• 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.

AMD Gaming Scientist: 4GB HBM is better than 8GB GDDR5

Status
Not open for further replies.

FuryXGuy

FuryXGuy

FuryXGuy

Associate
Joined
8 Jan 2016
Posts
94
Location
Across the pond
In an interview AMD's Richard Huddy states that 4GB HBM is better than 8GB GDDR 5.

https://www.youtube.com/watch?v=NQ8YlXh-jbI

As is stated in the article and the transcript:

According to him 4GB HBM:

"Effectively get rid of frame buffer size.” He said it “exceeds the capability of 8GB or 12GB of memory and the reason for that is that there is so much bandwidth inside HBM that if you have system memory we can swap memory around inside the machine, swap between HBM and system memory and keep the working set in the 4 Gigabytes and it never get’s in the way of the GPU. “
He goes on to say: “What happens is you effectively get rid of the problems of frame buffer size and the extraordinary result that comes on the back of that is when you benchmark a Fiji chip is..when start to wind up the resolution higher and higher you would start to think that our 4GB would come to the limit to the headroom a bottleneck but far from that it actually as you wind the resolution up we get better and better we start beating a Titan X and indeed we consistently beat it if you go to high enough resolution so HBM is actually the future of memory."

Note AMD also did driver optimization for the 4GB HBM Limit (as noted by AnandTech review) by assigning engineering resource to it. As such you can see that in Shadow of Mordor despite it pushing 4GB it doesn't have any stutter maxed out.

https://www.youtube.com/watch?v=PPs96gghULY

Source:
TechAvenue

I would choose 4GB HBM over 8GB GDDR5 anyday!
 
Last edited:
I've only ever seen you post about why AMD/HBM is better than Nvidia at the moment, do you regret your purchase and trying to convince yourself by reading up on this stuff?
 
savagesam117 said:
I've only ever seen you post about why AMD/HBM is better than Nvidia at the moment, do you regret your purchase and trying to convince yourself by reading up on this stuff?
Click to expand...

No. But I did read up on this stuff (especially about the HBM stuff). And I do find it fascinating. I mean it is a revolutionary departure from how memory was utilized in the past 20 years or so. For the first time even we are moving the memory on two the GPU via an interposer instead of having the GDDR5 on the PCB with traces running to the GPU.

I believe he is right. HBM is the Future. I expect that GPU's will mostly have HBM over the next 10 years.

How about you? You like your Fury X?
 
Last edited:
FuryXGuy said:
I believe he is right. HBM is the Future. I expect that GPU's will mostly have HBM over the next 10 years.
Click to expand...

On that front, he probably is right, but to say flat out that 4GB HBM is better than 8GB GDDR5 is just wrong. In certain situations it might very well be better, but in others it would be worse. It is not a blanket one is better than the other situation.
 
andybird123 said:
Where is Richard Buddy these days, he was pretty prolific 6 months ago but has gone strangely silent of late. With statements like these its not hard to see why.
Click to expand...

This article is old.

Pretty sure he was sent to the same farm as Roy after RTG. Raja has been doing all the interviews since.

But what exactly is wrong with what he's saying? He's explaining to gamers why his company's product beats TX in VRAM hog games. AKA his job. Countering the Nvidia brainwashing field that emanates from every GeForce product.
 
Last edited:
So does this mean Polaris will only have 4GB HBM2?
Why put more on there if 4GB isn't a limit anymore? Is it just so they can hike the price up and charge us more? Got to keep up with Nvidia on the pricing after all or they'll be the "value" brand not a "premium" brand!
 
There has got to be some sort of resource costs swapping data around from system memory to GPU memory on the fly. I get what he is saying, that its much better at this than GDDR5 but isn't it just better to have more capacity than rather relying on swapping data around in the system. From Ram to Vram?
 
GoogalyMoogaly said:
So does this mean Polaris will only have 4GB HBM2?
Why put more on there if 4GB isn't a limit anymore? Is it just so they can hike the price up and charge us more? Got to keep up with Nvidia on the pricing after all or they'll be the "value" brand not a "premium" brand!
Click to expand...

That's what I am wondering too. I read about Samsung now ramping up 4GB HBM2 to be introduced later this year with Pascal. I am like shouldn't it be 8GB HBM2 as that was the major difference between HBM1 and HBM2 is the ability to pile on extra GB. The only benefit I see with 4GB HBM2 is that bandwidth will now be doubled to 1TB/s. But wouldn't that be a bit redundant without the extra 4GB. Especially for games at 4K going on for the next 2 - 3 years.
 
FuryXGuy said:
That's what I am wondering too. I read about Samsung now ramping up 4GB HBM2 to be introduced later this year with Pascal. I am like shouldn't it be 8GB HBM2 as that was the major difference between HBM1 and HBM2 is the ability to pile on extra GB. The only benefit I see with 4GB HBM2 is that bandwidth will now be doubled to 1TB/s. But wouldn't that be a bit redundant without the extra 4GB. Especially for games at 4K going on for the next 2 - 3 years.
Click to expand...

I think 4GB HBM2 refers to each stack and you'll get 4 on a card (for 16GB) where as HBM is only in 1GB stacks.
 
FuryXGuy said:
In an interview AMD's Richard Huddy states that 4GB HBM is better than 8GB GDDR 5.

https://www.youtube.com/watch?v=NQ8YlXh-jbI

As is stated in the article and the transcript:

According to him 4GB HBM:

"Effectively get rid of frame buffer size.” He said it “exceeds the capability of 8GB or 12GB of memory and the reason for that is that there is so much bandwidth inside HBM that if you have system memory we can swap memory around inside the machine, swap between HBM and system memory and keep the working set in the 4 Gigabytes and it never get’s in the way of the GPU. “
He goes on to say: “What happens is you effectively get rid of the problems of frame buffer size and the extraordinary result that comes on the back of that is when you benchmark a Fiji chip is..when start to wind up the resolution higher and higher you would start to think that our 4GB would come to the limit to the headroom a bottleneck but far from that it actually as you wind the resolution up we get better and better we start beating a Titan X and indeed we consistently beat it if you go to high enough resolution so HBM is actually the future of memory."

Note AMD also did driver optimization for the 4GB HBM Limit (as noted by AnandTech review) by assigning engineering resource to it. As such you can see that in Shadow of Mordor despite it pushing 4GB it doesn't have any stutter maxed out.

https://www.youtube.com/watch?v=PPs96gghULY

Source:
TechAvenue

I would choose 4GB HBM over 8GB GDDR5 anyday!
Click to expand...

Total and utter garbage.

4 Fury Xs running the Shadow of Mordor bench @4k take about 3 minutes to load it as the memory is just not enough.

4 TitanXs running the Shadow of Mordor bench @8k (4x the resolution) take about 3 seconds to load it.

It is a total mismatch using the Shadow of Mordor bench to compare these cards as the TitanXs are so far ahead in the memory buffer size and Richard Huddy should know better.

I think we have been down this road before when someone on AMDs staff called the Fury X an overclockers dream at the launch.

The Fury X is a very good card but statements like the rubbish in the OP do nothing to help it at all and only give the haters more ammunition.
 
Kaapstad said:
Total and utter garbage.

4 Fury Xs running the Shadow of Mordor bench @4k take about 3 minutes to load it as the memory is just not enough.
Click to expand...

The Fury X stuttering issues in 4K has been fixed with newer drivers. Haven't seen any stuttering or frametime issues with Shadow of Mordor.
 
Had a curious experience with having a 970 sitting beside my 780 lately - in a lot of games with ultra settings @ 1440p the 970 sits at 3.4-3.6GB VRAM used the 780 at 2.8GB used and there is pretty much zero difference between them ingame - framerates aren't identical but as close as makes no odds that you can't tell a difference without the framerate counter up. Things get a bit more interesting at 4K and using mods and/or settings that can push past 5GB of (actual) VRAM usage - up until 4GB use very little difference aside from the 780 can take a bit longer to load stuff once you are getting close to 4GB used but once loaded runs pretty much identical to the 970 until VRAM use goes past 4GB and then pretty much on the dot of 4GB both turn into a slide show with noticeable stuttering as you are moving through the game world. Sadly don't have a true 4GB card like a 980 to see if it does the same at 4GB or if that has a bit more headroom before it starts to suffer.

Do have to hand it to AMD though - I was fully expecting to see the aggressive memory management required to make the Fury(X) work fall apart on them but they've mostly seemed to manage to stay on top of it.
 
Status
Not open for further replies.
Back
Top Bottom