones that never need that much memory anyway? Oh, right.

Yes we do know.
Click to expand...

Older games have already been found to of passed the 4gb mark guess they did need that much memory. :rolleyes:

So was a feature of Mantle and never implemented in drivers, Is it implemented in DX12 drivers yet? or just a feature as yet not used like mantle did ?
 
pmc25 said:
Why did you bother writing any of this? 1) If you're going CF, all DX12 / Vulkan games should pool memory. Therefore 2x Fury = 8GB. 2) HBM can swap larger files and more of them than GDDR5, and at lower latency, and DX12 / Vulkan can do the same, plus tiled resources. When the option is to go 4GB HBM or GDDR5 (and much slower GPU and less bandwidth), the answer is pretty damned obvious ... plus they make their transition to interposer, new memory, and new memory controller on a familiar and reliable process - they don't have THAT much to do for Arctic Islands ... NVIDIA have their work cut out with everything new on an unfamiliar process, and coming much later to the stacked memory (and certainly HBM) game.

Add to this GCN is far better at parallelisation than Maxwell or Kepler, and Vulkan & DX12 are all about parallelisation, and HBM and increased bandwidth suit parallelisation far more than a larger frame buffer ...... it's a total no brainer.
Click to expand...

Tiled/resource streaming is about the only thing that will give 4GB any breathing space in future titles on high res/multi display setups, no matter how fast HBM is when it comes to swapping its always going to be strangled by the slowest link in the chain.
 
pmc25 said:
If you mean 2xx and 7xxx series not supporting it, quite possibly. However it has nothing to do with PCI-E.
Click to expand...

PCI-E 3.0 is barely fast enough for high end cards now yet alone having to worry about pooled memory.

There are technical articles on the net that explain why PC hardware is not fast enough for pooled memory and I someone will probably post a link in a bit. Unfortunately I can not do so at the moment because of the PC I am using.
 
Kaapstad said:
PCI-E 3.0 is barely fast enough for high end cards now yet alone having to worry about pooled memory.

There are technical articles on the net that explain why PC hardware is not fast enough for pooled memory and I someone will probably post a link in a bit. Unfortunately I can not do so at the moment because of the PC I am using.
Click to expand...

;)

According to this even PCI-e 4 isn't going to be fast enough

"Coming to the final pillar then, we have a brand new feature being introduced for Pascal: NVLink.
NVLink, in a nutshell, is NVIDIA’s effort to supplant PCI-Express with a faster interconnect bus.
From the perspective of NVIDIA, who is looking at what it would take to allow compute workloads to better scale across multiple GPUs,
the 16GB/sec made available by PCI-Express 3.0 is hardly adequate.
Especially when compared to the 250GB/sec+ of memory bandwidth available within a single card.
PCIe 4.0 in turn will eventually bring higher bandwidth yet,
but this still is not enough. As such NVIDIA is pursuing their own bus to achieve the kind of bandwidth they desire."

source
 
I thought the HBM memory on these were stacked ? Could that not mean if you can see 4 modules there could be 8 so 8gb would be possible ?

I'm sure I have seen a pick of an 8GB HBM module showing 4 modules dual stacked....

Or something like that....
 
MOOGLEYS said:
I thought the HBM memory on these were stacked ? Could that not mean if you can see 4 modules there could be 8 so 8gb would be possible ?

I'm sure I have seen a pick of an 8GB HBM module showing 4 modules dual stacked....

Or something like that....
Click to expand...

Yeah I thought the same, the limit for HBM1 being 2GB per stack, i.e 4 x stacks would mean 8GB, but this 4GB rumour has persisted to the point where everybody is fairly certain these first chips are 4GB only.
 
i'm watching DIG DADDY on youtube it's billiant, so funny

it's miles better with a TITAN X, i'm so glad i brought it ..............LOL
 
Last edited:
bru said:
Yes but 4GB of textures is still 4GB of textures.

Or put it this way, a pint of water is a pint of water, whether it is in a bottle or a bucket, or even a super high pressure hose system, you may be able to empty some Quicker than others but if all three only hold 1 pint then you cannot squeeze more than that into it. all three will still be filled by the same tap as well, in this case the HDD/system ram.
Click to expand...

Did people using GDDR5 1GB cards suddenly have issues running games at 1080p 7 years ago over those with GDDR3 2GB cards, no they didn't. AMD have also added a new compression system for the HBM cards so could have a much higher effective VRAM limit than just 4GB. Look at nvidia's texture compression for instance.

The rate at which texture, geometry data and the frame buffer can be moved between the CPU, GPU, RAM and VRAM is far more important than the amount of storage available.
 
Last edited:
MadMossy said:
Did people using GDDR5 1GB cards suddenly have issues running games at 1080p 7 years ago over those with GDDR3 2GB cards, no they didn't. AMD have also added a new compression system for the HBM cards so could have a much higher effective VRAM limit than just 4GB. Look at nvidia's texture compression for instance.
Click to expand...
But if you play a game that use 2GB of vram the GDDR3 2GB card will do a better job compared to the GDDR5 1GB card.
 
AMD Radeon R9 390X and R9 390 GPUz data leaked

qR7vWDk.png


Yesterday Brazilian website leaked full list of MSI Radeon 300 graphics cards along with pictures and, more importantly datasheets. One ->datasheet<- in particular was very interesting. If this datasheet was correct then Radeon R9 390 would be in fact using full Hawaii silicon. Well unfortunately, that may not be true.
Click to expand...

http://videocardz.com/56373/amd-radeon-r9-390x-and-r9-390-gpuz-data-leaked
 
kylew_is_back said:
But if you play a game that use 2GB of vram the GDDR3 2GB card will do a better job compared to the GDDR5 1GB card.
Click to expand...

Very few games actually use a full 4GB of VRAM for textures, when you factor in texture compression as well that's a hell of a lot of VRAM to fill.

The frame buffer itself doesn't matter as its a tiny fraction of the available VRAM, 4K is barely 34MB per frame, 8MB for 1080p or 15MB for 1440p.
 
Status
Not open for further replies.
Back
Top Bottom