390X coming soon few weeks

again, that doesnt mean its all being used.

Just like GPU usage, at certain times of a game all will not be used and at other times it will hit the projected limit.
Theories or facts in some cases that we are carrying over from 2013 and 2014 are slowly shifting.

In no way my gaming experience is greatly affected by VRAM, not by a mile.
Games I bought last year run near their max settings, but I don't have the full quality of the games I bought this year enabled like in GTA V and Project Cars.

I remember a guy telling me dual core was all that was needed for gaming before Sandy was released.
Last year I said 3GB was enough for high-end gaming. We were both wrong.

I like how PC gaming always goes beyond the limits we try to set for it.
 
Just like GPU usage, at certain times of a game all will not be used and at other times it will hit the projected limit.
Theories or facts in some cases that we are carrying over from 2013 and 2014 are slowly shifting.

In no way my gaming experience is greatly affected by VRAM, not by a mile.
Games I bought last year run near their max settings, but I don't have the full quality of the games I bought this year enabled like in GTA V and Project Cars.

I remember a guy telling me dual core was all that was needed for gaming before Sandy was released.
Last year I said 3GB was enough for high-end gaming. We were both wrong.

I like how PC gaming always goes beyond the limits we try to set for it.

Yeah, I'm not suggesting that 4GB is necessarily enough. It Seems like it might be right now, but who knows.

All I am saying is the fact that someone measured VRAM utilization above 4GB doesn't in and of itself mean that 4GB is insufficient.

The only way to know for sure is testing.

We obviously don't have any tests for Fiji XT yet, but judging by how well 4GB Geforce 980's perform in SLI in GTA V at 4k, I'm thinking we are probably OK with 4GB for now:



Who knows for how long though.
 
Simply reading memory utilization is not a valid test to show the actual benefits of more memory. This is the same issue as people reading CPU utilization to extrapolate the benefits of multithreaded performance.

However typical benchmark runs are also problematic in that they do not necessarily expose the issue either. Benchmark runs are typically relatively short and rather limited in terms of area and scope of the game.

This also leads the issue of what the qualifying threshold is to actually be considered beneficial? What if only certain segments of the game will run into memory issues, what threshold is it considered a problem? What if stutters due to lack of memory only occur occasionally during game play, what the frequency that is considered acceptable?

GTA V also really is still a last generation game, I don't feel using it (especially as a single sample) can be predicative of any trends going forward.

The greater issue though is products need to be compared as a whole and not piece by piece. While some people may be fixated on certain singular aspects generally speaking the consideration is about the whole product. So if Fiji is limited to 4GB VRAM and ends up being a lower amount than its competitors what is it offering in place of that? Will actually offer a significant performance advantage (as opposed an arguable advantage that simply trades depending on workload)? Price/performance advantage? Something else? The issue will be if Fiji only offers other marginal advantages (if at all) but a clear disadvantage in this area. Simply having it be limited to 4GB by itself is not really indicative of anything. Hypothetically if it comes in much cheaper and the same performance but has less VRAM that is fine (its a valid tradeoff), if it comes at roughly the same price and performance with less VRAM than that would be an issue.
 
oh boy!

n6hWmEk.jpg

Most of them are already in their warehouse, only difference is the first digit is a 2....
 
I assume you were tyring ot be funny, but that would mean they would have to rebadge them themselves.

It doesnt work that way.
 
 
Now we get into the length of ownership question.
How long is the user planning to keep these cards? 1 year, 2 years etc.?

Is a ~$700 4GB card a good investment if an upgrade is needed within 18 months?
Maybe a $1000 12GB card for 2-3 years is a better investment, but it's more money upfront.

For me, I would like to skip the yearly upgrade cycle, one of the reasons why I went X99 for the next 3-5 years.
4K will keep me in that cycle if I don't by the top card available right now.
Any GPU less than a Titan X, I will replace next year with a 8GB card guaranteed.

Would you say 8gb of ddr3 is future-proof too?
 
Would you say 8gb of ddr3 is future-proof too?

futureproof in the PC world is like 2 years max.
ssd I bought I expect to be 4 years or so maybe 6.
but normally I find futureproof not really existing in this enviroment.
since die shrinks are slowing down a gpu might last 4 years soon even though I think my sister wouldnt know any difference anyhow.
 
Now we get into the length of ownership question.
How long is the user planning to keep these cards? 1 year, 2 years etc.?

Is a ~$700 4GB card a good investment if an upgrade is needed within 18 months?
Maybe a $1000 12GB card for 2-3 years is a better investment, but it's more money upfront.

For me, I would like to skip the yearly upgrade cycle, one of the reasons why I went X99 for the next 3-5 years.
4K will keep me in that cycle if I don't by the top card available right now.
Any GPU less than a Titan X, I will replace next year with a 8GB card guaranteed.

That is a good question.

In the past I used to never get more than 6 months out of a video card before feeling the upgrade itch.

Now I have 2 years and 3 months on my Titan, something I previously would have thought to be unheard of, and once I'm done with the Titan it might continue in my stepsons 1080p rig (unless it makes more sense to sell it to a number crunching sciency type, and buy him something different instead)

I don't know for sure. I'd guess everyone's situation is different, and it depends on what hits the market next time around.

My best guess would be that this generation that is about to launch will have much shorter longevity than my 2013 Titan, based solely on the fact that they are likely the last gen of 28nm GPU's.

I lucked out and got that Titan at the perfect time for it to last a while. I very much doubt this is a perfect time for a high end GPU, based on the fact that 28nm is a bit long in the tooth.

You never know with these things though. I had buyers remorse the minute after clocking "confirm" when I bought the Titan.

If you had asked me if more than two years later top end GPU's would still be coming out on the 28nm process, I would have laughed at you.
 
Simply reading memory utilization is not a valid test to show the actual benefits of more memory.

No but playing the game like [H] does is valid.


At 4K is where the GeForce GTX 970 SLI struggled, unlike 1440p. Here, the reduced VRAM capability was felt, as the game got choppy and stuttered with higher settings. We had to turn off all of the advanced graphics options on the GTX 970 SLI. These settings eat into VRAM, and we had to disable these to make GTA V smooth. We also had to lower PostFX and Grass setting. We would not recommend the GeForce GTX 970 SLI for 4K gaming in GTA V.
 
Not sure how legit this is but looks like the rad is coming out the back and not from the side... full water block?

original.jpg
 
Last edited:
I think most of us who've been around the block know we're at the cusp of 4GB not being enough. It will be fine if you're playing at 1080p, much like 512mb was more than enough at 1600x1200 when some of us were transitioning to 1080p in 2008. 4GB might be OK in most situations for the next 6 months. However, I plan on holding on to my next card for over a year. I am by no means in the market for a card that will likely have a very hard bottleneck over the next year.
 
You can tell they manufactured these cards before naming it. It just says "RADEON" all over the place. Tacky.
 
You can tell they manufactured these cards before naming it. It just says "RADEON" all over the place. Tacky.

Maybe they are trying to make Radeon a brand name for commercial purposes, I mean geforce gtx is no the side of a lot of Nvidia reference cards.
 
surprised the radiator is out the back instead of top

wonder how it overclocks considering dual 8 pin & assuming VRMs are also being water cooled
 
surprised the radiator is out the back instead of top

wonder how it overclocks considering dual 8 pin & assuming VRMs are also being water cooled
Since the card is so short, there's more clearance in that extra space rather than aligning the tubes with the power connectors. Seems weird at first but it makes more sense when you consider the card's dimensions. Otherwise they risk the tubes blocking your case's side panel.
 
Really is a strange looking card to say the least, sooner we get a pic of the full thing with the rad (preferably in a chassis) the better, right now it just seems weird how its looking in the pics.
 
It's sooo damn tiny.
PC cases have grown so large in recent years, so in 2-3 years from now when HBM is standard, everyone will have these little baby video cards in their rigs.
 
The 'water-cooled edition' uses 8+8 but the air versions are supposed to be 6+8.
Time will tell.

My 280X is 8+8, lol. Strictly for marketing purposes I imagine. Funny thing -- it's voltage locked. I gave Sapphire a stern talking-to via their forums and email, don't think they care.
 
So if we assume the rumored $850 was for the WCE... We take out $50-$100 for whatever custom block they're using on the card, since I don't see any fans. Take out another $50-$100 for the Asetek CLC. That puts the air-cooled Fiji flagship around $650-$700, and the cut-down Fijis around $500-$600. The price difference might be more drastic since the WCE will have an inflated premium just due to being a special edition.
 
So if we assume the rumored $850 was for the WCE... We take out $50-$100 for whatever custom block they're using on the card, since I don't see any fans. Take out another $50-$100 for the Asetek CLC. That puts the air-cooled Fiji flagship around $650-$700, and the cut-down Fijis around $500-$600. The price difference might be more drastic since the WCE will have an inflated premium just due to being a special edition.

Next week we can stop speculating :)
 
Looks as though the card is flipped horizontally as well, so the "face" of the card will be pointing upwards in standard cases.
 
Looks as though the card is flipped horizontally as well, so the "face" of the card will be pointing upwards in standard cases.
The gray plate you see in that picture is normally where fans would be, the PCB is at the bottom of this picture. Lots of people have mistaken that cover as a backplate; wrong side.
 
The gray plate you see in that picture is normally where fans would be, the PCB is at the bottom of this picture. Lots of people have mistaken that cover as a backplate; wrong side.

Thats what i mean, i can't see the face of the card being as blank as that. Reminds me more of a back plate than the spot where we would usually have a fan.
 
Back
Top