Vega can use any type of memory.
It won't have 4gb. Releasing a 4gb card would be worse than 780Ti 3GB.
Even GDDR3?
What bothers me is, Raja confirmed it will be called RX Vega. It means no numbers, and therefore not so many different configurations. We know there is 64 CU chip, but are there any 44 CU or 56 CU parts? Does that mean they'll only release one or two SKUs?
Even if this works, people won't understand it. They will see 4gb and run for the hills and buy Nvidia.You have not yet understood how Vega will work.
In short: Completely different than current cards.
Why? Because currently most data in VRAM is never actually used. The actually amount of data you need is even at 4k below 4 GB. Vega with the High-bandwidth cache controller (HBCC) will instead of filling up VRAM with useless stuff stream in the data that actually is required and hence will need far less vram for same performance. AMD has demoed this (albeit controlled demo) and such a vega card works perfectly fine even with only 2 GB of vram.
Even if this works, people won't understand it. They will see 4gb and run for the hills and buy Nvidia.
This unfortunately is true.
Even if this works, people won't understand it. They will see 4gb and run for the hills and buy Nvidia.
No they don't, and that's the big problem AMD are trying to solve, or at least alleviate on behalf of the devs. Many games are in fact memory inefficient, especially some console ports.Games already try to make the most efficient use of what memory they have
If you had seen the Radeon Pro SSG demo, slapping M.2 drives with a PLX chip directly to the PCB yielded an increased throughput for 8K video processing from 900MB/s to over 4GB/s. That should give you an indication of the potential of HBCC.Probably because they assume the whole concept is marketing fluff, which is what it is. Games already try to make the most efficient use of what memory they have, and streaming too/from gpu uses the same slow PCIE interface whatever type of memory the gpu possesses.
Probably because they assume the whole concept is marketing fluff, which is what it is. Games already try to make the most efficient use of what memory they have, and streaming too/from gpu uses the same slow PCIE interface whatever type of memory the gpu possesses.
HBM2 is two orders of magnitude faster than flash.HBCC only make sense if the card has onboard flash. Otherwise it sounds like smoke and mirrors. Like GTX 970 with its "4GB" of RAM. Works great, until you actually use and need all 4GB. But considering how well the 970 sold, maybe AMD are on to something!
HBCC only make sense if the card has onboard flash. Otherwise it sounds like smoke and mirrors. Like GTX 970 with its "4GB" of RAM. Works great, until you actually use and need all 4GB. But considering how well the 970 sold, maybe AMD are on to something!
And secondly there is a significant software overhead for a GPU to access RAM over PCIe, which is part of what the HBCC is there to address; So your claims of [speed over PCIe] equality are off target as well.
3072 GCN core chip with 4 GB of HBM2 with 512 GB/s bandwidth will be enough to play in 4K@60Hz in epic/very high settings.
Seems like you are a bad listener. This is literally the post above yours and makes yours sound like nonsense, lol.
HBM2 is two orders of magnitude faster than flash.
3072 GCN core chip with 4 GB of HBM2 with 512 GB/s bandwidth will be enough to play in 4K@60Hz in epic/very high settings.
One word. BS.It's also two orders of magnitude more expensive. Which is why you use both, connected together by an intelligent memory controller. If that's what this is then it has potential. Otherwise it is smoke and mirrors.
I am not saying that this will be the case at launch. Even if IPC clock for clock, core for core will increase from the beginning for Vega GPUs, they software is not designed so far to use feature that has biggest impact on performance of the GPU: Primitive Shaders. With time, as devs will get their job sorted you will see that Vega will start walking away from any previous generation architecture. We are talking about up to 50% performance uplift just from this one architecture change. You do not draw massive amount of data, that are not visible to the user. All of the power saved is available to you to create higher fidelity and quality of graphics.You are setting expectation very high with statements like that. If you haven't played a AAA title with said GPU @ 4K, you might want to be careful.
Actually it seems like you are a bad listener. A cache controller can ONLY access onboard HBM or system RAM over PCIe, unless there is another type of memory onboard the video card. There is nothing else for it to do. Is it really that hard to understand? If it is accessing system RAM over PCIe then its smoke and mirrors. It isnt going to do anything for performance.
It does for performance more than you think it does. HBCC is part of revolutionary Memory Paging System. We have not seen anything like this before. Data is delivered when its needed, where it is needed. It saves power, allows for higher performance, less stalls, higher efficiency...Actually it seems like you are a bad listener. A cache controller can ONLY access onboard HBM or system RAM over PCIe, unless there is another type of memory onboard the video card. There is nothing else for it to do. Is it really that hard to understand? If it is accessing system RAM over PCIe then its smoke and mirrors. It isnt going to do anything for performance.
Actually it seems like you are a bad listener. A cache controller can ONLY access onboard HBM or system RAM over PCIe, unless there is another type of memory onboard the video card. There is nothing else for it to do. Is it really that hard to understand? If it is accessing system RAM over PCIe then its smoke and mirrors. It isnt going to do anything for performance.
there is a significant software overhead for a GPU to access RAM over PCIe, which is part of what the HBCC is there to address; So your claims of [speed over PCIe] equality are off target as well.
too bad they can't do this in pure software for Fury X. Card still has tons of bandwidth and shading power (more cores than any other chip ever released!) but the 4gb holds it back at higher res