When will we see Fury reviews?

[bryanW1995]

The true story here is that Nvidia has developed a culture of almost rabid type fan support. They have to be careful attacking the "weakness" of the Fury (4G Vram) because as I said they did the very same thing with the GTX970 (Except they did NOT truly disclose the 3.5 +.5 memory configuration until called out on it by reviewers brave enough to challenge them.) When "outed" they daid they could deal with the issue with software compression.

You don't think that this forum is covered up with die hard Red Team members, too? Neither camp has to pay people any money at all to sing their praises and trash the other Team. Focus Group's days of hiding in the shadows are long gone.

The other cards look to be straight rebrands:

http://forums.anandtech.com/showpost.php?p=37488414&postcount=1

Anyone here already knows Fury is the best cards ever.

^^^^ See what I mean?

FYI, adding anything at all to card makes it a refresh, not a rebrand. Even if all they did was add 50 mhz and 4 gb ram, that would still be a refresh. It looks like they did a bit more than that even, but it doesn't matter.

Excellent post RS. I'll referencing it next time someone who doesn't understand the simple concept of vram usage in their monitoring software doesn't equate to vram required.

I thought it was lol-worthy that [H]'s article and phrasing "6GB is the MINIMUM for 4K" is so well timed, even when their own results show otherwise (980SLI & R295X faster than 980Ti/Titan X!). Those guys are a bunch of sellouts.

I don't think they're sellouts at all, they're just lazy and not particularly intelligent.

So I guess the next question is , why do you care so much what other people buy?:whiste:

Pot: kettle.

 

[RussianSensation]

How are you going to determine how much a game actually "needs"? Serious question.

That is really easy actually. When I put up 4xMSAA at 1600x1200 with 8800GTS 320MB my performance dropped to 4-5 fps in the original Dirt. When I removed AA, fps shot up to 25-30. Then you can test the same scene with an 8800GTS 640MB or 8800GTX 768MB and see what happens.

VRAM bottleneck can be seen with major drops in minimums, chugging and stuttering that occurs constantly, not just once in 3 hours of gaming. The performance drop with a real VRAM bottleneck is not 5% not 10%, it's usually massive like 30-200%. Any professional GPU reviewer would be able to tell from gameplay.

For example:

FPS drops from 50 to 30 (40% drop on the same GPU)

Or this, at 1920x1200 performance drops from 82-83 fps to ~57 fps (38% drop or so on the same GPU). Moving to the highest resolution, the 512MB card is very close to 60 fps and the 256MB version is barely above 40 fps.

So yea, putting the focus on VRAM now is a great strategy to diminish enthusiasm for Fury.

Right, but performance has to show it. You can't make statements like the reviewers at HardOCP did about VRAM bottlenecks and show superior performance and minimum fps on 4GB vs. 6-12GB cards. None of the games they showed to use > 4GB of VRAM require that amount to run which is why 295X2/980 SLI outperform Titan X at 4K in each of those titles. Therefore, their conclusion that 4GB holds back certain cards is flawed. They don't understand the difference between a game requiring 4GB of VRAM and a gaming dynamically using more than that for caching.

I'll provide some examples how to spot VRAM bottleneck - You gauge relative performance of 320MB vs. 512MB vs. 768MB cards from the same gen:

No VRAM bottleneck on that 320MB card:

Now looking at that graph above, let's pay attention to how 8800GTS 320MB performs relative to 2900PRO 512MB or 8800GTX 768MB. Now the reviewer turns on AA in the same game - huge VRAM bottleneck exposed! You cannot have an 8800GTX performing 4X faster than an 8800GTS and 2900PRO 2X faster. We can say without a doubt it's a VRAM bottleneck:

Here is another perfect example - 8800GT 512MB is 2X faster than 8800GTS 320MB with AA in COH = 100% VRAM bottleneck. On average, 8800GT was only about 20% faster than the 8800GTS 640MB so we cannot have a situation of 8800GT outperforming 8800GTS by 2X unless the 320MB version of GTS was VRAM bottlenecked.

http://www.xbitlabs.com/articles/graphics/display/leadtek-8800gt-extreme_11.html#sect1

 

[bryanW1995]

Not really a review, but AMD did have a performance related slide in their Chinese presentation. You are seeing that correct. Fury X MIN FPS is higher than Titan X MAX FPS in FC4 4k.

http://live.pconline.com.cn/483.html

Weird that their Chinese presentation was in English.

 

[exar333]

Excellent post, though that last part...I mean, come on. How many "professional review sites" have a clue about this? Give me computerbase.de, tpu, and BFG10k, almost everybody else is just noise with either an agenda for one camp or the other or a clueless rig-tinkerer who's only posting the review for a paycheck.



Huh? RS used [H]'s own reviews to show how unimportant the Ram amount was in those games. GTX 980 sli was 20% faster than Titan X on 4k Ultra settings. Go back and read his post.

That point isn't even relevant though. Flawed or not, that only shows that 6GB vs 12GB doesn't matter. Making any assertions on Fury or 4GB based on that is illogical at best, foolish at worst. The point of the argument is whether 4GB is enough, or 6GB is needed, etc.

I could go back and find literally a hundred posts of AMD fans stating a 8GB model would be announced at launch. Where is that? 4GB or bust. It is what it is. Hopefully the reviews we get soon will truly answer if its enough.

If I see enough evidence that its not an issue (or just in some more uncommon cases) that's my next card. Period.

Making any assumptions that 4GB IS or IS NOT a limiting factor for Rury is impossible right now. We don't know enough and I am surer that NV fanboys will find those edge-cases where maybe it does in specific conditions for a specific title. Thats fine, but is that a true exception, or the norm? I am leaning toward the exception, but again we don't enough yet.

Edit: Even extrapolating on current VRAM usages is tough because of HBM and what AMD may have written in their drivers to assist. Obviously AMD is doing what they can to make it work, but it remains to be seen if it does that in all cases.

 

[Stuka87]

That point isn't even relevant though. Flawed or not, that only shows that 6GB vs 12GB doesn't matter. Making any assertions on Fury or 4GB based on that is illogical at best, foolish at worst. The point of the argument is whether 4GB is enough, or 6GB, etc.

I could go back and find literally a hundred posts of AMD fans stating a 8GB model would be announced at launch. Where is that? 4GB or bust. It is what it is. Hopefully the reviews we get soon will truly answer if its enough.

If I see enough evidence that its not an issue (or even just in VERY rare cases) thats my next card. Period.

Making any assumptions that 4GB IS or IS NOT a limiting factor for Rury is impossible right now. We don't know enough and I am surer that NV fanboys will find those edge-cases where maybe it does in specific conditions for a specific title. Thats fine, but is that a true exception, or the norm? I am leaning toward the exception, but again we don't enough yet.

The GTX 980 has 4GB, not 6GB. You are thinking of 980Ti.

 

[Kenmitch]

You don't think that this forum is covered up with die hard Red Team members, too? Neither camp has to pay people any money at all to sing their praises and trash the other Team. Focus Group's days of hiding in the shadows are long gone.



^^^^ See what I mean?

FYI, adding anything at all to card makes it a refresh, not a rebrand. Even if all they did was add 50 mhz and 4 gb ram, that would still be a refresh. It looks like they did a bit more than that even, but it doesn't matter.

There are multiple types of members here.

100% Nvidia
100% AMD
Bang for buck
Benchmark freaks
The latest and greatest...Got to have the absolute highest performance available crowd.
Logical decision makers
Green(Tree huggers).... Per/watt trumps fps
Trouble makers...Just push buttons for fun. Not going to buy anyways. Goal is to cause infractions to others
Undecided/Impressionable

Most of the bickering back in forth is more like a chess match. The fud is mostly targeted at the last group as there the ones that can recruited I'll call it.

Reviews can't come soon enough! Looking forward to some in depth details of Figi along with performance figures.

 

[bryanW1995]

That point isn't even relevant though. Flawed or not, that only shows that 6GB vs 12GB doesn't matter. Making any assertions on Fury or 4GB based on that is illogical at best, foolish at worst. The point of the argument is whether 4GB is enough, or 6GB is needed, etc.

I could go back and find literally a hundred posts of AMD fans stating a 8GB model would be announced at launch. Where is that? 4GB or bust. It is what it is. Hopefully the reviews we get soon will truly answer if its enough.

If I see enough evidence that its not an issue (or just in some more uncommon cases) that's my next card. Period.

Making any assumptions that 4GB IS or IS NOT a limiting factor for Rury is impossible right now. We don't know enough and I am surer that NV fanboys will find those edge-cases where maybe it does in specific conditions for a specific title. Thats fine, but is that a true exception, or the norm? I am leaning toward the exception, but again we don't enough yet.

Edit: Even extrapolating on current VRAM usages is tough because of HBM and what AMD may have written in their drivers to assist.

I'm not saying that 4gb vs 6 or 12gb is unimportant. It very well could be, enough so that people might flock to 980ti in droves even if it has 10% worse typical performance. I'm just saying that [H] said, specifically in those games, that you needed 6gb to play them, in spite of the evidence from their own review to the contrary. For you to say that what I wrote isn't relevant is ridiculous, for RS was pointing out a logical fallacy from a (supposedly) professional review site. I've had issues in the past with a lot of other sites, but logical fallacies like that wouldn't happen at a reputable site (at least not one that had earned a positive reputation).

There are multiple types of members here.

100% Nvidia
100% AMD
Bang for buck
Benchmark freaks
Logical decision makers
Green.... Per/watt trumps fps
Trouble makers....These guys just push buttons for fun. Not going to buy anyways.
Undecided/Impressionable

Most of the bickering back in forth is more like a chess match. The fud is mostly targeted at the last group as there the ones that can recruited I'll call it.

Reviews can't come soon enough! Looking forward to some in depth details of Figi along with performance figures.

Perhaps you should change "Green" to "Environmentally Friendly"...it looks like Red might be a bit more Green this time.

 

[el etro]

What is right is right.

Very much probably you will hit the vram wall at resolutions higher than 1600p, with 4 gigs of vRAM. But this will not happen in every case.

 

[2is]

snip

AA requires processing power too, it's not merely a VRAM hit so simply upping or lowering AA to gauge performance isn't sufficient. Your performance will go down when you enable AA even if you're not using up all your VRAM.

So unless you have a reliable way around this "issue" you're complaining about, you really shouldn't be crying that reviewers are documenting vram usage.

 

[2is]

Play/bench the game and if it's a stuttering mess and reducing the texture quality or AA returns it to smoothness then it "needs" that amount of VRAM. If it uses for example 5.5GB on a GTX 980ti but the relative framerates/frametimes are still good on a GTX 980 at the same settings then it's just caching assets it can and doesn't "need" that much VRAM. That would be my guess anyway.

That's great for cards that ACTUALLY run out of vram, what about cards that don't? If you have a 12GB titan and you're using 6GB how are you going to determine that it doesn't "NEED" to use that 6GB it's using? The complaints are stemming from that type of reporting. Bottom line is you can't make any sort of reliable determination with the tools we have at our disposal, not without swapping GPU's and even then you'd need one that has a similar performing GPU just with less ram to get any sort of reliable data.

The question isn't "how can I tell if I'm running out of vram" the question is, how do you know the game NEEDS as much as it's actually using.

 

[jackstar7]

AA requires processing power too, it's not merely a VRAM hit so simply upping or lowering AA to gauge performance isn't sufficient. Your performance will go down when you enable AA even if you're not using up all your VRAM.

So unless you have a reliable way around this "issue" you're complaining about, you really shouldn't be crying that reviewers are documenting vram usage.

I think you didn't read what he wrote. Maybe I'm wrong...

He showed non-AA performace, then AA performance all with the same cards. The hits should be consistent across the board, but are not. They align more with VRAM limits.

 

[Grooveriding]

RS nailed it.

When you actually run out of VRAM your game comes to a full stop and you see sub 5 fps while information is churned from the disk into VRAM. Memory usage measurements in monitoring software are worthless in representing what you actually need before you run out of VRAM. I think most users don't know that because it rarely happens that you literally run out of VRAM. I experienced it on my 1GB 5870s in Crysis at 2560x1600 and again in Shadow of Mordor at the same resolution with my 3GB 780tis.

I'm sure this will be a hot talking point and you can be sure nvidia will want to make it one, but I think they'll be hard pressed to find any instances where 4GB is not enough. I never saw the 980 or 290X run out of VRAM in any benchmarks at 4K. You can do it for sure, but you'll need to push some insane settings to show it happening, at which point the logical question will be why does that matter when you're running at 15fps on the 6GB card and 5fps on the 4GB card. I'm sure pcper will put out a nice bench of GTA V at 4K with 8xMSAA and supersampling if necessary to try and make 4GB look inadequate. Running at 15fps!

I'll be buying whichever is faster, the VRAM is a non issue.

 

[Gikaseixas]

None of the games they showed to use > 4GB of VRAM require that amount to run which is why 295X2/980 SLI outperform Titan X at 4K in each of those titles. Therefore, their conclusion that 4GB holds back certain cards is flawed.

Great post as usuall. GTX 980 SLI or 295x2 do outperform Titan X primarily due to GPU horsepower at their disposal

 

[Gikaseixas]

I'm sure this will be a hot talking point and you can be sure nvidia will want to make it one

I think they rallied their troops already

 

[xthetenth]

That's great for cards that ACTUALLY run out of vram, what about cards that don't? If you have a 12GB titan and you're using 6GB how are you going to determine that it doesn't "NEED" to use that 6GB it's using? The complaints are stemming from that type of reporting. Bottom line is you can't make any sort of reliable determination with the tools we have at our disposal, not without swapping GPU's and even then you'd need one that has a similar performing GPU just with less ram to get any sort of reliable data.

The question isn't "how can I tell if I'm running out of vram" the question is, how do you know the game NEEDS as much as it's actually using.

It's very hard to show how much VRAM isn't needed, but it's very easy to show that there is some VRAM being used that isn't needed. Does your framerate tank a dive that can be fixed by lowering the memory consuming settings? If not, you haven't run out of memory! :thumbsup:

So grab that 295X or those 980s, SLI em up, and show that 4 GB isn't small enough to run out of memory!

 

[Headfoot]

RS nailed it.

When you actually run out of VRAM your game comes to a full stop and you see sub 5 fps while information is churned from the disk into VRAM. Memory usage measurements in monitoring software are worthless in representing what you actually need before you run out of VRAM. I think most users don't know that because it rarely happens that you literally run out of VRAM. I experienced it on my 1GB 5870s in Crysis at 2560x1600 and again in Shadow of Mordor at the same resolution with my 3GB 780tis.

I'm sure this will be a hot talking point and you can be sure nvidia will want to make it one, but I think they'll be hard pressed to find any instances where 4GB is not enough. I never saw the 980 or 290X run out of VRAM in any benchmarks at 4K. You can do it for sure, but you'll need to push some insane settings to show it happening, at which point the logical question will be why does that matter when you're running at 15fps on the 6GB card and 5fps on the 4GB card. I'm sure pcper will put out a nice bench of GTA V at 4K with 8xMSAA and supersampling if necessary to try and make 4GB look inadequate. Running at 15fps!

I'll be buying whichever is faster, the VRAM is a non issue.

Perfect post. This is exactly right.

 

[jackstar7]

I'm sure pcper will put out a nice bench of GTA V at 4K with 8xMSAA and supersampling if necessary to try and make 4GB look inadequate. Running at 15fps!

And the graphs will all be done in %'s so there are no actual FPS numbers to compare.

But don't let them hear us. They've been over on [H] lately defending themselves from the bias accusations. Makes for interesting* reading.



*Depending entirely on how much time you have and how much you value said time.



EDIT: Since I was asked by someone who cannot receive PM's back...(sigh)... here is the thread at [H] to which I referred:
http://hardforum.com/showthread.php?t=1865193

Poster: Vigile says he is Ryan Shrout of pcper and discusses things/defends himself.

 

[IEC]

RS nailed it.

When you actually run out of VRAM your game comes to a full stop and you see sub 5 fps while information is churned from the disk into VRAM. Memory usage measurements in monitoring software are worthless in representing what you actually need before you run out of VRAM. I think most users don't know that because it rarely happens that you literally run out of VRAM. I experienced it on my 1GB 5870s in Crysis at 2560x1600 and again in Shadow of Mordor at the same resolution with my 3GB 780tis.

I'm sure this will be a hot talking point and you can be sure nvidia will want to make it one, but I think they'll be hard pressed to find any instances where 4GB is not enough. I never saw the 980 or 290X run out of VRAM in any benchmarks at 4K. You can do it for sure, but you'll need to push some insane settings to show it happening, at which point the logical question will be why does that matter when you're running at 15fps on the 6GB card and 5fps on the 4GB card. I'm sure pcper will put out a nice bench of GTA V at 4K with 8xMSAA and supersampling if necessary to try and make 4GB look inadequate. Running at 15fps!

I'll be buying whichever is faster, the VRAM is a non issue.

Same here. If it's +/- 5% I'll favor dual Fury X simply because the stock AIO CLC is a big plus at the same pricepoint. Otherwise I'd be tempted to go dual 980 Ti.

And even then, I'm not going to be using MSAA @ 4K... too big a performance hit in many demanding games, and I'd rather not be playing a 5 FPS slideshow

 

[2is]

It's very hard to show how much VRAM isn't needed, but it's very easy to show that there is some VRAM being used that isn't needed. Does your framerate tank a dive that can be fixed by lowering the memory consuming settings? If not, you haven't run out of memory! :thumbsup:

So grab that 295X or those 980s, SLI em up, and show that 4 GB isn't small enough to run out of memory!

Still doesn't answer the question. If I have a 12GB and a game is using 5GB of VRAM and my performance is perfectly fine, how do you know a 4GB card will be enough? That's the question. THe only way you would know is to have a card with 4GB of vram that has a GPU that's similar in performance. Like a 980Ti and Titan X or even more similar than that, an 8GB 290x and a 4GB 290x. But that isn't practical.

So again, unless the people complaining about VRAM reporting in reviews have a way around this, they really shouldn't be complaining.

Also, when you run out of VRAM your performance doesn't instantly go to 5fps. There are a lot of factors. For one, it's not streaming from your disk to the card, it's going to stream from main memory. Second, a HUGE factor in just how much your performance drops is how much data is needing to get swapped from your GPU to main memory. If you have 2GB vram and need 2.5 for your settings you aren't going to see as big of a hit as having 2GB but needing 4GB. Not that this matters anyway, because again, that isn't really answering the question being asked but just wanted to point it out.

 

[happy medium]

If you have a 4k setup don't you usually have 2 or 3 cards anyway?

So the question is can you keep about 40 playable fps with 2 or 3 high end cards at settings that will tank 4gb of memory?
An enthusiast gamer with a high end 4k rig doesn't just have 1 or 2 gtx980's right?

 

[3DVagabond]

If you have a 4k setup don't you usually have 2 or 3 cards anyway?

So the question is can you keep about 40 playable fps with 2 or 3 high end cards at settings that will tank 4gb of memory?
An enthusiast gamer with a high end 4k rig doesn't just have 1 or 2 gtx980's right?

Why 40fps? How about 60fps? Isn't that where we want to be? 40fps avg. is minimum acceptable.

 

[2is]

There are 4K reviews out there. 4GB does not appear to be a limiting factor in today's games. The games of tomorrow however... That's the question.

 

[happy medium]

Why 40fps? How about 60fps? Isn't that where we want to be? 40fps avg. is minimum acceptable.

I'm sure most games dip into the thirties every once in a while ,so lets say 35 min, 48 avg, 65max. That is certainly playable.

I'll fix it.
quote"
If you have a 4k setup don't you usually have 2 or 3 cards anyway?

So the question is can you keep about 35 min, 48 avg.,fps with 2 or 3 high end cards at settings that will tank 4gb of memory?
An enthusiast gamer with a high end 4k rig doesn't just have 1 or 2 gtx980's right?

 

[chimaxi83]

There are 4K reviews out there. 4GB does not appear to be a limiting factor in today's games. The games of tomorrow however... That's the question.

Well in that case, why buy anything, ever? The "games of tomorrow" will always increase in requirements. That goes for pretty much anything.

 

[2is]

Well in that case, why buy anything, ever? The "games of tomorrow" will always increase in requirements. That goes for pretty much anything.

That isn't the case at all. I'm not talking about 5 years from now. With DX12 in everyone's hands in July, who knows how much more complex games are going to get. VRAM requirements could easily rise in a very short period of time. We already have >2GB NEEDED even at 1080p. If you feel comfortable that 4GB is going to be enough before you decide to buy your next GPU then you have nothing to concern yourself with. I'm not that confidant.

I'm sure you've heard all these arguments before, so not sure why you're asking the same questions again. Nobodys mind is going to change here.