rx570: 4gb vs 8gb?

[qu4k3r]

No utility actually reports "how much VRAM the GPU is actually using — instead, it reports the amount of VRAM that a game has requested. We spoke to Nvidia’s Brandon Bell on this topic, who told us the following: “None of the GPU tools on the market report memory usage correctly, whether it’s GPU-Z, Afterburner, Precision, etc. They all report the amount of memory requested by the GPU, not the actual memory usage. Cards will larger memory will request more memory, but that doesn’t mean that they actually use it. They simply request it because the memory is available.”

I found a review from a brazilian site with exactly these two cards. For the same game, sometimes they allocated more or less the same amount of vram, but sometimes not. However they perfrom almost the same.

I don't play the latest games anyways, so I guess the powercolor rx570 4gb will be enough for what I need atm. Thanks for all your anwsers

 

[SoNic67]

The fact that both Windows 10 "Task Manager - Performance tab" and GPU-Z show same amount of GPU memory tells me that what nVidia guy said might have to be taken with a grain of salt. Maybe is true for nVidia cards due to their compression algorithm (but I doubt it).
Also, the fact that a random reviewer says "that there is no quality difference" in not proof. I can say it is in a Youtube video, so it's he said, I said type of deal. Without actual tools to measure the texture quality, I tend not to believe it.

 

[EarthDog]

Games are different in how they are coded to use memory. That's a pretty interesting blanket statement (by NVIDIA) honestly...

I then wonder why in games would the allocation change. For example, in those utilities during a game the "allocated" memory is incredibly dynamic going up and down throughout. Why would it use/waste the extra cycles to constantly reallocate space instead of acting more like a page file? If it peaks at 6gb, why does it appear to instantly 'reallocate' memory to a lesser amount? That doesn't make sense to me.

A game knows how much it needs (GTA V lists the amount, so does SOTR)...why wouldn't it just make a slightly larger pool? I don't understand what efficiencies that method would have.

I also wonder how much it is off from the "allocated" amount the utilities show. I'll have to check to confirm, but I recall in SOTR and GTA V, when benchmarking, what it listed was damn close to what the utilities read. So....how off is it really? While that statement is true, is it just a couple percent so what we see is 'good enough'for most? Or are we talking like 20%???

How did that site capture actual ram use???

I've seen it mentioned when it happens. It isn't rare. And many just run a canned bench and mark the score, so there's that.


Again, its not so much about average than it can be about minimums and the game experience being as affected in some cases. The severity of things depends on the title and how it manages memory as well. It isn't a black and white issue.

@John Naylor - Thoughts......

 

[John Naylor]

It's not about how the game is coded, it's about access to the information ... the info is not available. When installing, the routine looks at the physical amount of RAM on the card ... and says OK, he has 8, lets grab 6 for us. The validity of this presumption initially was that Max Payne 2 would refuse to install at 5760 x 1080 with 2 GB card ... swap in a 4 GB model and it was fine ... swap back... same avg fps, same min fps, same image quality, same everything.

Of course two different things will record close to the same numbers ... that's the only information available.

As to why peeps have **seen** problems, well it's pretty well known how you can create such problems.... load up max settings at 4k, zoom in and out repetaedly and you get your 15 minutes of fame.

Yes, these have all been addressed in the referenced reviews. To my eyes, when Guru3d, PugetSound, ExtremeTech and Alienbabeltech all get the same results on 6xx, 7xx, 9xx cards, ... and yes, same avg fps, same min fps, same image quality, same everything.... it's passed "peer review". They did the same thing to get their 15 minutes of fame with the 3.5 GB **problem** with the 970 ... except as was shown, anytime you could ***create*** the problem on the 970, doing the same thing created the same problem on the 980.

 

[EarthDog]

@John Naylor

When installing, the routine looks at the physical amount of RAM on the card ... and says OK, he has 8, lets grab 6 for us.

Ok, that is upon installation. But what we seen in tools like MSI AB is that the RAM "allocation" CONTSTANTLY changes. So while I do believe there are things that check requirements and prevent installation, I think it is bullocks that these tools aren't reading an actual vRAM amount.

Assuming it is true, how much is MSI AB and such off? How did that website which called out those applications for not being accurate actually measure the RAM?

The information is nice, but lacks logic and sense to me...