• 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.

Poll: The Vega Review Thread.

What do we think about Vega?

  • What has AMD been doing for the past 1-2 years?

  • It consumes how many watts and is how loud!!!

  • It is not that bad.

  • Want to buy but put off by pricing and warranty.

  • I will be buying one for sure (I own a Freesync monitor so have little choice).

  • Better red than dead.


Results are only viewable after voting.
I think at the moment HBCC only doubles your vram presently. That's what I remember hearing/reading.

Both HBCC and Pascal unified memory support 49-bit address space (which is 512TB or something) though the CPU IIRC has slightly lower limits - I'm guessing this is to cover both GPU resident storage as well as system storage in general.

The biggest impact of this will be professional compute - I'll be very surprised if there is any meaningful impact on gaming any time soon.
 
The answer to your question as to why they didn't put 4GB is probably that they found 8GB to be good a buffer, between feeding the gpu and moving data across and power. The issue with the 970 was that Nvidia hid this information and didn't mention it. Also I'm pretty certain that everyone i've seen when talking about the fury say that it should have had more memory so stop trying to twist history to make a point. Your penultimate point has nothing to do with this.

Overpriced for game. Yes I agree, but for compute its right where it should be.
Well I'm sure that not everyone I've seen seemed to think 4GB HBM was a problem. So from my point-of-view/memory I'm not twisting history. What you've seen may lead you to a different conclusion, I can't base my posts on what you saw.

What a silly post.
Ditto

They is a massive difference between running from system ram/HDD vs the High Bandwidth "Cache Controller"

The key difference between then and now is how the cache controller handles the data.

Just like how HDD with SSD cache work to speed up transfer speeds.
But there's got to be a performance hit from using HDD space, even with HBCC hasn't there?
My point being that because 0.5GB of the 970's VRAM ran slower a huge number of people complained about it. When AMD implement a system to use slower memory it's a 'feature' and people love it. I realise that at least some people were annoyed because Nvidia misrepresented/lied about some of the numbers, but a lot of people seemed happy with the cards until this issue was raised, it's not like they noticed the problem.
If a game needs 10GB VRAM I'd rather have an 11GB DDR5X 1080Ti than an 8GB RX Vega 64 with HBCC. If a game needs 12GB VRAM, well then I'd probably want a new card with 16GB VRAM rather than 8GB and HBCC. Much like if a program needs 10GB system RAM I'd rather have 16GB DDR3 than 8GB DDR4 and use the swap file.

I guess some people will just love everything AMD do because AMD do it. Currently I think Vega is decent but 12 months late and probably too expensive. Also, it'd be nice if it was more efficient.
If I was buying 1 card I'd buy a 1080Ti. If I was then buying a 2nd card for a different PC I'd be tempted with a Vega, but currently for me it's looking like a repeat of the last gen card, Nvidia have just pipped it.
 
Yep, my first ever computer 'mod' was feeling like a God when upgrading it to 1mb ram!

/OT

Likewise, in my case it was 128MB though.

I ran a BBS in the background in case you were wondering and it was the almighty A3000. ;)

I might look at getting one of those new Vulture cards if I can get SCSI attached to it.

\OT
 
But there's got to be a performance hit from using HDD space, even with HBCC hasn't there?
My point being that because 0.5GB of the 970's VRAM ran slower a huge number of people complained about it. When AMD implement a system to use slower memory it's a 'feature' and people love it. I realise that at least some people were annoyed because Nvidia misrepresented/lied about some of the numbers, but a lot of people seemed happy with the cards until this issue was raised, it's not like they noticed the problem.
If a game needs 10GB VRAM I'd rather have an 11GB DDR5X 1080Ti than an 8GB RX Vega 64 with HBCC. If a game needs 12GB VRAM, well then I'd probably want a new card with 16GB VRAM rather than 8GB and HBCC. Much like if a program needs 10GB system RAM I'd rather have 16GB DDR3 than 8GB DDR4 and use the swap file.

I guess some people will just love everything AMD do because AMD do it. Currently I think Vega is decent but 12 months late and probably too expensive. Also, it'd be nice if it was more efficient.
If I was buying 1 card I'd buy a 1080Ti. If I was then buying a 2nd card for a different PC I'd be tempted with a Vega, but currently for me it's looking like a repeat of the last gen card, Nvidia have just pipped it.

From my understanding the point of HBCC is to swap out data that is not being actively used. In theory it should fetch data from the RAM/HDD before the GPU needs to run calculations on it (There will be misses at times). This is different to the 970 issue because the slow down came because the GPU was trying to access data that it wanted to process at that point in time. They are in no way equivalent situations.

What the point in having 16GB of RAM on you system and on your GPU when most of it isn't being actively used? All you do is increase system cost and power usage.

Edit: This guy rendered a 50GB scene with a VEGA card thanks to HBCC. Such a scene wouldn't even load on Nvidia card unless it has enough VRAM. Wonder how much that would cost.

https://blenderartists.org/forum/showthread.php?426522-AMD-RX-Vega/page7&p=3230126#post3230126
 
Last edited:
Both HBCC and Pascal unified memory support 49-bit address space (which is 512TB or something) though the CPU IIRC has slightly lower limits - I'm guessing this is to cover both GPU resident storage as well as system storage in general.

The biggest impact of this will be professional compute - I'll be very surprised if there is any meaningful impact on gaming any time soon.


Agreed.

In theory you could design games to require far more memory with lots of high resolution assets. Considering Vega users will be in such a minority that you it would be a huge waste of resource to cater just for them, and its not just the software side which is relatively easy but in generating the assets that only a few users would use is just not worth it, or you would need to create a software-ware workaround. This leads to the assets creation problem. The ID4 engine's biggest advancement was the use of Megatexture technology, so every surface could have a unique and high resolution texture.. MultiGB of texture data could be used in a single level, way more than would fit in VRAM. The problem was that creating this giant texture took a huge amount of artistic resources and the visual gains did not relate well to the extra costs. The artists also tended to get lazy, so although different areas could have completely unique textures it is much easier to splatter the same textures about. The huge asset size also makes for very big downloads, which is not ideal the way PC games are now distributed. Furthermore, John Carmack proved, you don't actually need a hardware solution to stream in huge amounts of asset in real time. HBCC will make it faster, more seamless, easier to code etc., but that is likely not where the costs lie but in the asset creation in the first place. And finally, use of HBCC will mean the game is churning through more than 8Gb of data per frame (otherwise it woudlnt be useful), vega just simply isn't powerful enough to cope or has enough bandwidth.



It is a technology for compute and HPC that AMD are spinning as a feature for gaming. maybe in the future when mnay users have GPUs with similar technology, asset creation has been made cheaper (perhaps better tools for automating texturing etc? and perhaps memory prices have sky rocketed HBCC will make sense.
 
From my understanding the point of HBCC is to swap out data that is not being actively used. In theory it should fetch data from the RAM/HDD before the GPU needs to run calculations on it (There will be misses at times). This is different to the 970 issue because the slow down came because the GPU was trying to access data that it wanted to process at that point in time. They are in no way equivalent situations.

What the point in having 16GB of RAM on you system and on your GPU when most of it isn't being actively used? All you do is increase system cost and power usage.

Edit: This guy rendered a 50GB scene with a VEGA card thanks to HBCC. Such a scene wouldn't even load on Nvidia card unless it has enough VRAM. Wonder how much that would cost.

https://blenderartists.org/forum/showthread.php?426522-AMD-RX-Vega/page7&p=3230126#post3230126



Unless the HBCC includes the cape ability of time travel then it can't know beforehand what data is needed. the HBCC just tries to make this smoother and faster but is still limited by the disk/system ram speed.
 
From my understanding the point of HBCC is to swap out data that is not being actively used. In theory it should fetch data from the RAM/HDD before the GPU needs to run calculations on it (There will be misses at times). This is different to the 970 issue because the slow down came because the GPU was trying to access data that it wanted to process at that point in time. They are in no way equivalent situations.

What the point in having 16GB of RAM on you system and on your GPU when most of it isn't being actively used? All you do is increase system cost and power usage.

Edit: This guy rendered a 50GB scene with a VEGA card thanks to HBCC. Such a scene wouldn't even load on Nvidia card unless it has enough VRAM. Wonder how much that would cost.

https://blenderartists.org/forum/showthread.php?426522-AMD-RX-Vega/page7&p=3230126#post3230126
50GB scene of what?
A game? How quickly did it do that?
 
Might be a simple question for some of you but does the Vega56 now finally come with 4K HDR over HDMI?

Looking for something to replace my nano. Vega56 seems like the only worthwhile new gpu from AMD.
 
Might be a simple question for some of you but does the Vega56 now finally come with 4K HDR over HDMI?

Looking for something to replace my nano. Vega56 seems like the only worthwhile new gpu from AMD.
Maybe it's just me, but if I were going to get a Vega card I'd want a Vega 64 (either AIO, 3rd party or WC'd). The Vega 56, while better, doesn't seem like a huge step up from the Fury X I'm running currently (and currently I'm running Fury X CF).
Not sure why everyone seems so keen on the 56, what am I missing?
 
Maybe it's just me, but if I were going to get a Vega card I'd want a Vega 64 (either AIO, 3rd party or WC'd). The Vega 56, while better, doesn't seem like a huge step up from the Fury X I'm running currently (and currently I'm running Fury X CF).
Not sure why everyone seems so keen on the 56, what am I missing?

Yeah i don't think the 56 is worth upgrading to from the Furys, id go the 64, AIO cooled too.
 
This is absurd. Since when did people's relationship with AMD/NVIDIA develop so deeply as to start arguing points such as 'anything I do is never good enough" and "somehow you can never do wrong".

The virtual world is a very strange place indeed, I'm just glad I don't live here and only pop in to visit.
 
The problem for AMD will come when HBMx is actually cheap and widely available, then Nvidia can get the same power savings, just think how absolutely horrific Vega would seem if Pascal used HBM2 and draw 40w less!

It already seems horrific if you make the AIO version the comparison point.
 
Maybe it's just me, but if I were going to get a Vega card I'd want a Vega 64 (either AIO, 3rd party or WC'd). The Vega 56, while better, doesn't seem like a huge step up from the Fury X I'm running currently (and currently I'm running Fury X CF).
Not sure why everyone seems so keen on the 56, what am I missing?

Brain Damage :D
 
There is no reason to think AMD cant make up the technical deficit between them and Nvidia like they did with Ryzen and Intel.

You can jump more than one step at once.

No doubt, no different to intel and netburst, amd and piledriver, or nvidia and the fx series. The tide will turn at some point.
 
Back
Top Bottom