yeah funny indeed... but NVIDIA is telling/pressure Oxide to disable async compute/shader feature on the bench.

" The rather startling news is that Nvidia's Maxwell architecture, and yeah that would be the entire 900 range does not support it in the way AMD does. I can think of numerous scenarios as to where asynchronous shaders would help."

http://www.guru3d.com/news-story/nvidia-wa...n-settings.html

add-on: oh bro your signature correction on "MSI NVIDIA GeForce GTX970 Gaming 4GB GDDR5" is 3.5GB GDDR5

This post has been edited by Unseen83: Sep 1 2015, 10:20 PM

Huh? You mean the GTX970 has physically only 3.5GB GDDR5? Ok...thanks for info

This post has been edited by Demonic Wrath: Sep 1 2015, 11:22 PM

I've seen some of these for the past months since the GTX970's "RAMgate"...

It has 4GB GDDR5,but 3.5GB is fast,0.5GB is slow.

No need to make a big problem out of it.

Wait until you tell him that GTX970 has only 1.75 MB L2 Cache and 56 ROP instead of marketed 2MB and 64 ROP.

Because of 970 memory segmentation, the GPU VRAM bandwidth is effectively 196GB/s due to 224bit bus width.
Originally it is advertised to have 224GB/s and 256bit bus width.

oh so total 4GB but on separate partition/speed 3.5GB on faster or ads speed meanwhile 0.5gb on slow or half ads speed okay thanx for that info

naah i was wrong.. gtx 970 have total 4gb gddr5

really.. wow.. i did not know that also.. hmm imagine i almost go for GTX970 Sli in exchange R9 290x crx

Lucky for you, you should know the weird frame time issue when GTX970 are SLI-ed..........?

http://www.pcper.com/reviews/Graphics-Card...sues-Tested-SLI

While GTX970 can use and address the 0.5GB (32bit) just fine, but using it will block the access to the faster 3.5GB(224bit).
If program reads 3.5gb (default drivers behavior is to prioritize this 224bit segment), then it cant read 0.5gb and vice versa for writing operation.

It can only access both partitions if first partition is reading while the second partition is writing.

To expensive and GM200 is not best overclocker card. Buy reference GTX 980ti+G10+H55 is enough if u dont like aftermarket cooler. Can get 55-60 celcius max OC load

This post has been edited by crash123: Sep 2 2015, 12:36 AM

looking into this async compute issue.... regarding how nvidia cards would perform in dx12 games has got me thinking twice about pascal.

i'm just gonna have to wait and see what the reviewers say on that matter.

when it comes out and they test on ash singularity or another newer dx12 game, then can know whether pascal will last me a few years (with being able to play dx12 games well in mind) or not

if not then choice would be go for amd (which apparently was not a total flop because they actually did async compute the right way despite being a power guzzler), whereas nvidia probly did some emulation method that could do it albeit poorly

nvidia telling ash to disable async compute ... does not inspire much confidence

This post has been edited by Moogle Stiltzkin: Sep 2 2015, 03:23 AM

dammit i buying console after this no DX12 racing games will be out soon hahahaha

Since both all graphics hardware vendor doesn't have a card that can fully support DX12, just wait till 2016 or 2017 to get a full DX12 feature support card.. but then again, maybe that time got DX12.1 feature?

Edit: For those who really needs full DX12 support, just wait till DX12 games are released first then only see benchmarks..

This post has been edited by Demonic Wrath: Sep 2 2015, 09:35 AM

but but.. AMD never Claim to fully Support DX12, meanwhile Nvidio proudly claim it's Maxwell FULLY DX12 support" but like you said.. Nvidia own 82% market so got all money so maybe they can pay off game dev to delay using dx12.. while they conjure up pascal, but at the end is about one dishonest company fool its customer/fan...

https://www.reddit.com/r/nvidia/comments/3j...nvidia_cant_do/


:/

dx11 nvidia clearly wins, but come dx12 with all sorts of performance tricks you can do, this may change needs closer look when benchmarks are out. but judging by all the technical mumbo jumbo, it's not looking good for 980ti

sure hardly any dx12 games out yet, but most people buy a card to last 4 maybe 5 years. so you wouldn't want to change out your card lesser than that just to play a dx12 game as well it should have been

maybe the performance penalty isn't too bad, or better yet pascal fixes this ? so have to wait for benchmarks

This post has been edited by Moogle Stiltzkin: Sep 2 2015, 11:07 AM

980ti only?

all 9 series i guess.

i saw ur post so many times, ini kali-lah ur statement, i fully agreed with u. NV purposely pay them to delay. i believe pascal is fully support DX12 but for maxwell users lain kali-lah.

maxwell is partially support for testing purpose on DX12 only. this is a game goin release soon which is favor AMD but look at NV performance really disaster.

i'm curious 980Ti can hit 6GB vram at the peak???

i remembered NV TOM claimed above 970 is fully support DX12

not sure about others. only cared about the best single gpu nvidia has atm

even with this dx12 concern when it comes down to async compute which a game like rts genre that spawns many units like AOTS was demonstrating how rts could be like by increasing efficiency/performance via dx12 is possible. the gains from async compute if the developer uses it, seems to be quite good on paper :]

will come down to fps and latency, whether the difference is huge to warrant switching camps






http://www.pcper.com/reviews/Graphics-Card...abs_block_tab=1


from the benchmark it just clearly shows the amd cards have huge jumps in performance in dx12 compared to dx11. so although the difference in end result is by a small margin between the 2 cards, it brings up the question, could nvidia performance had been much better had they used better async compute like amd ? nvidia async compute now being scrutinized for being inferior to amds judging by this result.

might be that in pascal the dx12 leveraging would be the same as maxwell, and this is what people waiting to be answered :/

but if the results is more or less same, dx 11 will be better than amd, and dx12 only slightly worse than amd, so nvidia may still be a better choice albeit disappointing they didn't fully leverage on dx12 async compute (cause the benchmark shows it does make a difference, if a game decided to use it e.g. rts especially).

for VR, people may prefer amd for the lower latency compared to nvidia. not sure the latency will be that big to be a problem for non vr games in regards to async compute

From AoTS benchmark, it seems that NVIDIA's DX11 without async support performs faster/similarly to AMD's DX12 with async compute. (source) edit: see frame rate by batch type.

Secondly, Anandtech (source) previously show that NVIDIA does benefit from DX12 implementation too. StarSwarm is also developed by Oxide.

This post has been edited by Demonic Wrath: Sep 2 2015, 12:35 PM

or it could be lie from Oxide and AMD as they working together to bring down Nvidia... is like how Nvidia told game dev add gamework that cripple gpu performance hmm ..



https://www.youtube.com/results?search_query=FaWbDpEuuk

This post has been edited by Unseen83: Sep 4 2015, 05:47 PM

nvm i still enjoy playing DX11 games. too many can't finish at all..