AMD Radeon Fury X beats Titan X in leaked CompuBench OpenCL benchmark

I wasn't aware AMD had ever "left the grid"... If you for instance mean the "$1500 grid," well, I'd prefer it if AMD never gets on that grid. Hmmmm.... (Benchmarks & rumors...the domain of the gullible, eh?)

 

Don't think I've ever heard about " HBM " before, how is it different from GDDR5?

Code:

Graphics memory 	4 GB HBM

EDIT: oh, decided to check up on it since I hadn't heard about HBM before. Supposedly its 9X Faster Than GDDR5

http://www.amd.com/en-us/innovations/software-technologies/hbm & http://hothardware.com/news/amd-details-high-bandwidth-memory-hbm-interface-technology

 

Welcome back to life lol


http://en.wikipedia.org/wiki/High_Bandwidth_Memory

 

hehe, yeah. I should probably have just googled it before asking if I gotta be honest. Did do it though :nerd:

 

I mean, I guess, if you're going to start pulling numbers from different sites and comparing them. 780Ti isn't even on that slide. The 970 is, which is roughly the same performance, and also ~45% which still makes your post wrong.

It's twice as fast @ half the power. It's like the main selling point behind Fiji.

 

Interesting - Fury X with 4GB HBM (just) beats Titan X with its 12GB GDDR5 @ 4k res Ultra detail in those leaked Firestrike benchies

Point being it seems AMD is correct in claiming it has made massive improvements in the efficiency of its use of available video memory. So 4GB HBM could well be fine for one card 4K use.

Feeling better about this card already!

 

There will be that improved memory compression thing that could make it act more like a 5GB card?

But come on AMD and just release already. The wait.. my god, the wait. I can't take much more now!

 

Zz.. hope it has an ok price. 600€ MAX. Not paying a cent more than that.

 

I recon it's going to be mixed blows between Titan and Fury, application dependent. Games that adore memory bandwidth will mostly be where it excels, provided the 4GB is sufficient.

 

Yes, It's either that or I will have to look at other alternatives.

Hope they make the right respectful move and offer more performance for less price as they have done in the past.

 

Agreed, as they did with the 7950, amazing product, best ocability, best performance, best value.

 

Remember unlocking those Sapphire 6950's for the same price of a single GTX 580 configuring them in to crossfire , whilst seeing the same numbers in crysis as two 580 or very close. That made me feel good about that buy

 

Man... Epic clueless comment LOL! GDDR5 and what amd calls HBM store data differently, or rather address memory differently. AMD claims that HBM, this new memory technology that they mostly developed by themselves, is more efficient space-wise than GDDR5, which was also developed by AMD. The impact HBM is gonna have on performance is going to be mostly due to its lower access latencies compared to GDDR5.


While I don't want to make any performance estimates like you seem to be doing, I think Fury will beat the Titan X by a bit. It will probably beat it in Tessellation too (polygon count generated per second benchmark) and that is impressive for a chip that is around 66mm^2 smaller.

Now...if AMD acts smart and prices Fury cards properly, then they win this round, big time!

 

Dude im talking about the generational leap here by comparing the flagship Gpu of both the companies over the two generations not that how was the performance over a single generation.

 

You really don't know if they are comparing air-to-air or hybrid-to-hybrid so doesn't mean much right now. If this is a Fury X hybrid then comparison should be done against a card like the Evga Titan X hybrid if trying to determine performance differences.

 

May not be in full use yet or not at all, but has not the current buzz been about streaming textures instead of storing it in the buffer? Would this not be a reasonable assumption as to why 4GB HBM is all that is needed? They are just too far ahead of the curve again. Timing is off as usual.

 

You said a Fury X is ~45% faster then a 290x and that a 980Ti is only ~30% faster then a 780Ti. You said this as if generational gaps in performance is some kind of indicator of the final worth of a card(?). Anyway, you then told me to go to WCCFTech and look at the benchmark there. I did.

Fury X - 7873/3960
290 X - 5091/2617

42.9% Difference @ Extreme, 40.83% at Ultra

The 780Ti is not on the picture, but looking at other benchmarks show it's roughly the same as a 970, so I'm going to use those numbers...

(http://i.imgur.com/de8zTvf.jpg) 6.7% difference so I'll just subtract that from final result, despite the fact that newest drivers actually put the 970 slightly ahead.

980 Ti - 7781/3867
970 - 4928/2058

38.1% difference @ Extreme, 54.3% @ Ultra (6.7% removed from each)

So essentially the gap between generations is roughly the same between both companies, which again, is not what you said.

One of the companies, i think it was PC Perspective, interviewed AMD about the 4GB issue. Basically they said they threw a few engineers at the problem and found that video memory is stored incredibly inefficiently at both the OS and the driver level. That no one really looked at it before because it was easy to just throw more ram at the problem instead of fixing it. So I'm assuming they are going to have driver level fixes for ram management in order to combat the 4GB problem. Honestly the number of games that exceed 4GB of ram, even at 4K is overstated on this forum. It's not many at all. I doubt Fiji is going to run into issues with that with the current generation of games.

 

ok so u actually admitt that amd is now on par with nvidia??

 

Its not really compressing textures any more to get more in memory, its compressing how they access and stream the texture information, to and from the GPU/Memory. There is also some extra compression possible for memory based non-texture data, which i suppose would free up some more space for textures.
All of the compressed data can be accessed and processed without being decompressed also.

 

http://techreport.com/review/28294/amd-high-bandwidth-memory-explained/2

Sounds like what they are planning to do is definitely going to help improve capacity problems.

I never said that they weren't.