A long time ago we used to think of video memory as just the framebuffer. But now, even with triple buffering at max res, the amount needed to hold frames ready for display is trivial compared to what is being used for other things.
If HBM has lower latency it will probably help reduce instances of immersion disruption caused by "pop in".
along with those lower leaks and lower power consumption at GF you get lower yields.... :whiste:
The way I see things is that if you have a more unique or esoteric set up, stay away from Nvidia as they don't really follow the design philosophy PC hardware where different components are supposed to play nice together.
I have a fairly unique setup, with HDMI audio running out of Intel integrated graphics (Nvidia's HDMI audio support is so bad it may as well be non-existent) and triple monitor + Oculus Rift running out of a GTX 980. I had more problems in the first month of Nvidia ownership than the past three generations of AMD cards I ran. Nvidia drivers even bricked my Windows installation once. Almost every option from the Nvidia control panel will periodically disappear until I disconnect and reconnect my multi-monitor array. It's frankly hellish.
If your setup is pretty bog standard, then Nvidia has the advantage of putting in more effort into their drivers. If you're doing anything unique, or anything with multiple monitors, then AMD is the massively better choice. Their philosophy is to work with other IHVs, Nvidia lacks this entirely in their corporate culture.
I can`t help but wonder if these slides is actually 100% legit.
"GM200 Cut" is GTX 980 Ti and is 10% slower than Titan X which makes sense since we learned that GTX 980Ti got 22SMM while Titan X got 24SMM.
The slide also got the HBM capacity of 4GB correct which we know Fiji XT got, plus they call it Fiji XT and not 390X which also may be correct since Fiji XT is not 390X but a named card like Titan.
What is more shocking is that it means that Fiji XT only draw 33W more than Titan X, which means there got to be a lot of improvement on the chip vs previously R9 200 cards. Fiji XT draw less than 290X despite having 45% more cores.
HBM make up for 15-20W of that, the rest must be from architecture and process.
No way these can be right, they show the 780TI beating 970 and the 290x.:sneaky:Sarcasm Alert
It is shocking how that slide is becoming more accurate as time goes on.
Well 780Ti is neck in neck with 290X @ 4K plus it does beat 970 there as well (Source).
I think I may lead towards the Fiji XT this time. The card looks stunning from what I`ve seen so far, HBM doesnt seem to help it against TitanX/980Ti in 4K resolutions though but neither does the 4GB seem to limit it.
I need to see temperature measurements and noise of this baby, but if they got that pinned down, AMD deserve my support.
There are other stuff like recent TressFX issue with Witcher3 and very few driver updates vs Nvidia`s countless driver updates which is nagging in the back of my head to go team green though.
I`m very conflicted at this stage
I prefer high quality quarterly stable driver updates, and maybe as many betas as possible.
Besides why do nvidia need to constantly update their drivers, why do they need to be constantly fixed? /sarc
On the other hand, I guess Fiji XT is just a little more efficient than 290X with power when Fiji XT is 300W TDP vs 290X`s 290W TDP.
So it may come down to:
AMD Fiji XT 4GB - $849 - 300W
GTX Titan X 12GB - $999 - 250W
GTX 980 Ti 6GB - $800 (-10% in performance) - 230W
decisions decisions :hmm:
You guys are seriously finding excuses for everything.Their latest "game ready" drivers messed up other games. It even lowered FPS for users who had older drivers before. I used to be a huge fan of Nvidia driver update frequency until I started actually reading users having issues.
Do we seriously need new drivers for every game that comes out? This is the kind of stuff that starts to make console gaming look attractive.
Do you think AMD defended 4GB in the recent HBM introduction if it was 8GB?If this card turns out to be 8GB and cheaper. I hope people rub it in hard.
lol noGet both...problem(s) solved!
You guys are seriously finding excuses for everything.
A few driver updates from Nvidia may had some issues, but many many other`s have been flawless with performance boost over previous driver for that specific new game plus they added SLI profiles.
That is something I and many many other people want and why many people buy Nvidia instead of AMD. Things like this is important believe it or not.
What is more shocking is that it means that Fiji XT only draw 33W more than Titan X, which means there got to be a lot of improvement on the chip vs previously R9 200 cards. Fiji XT draw less than 290X despite having 45% more cores.
HBM make up for 15-20W of that, the rest must be from architecture and process.
You guys are seriously finding excuses for everything.
A few driver updates from Nvidia may had some issues, but many many other`s have been flawless with performance boost over previous driver for that specific new game plus they added SLI profiles.
That is something I and many many other people want and why many people buy Nvidia instead of AMD. Things like this is important believe it or not.
Do you think AMD defended 4GB in the recent HBM introduction if it was 8GB?
By all means, I`d love 8GB too but I highly doubt it will happen. We will see
lol no
“You’re not limited in this world [die stacking world] to any number of stacks, but from a capacity point of view, this generation-one HBM, each DRAM is a two-gigabit DRAM, so yeah, if you have four stacks you’re limited to four gigabytes. You could build things with more stacks, you could build things with less stacks. Capacity of the frame buffer is just one of our concerns. There are many things you can do to utilise that capacity better. So if you have four stacks you’re limited to four [gigabytes], but we don’t really view that as a performance limitation from an AMD perspective.”
Why not buy both so you can enjoy brand exclusive games like nvidias upcoming batman title, or AMDs Star Wars battlefield (which would work good on both...sigh).
Also, on ChipHell there were other leaks showing a card that was 20% faster than GTX980 and was drawing around 185W of power.
You guys are seriously finding excuses for everything.
A few driver updates from Nvidia may had some issues, but many many other`s have been flawless with performance boost over previous driver for that specific new game plus they added SLI profiles.
That is something I and many many other people want and why many people buy Nvidia instead of AMD. Things like this is important believe it or not.
Do you think AMD defended 4GB in the recent HBM introduction if it was 8GB?
By all means, I`d love 8GB too but I highly doubt it will happen. We will see
lol no
The way HBM achieves its astounding bandwidth is by keeping all the interconnects in the silicon interposer. For more than one GPU chip to share memory at the expected bandwidth, all GPUs and memory chips would have to be mounted on a single interposer.
I have an odd thought regarding the high bandwidth of the 390X with HBM. Remember awhile back when AMD was talking about working on Unified Memory architecture? Is it possible that there is enough bandwidth available with the 390X that they could conceivably have UM when running CrossFire? Then two 4GB 390X's would have 8GB worth of usable VRAM. More than enough for the resolutions you'd need two GPUs for. Add a third GPU and you'd match a Titan X in VRAM and tremendously surpass it in processing power. Each tier (1 card at 4GB, 2 cards at 8GB, 3 cards at 12GB) would have enough memory to support the resolutions it would be purchased for and producing the 4GB 390X with 4GB of memory makes perfect sense.
The above is probably not possible, but would be an incredible game changer for AMD if it could be.