• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

One it's 10Gbps memory, meaning 320GB/s on a 256bit bus. 14Gbps is the highest speed they aim to achieve over time, not the speeds available from the first chips.

As for the 980ti stuff, what on earth, GDDR5x isn't a swap in replacement, it's not compatible with gddr5, it's pin compatible but the memory controller has to support it. 980ti has zero support for GDDR5x and again even if it did again with the chips actually available it would be 480GB/s, I have absolutely no idea at all where you came up with 886GB/s. The 980ti has a 384bit bus, divide by 8(bits in a byte) and times by the speed of the memory. 14Gbps, the stuff that isn't even close to available yet not planned any time in the next year, you would still only achieve 672GB/s.

Meant 386 typing on my phone isn't great


Very aware 5X isn't compatible, just looking at the raw numbers , 5X won't be the wonder ram some people think it will be. It's nit the bottleneck @4k anyway
 

Actually yes.

I am amazed GTX 970 is still the best GPU's ever made today, it still broke sales record. It is number 1 accorded to new Steam survey March 2016 and I break down total owner shares below:

166,102,006 active steam users

http://steamspy.com/#tab-graphs

Geforce 900 series

NVIDIA GeForce GTX 970 9.52% 15,812,910 owners
NVIDIA GeForce GTX 960 6.04% 10,032,561 owners
NVIDIA GeForce GTX 980 2.00% 3,322,040 owners
NVIDIA GeForce GTX 980 Ti 1.72% 2,856,954 owners
NVIDIA GeForce GTX 950M 1.02% 1,694,240 owners
NVIDIA GeForce GTX 970M 1.00% 1,661,020 owners
NVIDIA GeForce GTX 950 0.96% 1,594,579 owners
NVIDIA GeForce 940M 0.82% 1,362,036 owners
NVIDIA GeForce 920A 0.74% 1,229,154 owners
NVIDIA GeForce GTX 980M 0.52% 863,730 owners
NVIDIA GeForce GTX 965M 0.22% 365,424 owners
NVIDIA GeForce 930M 0.20% 332,204 owners
NVIDIA GeForce GTX TITAN X 0.16% 265,763 owners

Radeon Fury series

AMD Radeon R9 Fury 0.08% 132,881 owners

Very strange here was no R9 300 series in survey, AMD really need to naming each card properly in driver detection. I recalled a year after 780 Ti on sales, here were around 1 million 780 Ti owners at peak so £500 980 Ti had broke all sales record with nearly 3 million owners massive outsold Fury series all combined nearly 21 times. :eek:

Edit: I checked Steam survey twice, don't realised R9 360 is on list.

Radeon 300 series

AMD Radeon R7 360 0.24% 398,644 owners
 
Last edited:
Very aware 5X isn't compatible, just looking at the raw numbers , 5X won't be the wonder ram some people think it will be. It's nit the bottleneck @4k anyway

GDDR5 and 5X still have a much lower latency than HBM1. No hard latency figures for HBM2 yet, but I'll bet they're still not as good as GDDR5/5X

It'll be interesting to see how latency affects the new gpu's especially with the compute stuff.
 
snip! random stuff

What hilarious is that using your metrics,does that make my card,a GTX960 the second best card ever made?? :D

Woot! I have the SECOND BEST CARD EVER MADE!! :D

You are obviously somebody who has no clue about "best". The GTX970 is starting to look meh in more and more games - so in three years loads of people will running them fine??

If you really want to look at sales figures - want to look at the 8000 series cards??

I suspect the 8400GT and the rebadged versions have sold TENS of millions of cards in a bigger market.

Does that make them the BEST CARDS EVER?? :p:p

Will it be the new 8800GT,8800GTX or 9700 PRO?? They lasted for years.


As an owner of a 8800GTS 512MB and having owned a 9500 PRO,no I don't think it will be the best. The GTX970 and GTX960 are doing worse and worse in more games - the 8800GT was relevant for generations,just like the 8800GTX and 9700 PRO.

It just shows you how much modern enthusiasts,have such low standards. Really sad.

No wonder consoles like the PS4 have record breaking sales.
 
Last edited:
GDDR5 and 5X still have a much lower latency than HBM1. No hard latency figures for HBM2 yet, but I'll bet they're still not as good as GDDR5/5X

It'll be interesting to see how latency affects the new gpu's especially with the compute stuff.

Hmm, pretty strong call to say ggdr5x has 'much' lower latency than HBM1 with no hard figures on HBm1 or 2, nor GDDR5x for that matter which likely doubles latency or near enough compared to GDDR5.

http://www.hardwareluxx.com/index.p...han-just-an-increase-in-memory-bandwidth.html

This is a fairly good break down on the differences. Latency alone is a relatively meaningless measure. GPU's have always sacrificed latency for bandwidth because latency sensitivity is not hugely important. For system memory both the higher clock speeds of a CPU and the less predictable nature of what will run next means mis-predicting instructions leads to a direct wait for the next new instruction coming from memory so the lowest latency possible has trumped overall bandwidth in so many circumstances.

As the link above addresses, HBM access and usage is really quite dramatically different from GDDR5, a direct latency comparison is hard to make and not that important. GPUs queue up memory access in advance and know mostly what they'll need and when so overall bandwidth is really only what matters. What is important is memory efficiency, how the memory is accessed significantly effects efficiency. HBM should be more efficient over time though there are indications that at least Fury has a pretty inefficient memory controller. There are many possible reasons for this, for one thing a gen 1 HBM controller will absolutely be less efficient than a lets say gen 5 HBM controller. It's newer and less efficient but after several iterations(maybe only one more) it should be more efficient than GDDR5 access. By efficiency I mean while theoretical bandwidth is 512GB/s, effective bandwidth could be anything, 256GB/s or 480GB/s. Another issue is simply that GCN 1-1.2 wasn't designed for HBM so the inefficiency could be not on the memory controller but the internal connections of the GPU.

An architecture designed for GDDR5 with a max realistic bandwidth of about 350GB/s during the architectures lifespan won't waste internal wiring and communication channels to support 512GB/s of communication. So maybe it can draw 480GB/s effective from the memory but it's getting bogged down on the GPU side of the memory controller.

There is every reason to believe that with a fully designed for HBM architecture effective bandwidth and effective latency will be lower than GDDR5, not higher.
 
Nothing really said there that we have not heard in rumors and then adding more rumors (his opinions), total click bait.
The phrase 'click bait' has really been watered down and overused lately. He makes a pretty clear distinction between what is rumor and what isn't, which is more than most sites do trying to attract attention with bold headlines and article content.

And not everybody stays super up-to-date on the latest GPU gossip. This is more of an overview of what we might expect. Just because he hasn't added anything noteworthy for your own knowledge consumption doesn't make this 'click bait'. It might well be informative for others.
 
Last edited:
Back
Top Bottom