• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD responds to 4GB issue ...

i read somewhere that it has more to do with HBMs ability to release unused assets much quicker than GDDR5 effectively increasing the amount of info you can cram through PCIe when VRAM is constrained
 
1st gen HBM was limited to 4GB. If they could have put 8GB on there they would have. Just like they did with the 390X. Of course they are going to say it's not an issue, but you can see in certain benchmarks that it is an issue.

The amount of damage control AMD has had to do these last few days is unprecedented.
 
The Mac said:
i read somewhere that it has more to do with HBMs ability to release unused assets much quicker than GDDR5 effectively increasing the amount of info you can cram through PCIe when VRAM is constrained
Click to expand...

And how can VRAM decide that. It has to be built within the drivers.

HBM is nothing but stacked memory connected with an interposer. It cannot make memory management decisions :-/
 
This one sounds like a lot of BS (already posted in another thread too), cause of the PCI-e bus, as Mac said, it has more do with freeing up assets and buffers faster more then anything else, 4GB is 4GB doesn't matter if its HBM or GDDR5, what matters is the bandwidth and how its used.

Madgun, it does seem to be able to do those things faster because of the available bandwidth.
 
razor1 said:
This one sounds like a lot of BS (already posted in another thread too), cause of the PCI-e bus, as Mac said, it has more do with freeing up assets and buffers faster more then anything else, 4GB is 4GB doesn't matter if its HBM or GDDR5, what matters is the bandwidth and how its used.

Madgun, it does seem to be able to do those things faster because of the available bandwidth.
Click to expand...

That makes sense, but we'll still saturate the PCI-e bandwidth pretty fast in the need to continuously push and pull data into system ram.
 
AMD had released a bit of code to simulate the process of filling and emptying out the vram buffer it is was pretty impressive, once you understood what it was doing.

To me it is another way to do the same thing. Fitness vs. brute force.
 
Last edited:
Yeah I could see how they might be able to crank up how often the garbage collector is called, but its quite a stretch to say system ram will behave like vram because of HBM.
 
Taldren said:
Video interview:
https://youtu.be/DuA3T9MzNic?t=8m31s

Does that sound plausible to anyone else? I mean, I get what he is saying, but doesn't that just defeat the purpose?
Click to expand...

It makes sense as that's exactly what Shadows of Mordor does on my system when I enable the highest settings. I only have 8GB of ram and 4GB of VRAM on my R9 290 so it increases the size of my page file on my SSD. I could imagine that if I had more ram it wouldn't have to do this as much.

Nvidia does the same thing I suppose. If I remember right they recommend 2 times the VRAM memory for the Titan X. So having one Titan X means 24GB of system memory. Two Titan X is 48GB of system memory minimum. Four Titan X is 96GB of system memory minimum. Anyone that can afford four Titan X cards can easily afford 96GB of ram I'm sure. I looked again and it seems that Nvidia changed the recommended minimum to 16GB of system ram for a single Titan X. I could have sworn it was 24GB at first so that it can swap with the system ram.

Thanks for the video. Glad to see confirmation of what I observed and discussed in the Shadows of Mordor thread long ago. When I build my new PC next year I'll make sure that I stuff it with lots of ram. :)
 
cageymaru said:
It makes sense as that's exactly what Shadows of Mordor does on my system when I enable the highest settings. I only have 8GB of ram and 4GB of VRAM on my R9 290 so it increases the size of my page file on my SSD. I could imagine that if I had more ram it wouldn't have to do this as much.

Nvidia does the same thing I suppose. If I remember right they recommend 2 times the VRAM memory for the Titan X. So having one Titan X means 24GB of system memory. Two Titan X is 48GB of system memory minimum. Four Titan X is 96GB of system memory minimum. Anyone that can afford four Titan X cards can easily afford 96GB of ram I'm sure. I looked again and it seems that Nvidia changed the recommended minimum to 16GB of system ram for a single Titan X. I could have sworn it was 24GB at first so that it can swap with the system ram.

Thanks for the video. Glad to see confirmation of what I observed and discussed in the Shadows of Mordor thread long ago. When I build my new PC next year I'll make sure that I stuff it with lots of ram. :)
Click to expand...


Doesn't work that way, in SLI or multi card you still have 12 gb of vram and that's it, all assets are duplicated on each card. Now things that aren't duplicated are the buffers used to render to on screen or intermediate buffers for rendering and also AA an AF.
 
razor1 said:
Doesn't work that way, in SLI or multi card you still have 12 gb of vram and that's it, all assets are duplicated on each card. Now things that aren't duplicated are the buffers used to render to on screen or intermediate buffers for rendering.
Click to expand...

My quote from Shadows of Mordor testing last year.

http://hardforum.com/showpost.php?p=1041135757&postcount=214
Yes, when I go over the VRAM requirements @1080p my pagefile increases from 3GB to 11GB+. As soon as I end the benchmark and exit the game, my pagefile shrinks back to the 3GB - 4GB range instantaneously. I like the way that they handle the framerate when the VRAM is overextended. As shown above my fps doesn't alter in the slightest. I wonder if I had 16GB of ram would that help it swap textures better?


So either Shadows of Mordor is either making my page file huge or Windows just likes playing with the size when Shadows of Mordor runs.
 
cageymaru said:
My quote from Shadows of Mordor testing last year.

http://hardforum.com/showpost.php?p=1041135757&postcount=214
Yes, when I go over the VRAM requirements @1080p my pagefile increases from 3GB to 11GB+. As soon as I end the benchmark and exit the game, my pagefile shrinks back to the 3GB - 4GB range instantaneously. I like the way that they handle the framerate when the VRAM is overextended. As shown above my fps doesn't alter in the slightest. I wonder if I had 16GB of ram would that help it swap textures better?
Click to expand...


Sorry yes SOM and other games do do that with their page files, thought I removed the first part of your statement sorry.

have to add though the driver has control over the page file though, and so if we are to push Vram usage to insane amounts and there isn't enough vram to store all the assets, you will start seeing performance decreases because of the bandwidth limitations of the pci-e bus and if over pushed you will get slide shows.

Even if the the pci-e bus isn't limiting, there still will be some performance degradation (although minimal) because the GPU still has to process the files that are being read over to the vram even though its done in the background.
 
Last edited:
razor1 said:
Sorry yes SOM and other games do do that with their page files, thought I removed the first part of your statement sorry.
Click to expand...

Well to me it just makes sense to use the available system ram instead of a page file to swap textures when you run out of memory. Otherwise an Nvidia Titan X wouldn't need but the bare minimum that Windows 8.1 needs to run. Seems like even though my page file is on a SSD, that system memory is faster than a SSD. Thus the swap would be performed faster.

That's why we still make RamDisks even though there are SSDs on the market.
 
razor1 said:
Sorry yes SOM and other games do do that with their page files, thought I removed the first part of your statement sorry.

have to add though the driver has control over the page file though, and so if we are to push Vram usage to insane amounts and there isn't enough vram to store all the assets, you will start seeing performance decreases because of the bandwidth limitations of the pci-e bus and if over pushed you will get slide shows.

Even if the the pci-e bus isn't limiting, there still will some performance degradation because the GPU still has to process the files that are being read over to the vram even though its done in the background.
Click to expand...

Swapping on the page file in Shadows of Mordor is seamlessly smooth. I can play the game without lag, slowdown, or hiccuping when the game is using ridiculous amounts of VRAM and page file space. Of course I have a SSD to page file swap onto.
 
cageymaru said:
Well to me it just makes sense to use the available system ram instead of a page file to swap textures when you run out of memory. Otherwise an Nvidia Titan X wouldn't need but the bare minimum that Windows 8.1 needs to run. Seems like even though my page file is on a SSD, that system memory is faster than a SSD. Thus the swap would be performed faster.

That's why we still make RamDisks even though there are SSDs on the market.
Click to expand...

System ram is faster then the page file, think about it like this, when windows is put into hibernate it makes a page file for everything that windows has open at the time of hitting hibernate. It takes a lot longer to load up then from sleep which essential turns off everything other then the system ram.
 
cageymaru said:
Well to me it just makes sense to use the available system ram instead of a page file to swap textures when you run out of memory. Otherwise an Nvidia Titan X wouldn't need but the bare minimum that Windows 8.1 needs to run. Seems like even though my page file is on a SSD, that system memory is faster than a SSD. Thus the swap would be performed faster.

That's why we still make RamDisks even though there are SSDs on the market.
Click to expand...


Windows needs around 800 mb of vram and ram to be allocated to it.
 
cageymaru said:
My quote from Shadows of Mordor testing last year.

http://hardforum.com/showpost.php?p=1041135757&postcount=214
Yes, when I go over the VRAM requirements @1080p my pagefile increases from 3GB to 11GB+. As soon as I end the benchmark and exit the game, my pagefile shrinks back to the 3GB - 4GB range instantaneously. I like the way that they handle the framerate when the VRAM is overextended. As shown above my fps doesn't alter in the slightest. I wonder if I had 16GB of ram would that help it swap textures better?


So either Shadows of Mordor is either making my page file huge or Windows just likes playing with the size when Shadows of Mordor runs.
Click to expand...

IN many games I play the forums are full of posts about stuttering. I have never experienced it myself. I also have 24gb ram.

Gonna have to go fire up SOM later and see how that goes.
 
After reading about this debate since the Fury came out I've increased my ram from 8GB to 16GB. Screw it, ram is cheap and I'll do whatever I can to stop or minimize future stuttering.
 
AMD really should have used GDDR5 instead of HBM.

8 GB of GDDR5 instead of 4 GB of HBM would have made the product very competitive against the 980 Ti.

Just goes to show how poor AMD's engineering is these days.
 
AMD filled a niche by having a Smallform factor, something Nvidia doesn't have. GDDR5 would not have helped make that possible. Add to that having early access to HBM gives them some time to get a better understanding of it before Nvidia gets it. This is how early tech works. It isn't always out of the gate strong but generally makes a huge impact as it matures. At least we get to see it emerge and how it grows.
 
durquavian said:
AMD filled a niche by having a Smallform factor, something Nvidia doesn't have. GDDR5 would not have helped make that possible. Add to that having early access to HBM gives them some time to get a better understanding of it before Nvidia gets it. This is how early tech works. It isn't always out of the gate strong but generally makes a huge impact as it matures. At least we get to see it emerge and how it grows.
Click to expand...

"Before Nvidia gets it". Hate to burst your bubble but HBM is a JETEC spec - meaning it's not a secret to Nvidia, and they'll deliver HBM2 with up to 32GB on Pascal 16nm long before AMD has a die shrink followup to Fiji.

The SFF argument is also a little weak because that market is very narrow for a flagship part that AMD needs to sell more broadly in significant volume in order to stay in the game.
 
Last edited:
I have a feeling that even though "next year" AMD will have HBM2 and have 8GB, the games that need even more v-ram will out pace that 8GB.

Just my thoughts.
 
DPI said:
"Before Nvidia gets it". Hate to burst your bubble but HBM is a JETEC spec - meaning it's not a secret to Nvidia, and they'll deliver HBM2 with up to 32GB on Pascal 16nm long before AMD has a die shrink followup to Fiji.

The SFF argument is also a little weak because that market is very narrow for a flagship part that AMD needs to sell more broadly in significant volume in order to stay in the game.
Click to expand...

So... does Nvidia have it now? NO. Wait I didn't say that right... NO!

Fact of the Matter is AMD has right to HBM for a full year before Nvidia can even use it, per AMDs agreement with Hynix not JDEC... seriously cant believe you said that and spelled it wrong. Also white paper tells you little for how it actually runs. As others have alluded to GDDR5 with Nvidias first shot, terrible.
 
durquavian said:
So... does Nvidia have it now? NO. Wait I didn't say that right... NO!

Fact of the Matter is AMD has right to HBM for a full year before Nvidia can even use it, per AMDs agreement with Hynix not JDEC... seriously cant believe you said that and spelled it wrong. Also white paper tells you little for how it actually runs. As others have alluded to GDDR5 with Nvidias first shot, terrible.
Click to expand...

Yeah, I was about to ask why a company would spend 7 years co-developing a new technology just to instantly give it to the competition. That would have been right up there to when Xerox just gave apple/microsoft the mouse/windows design. :p
 
durquavian said:
Fact of the Matter is AMD has right to HBM for a full year before Nvidia can even use it, per AMDs agreement with Hynix not JDEC...
Click to expand...

Wow, Nvidia sure got the short end of that agreement, not being able to use HBM while it was limited to 4GB. They're really going to regret that when... when... uh...
 
durquavian said:
So... does Nvidia have it now? NO. Wait I didn't say that right... NO!

Fact of the Matter is AMD has right to HBM for a full year before Nvidia can even use it, per AMDs agreement with Hynix not JDEC... seriously cant believe you said that and spelled it wrong. Also white paper tells you little for how it actually runs. As others have alluded to GDDR5 with Nvidias first shot, terrible.
Click to expand...

No, they don't, source: http://wccftech.com/sk-hynix-begins-hbm-production-q1-2015-feature-amd-r9-390x380x-fiji-gpu/
Basically this memory is available for use both for AMD and Nvidia, however I would be willing to wager that it will appear first in Pirate Islands, specifically Fiji. On the off chance it doesn’t appear in Pirates Islands, it will probably then debut in an Nvidia GPUs. In my opinion, the chances of that happening (Green Debut) are very slim...
Click to expand...

NVIDIA were waiting for higher capacity modules.


​
You spelled it wrong too, it's JEDEC not JDEC.
 
Last edited:
RAM is RAM, unless there are some crazy compression technique, 4GB of VRAM can't store enough data for some games that already use more than 4GB of VRAM.
 
durquavian said:
So... does Nvidia have it now? NO. Wait I didn't say that right... NO!

Fact of the Matter is AMD has right to HBM for a full year before Nvidia can even use it, per AMDs agreement with Hynix not JDEC... seriously cant believe you said that and spelled it wrong. Also white paper tells you little for how it actually runs. As others have alluded to GDDR5 with Nvidias first shot, terrible.
Click to expand...

Why are you such a die hard AMD fan. It doesn't leave an ounce of credibility in what you say either. So please next time, don't point it for other people. You are no better.
 
I like how Richard Huddy's anwser basically says if you don't have enough GPU memory just use more system memory, without mentioning that the GPU has to talk through the ~16GB/s PCI-E bottleneck. That's great so you've got 512GB/s of nearly useless bandwidth if you have to swap resources out constantly because you can only do so at 16GB/s. Ah well, should've seen that coming a mile away. It would be interesting to see someone do some benchmarks with GPU-Z open on the sensors tab showing Bus Interface Load on a game that definitely uses more than 4GB of VRAM.
 
Mmm I was wondering how they will try to spin that 3,5 GB vram is not enough but 4GB is plenty even for 4K :D
 
xLokiX said:
No, they don't, source: http://wccftech.com/sk-hynix-begins-hbm-production-q1-2015-feature-amd-r9-390x380x-fiji-gpu/


NVIDIA were waiting for higher capacity modules.


​
You spelled it wrong too, it's JEDEC not JDEC.
Click to expand...
Actually your article has no facts to who can and can not use HBM now. For a number of years it was said that AMD would get early production and Nvidia would have to wait. Whether it is in fact the case or not doe not matter. AMD still gets a year or more lead to test.

madgun said:
Why are you such a die hard AMD fan. It doesn't leave an ounce of credibility in what you say either. So please next time, don't point it for other people. You are no better.
Click to expand...
Not being diehard anything here. Fact is they were just spouting inaccurately and well as you put diehard fanboy-like. AMD has it out in production NOW. Nvidia wont till next year at best, NOT NOW. Simple as that.
 
flak-spammer said:
I like how Richard Huddy's anwser basically says if you don't have enough GPU memory just use more system memory, without mentioning that the GPU has to talk through the ~16GB/s PCI-E bottleneck. That's great so you've got 512GB/s of nearly useless bandwidth if you have to swap resources out constantly because you can only do so at 16GB/s. Ah well, should've seen that coming a mile away. It would be interesting to see someone do some benchmarks with GPU-Z open on the sensors tab showing Bus Interface Load on a game that definitely uses more than 4GB of VRAM.
Click to expand...

He was more or less likely speaking to this:

To substantiate my comment about driver trickery, this is a quote from TechReport's HBM article:

"When I asked Macri about this issue, he expressed confidence in AMD's ability to work around this capacity constraint. In fact, he said that current GPUs aren't terribly efficient with their memory capacity simply because GDDR5's architecture required ever-larger memory capacities in order to extract more bandwidth. As a result, AMD "never bothered to put a single engineer on using frame buffer memory better," because memory capacities kept growing. Essentially, that capacity was free, while engineers were not. Macri classified the utilization of memory capacity in current Radeon operation as "exceedingly poor" and said the "amount of data that gets touched sitting in there is embarrassing."

Strong words, indeed.

With HBM, he said, "we threw a couple of engineers at that problem," which will be addressed solely via the operating system and Radeon driver software. "We're not asking anybody to change their games.""


Just he does a bad job of it and likely doesn't have a full understanding of it either.
 
durquavian said:
Actually your article has no facts to who can and can not use HBM now. For a number of years it was said that AMD would get early production and Nvidia would have to wait. Whether it is in fact the case or not doe not matter. AMD still gets a year or more lead to test.
Simple as that.
Click to expand...

You know what, AMD does need a year's lead to straighten things out on the drivers side. That's the usual amount they take anyways with a new product release.

NVidia, whenever they launch Pascal, should be on top of things from the very first day.

But I agree with you, AMD definitely needs more time!
 
madgun said:
You know what, AMD does need a year's lead to straighten things out on the drivers side. That's the usual amount they take anyways with a new product release.

NVidia, whenever they launch Pascal, should be on top of things from the very first day.

But I agree with you, AMD definitely needs more time!
Click to expand...

Now see the inherent negativity in your post? Not necessary is it? Besides this is the first use in GPUs of this tech. When Pascal comes HBM will be a tried tech and HBM2 will have the luxury of learning from HBM1. Not sure why more of you aren't enjoying the discussion of the tech even if the release is not up to whatever level of performance you were hoping for.
 
durquavian said:
Now see the inherent negativity in your post? Not necessary is it? Besides this is the first use in GPUs of this tech. When Pascal comes HBM will be a tried tech and HBM2 will have the luxury of learning from HBM1. Not sure why more of you aren't enjoying the discussion of the tech even if the release is not up to whatever level of performance you were hoping for.
Click to expand...


So you are saying HBM1 memory tech isn't mature? Come on dude, that doesn't make much sense either.
 
You must log in or register to reply here.
Back
Top