blue1512: It seems that the HBMs on FuryX are only locked by driver. It DID be overclocked to 600 MHz.
AMD is releasing a two-phase driver release with Catalyst 15.15 and 15.20. The 15.15 drivers are available immediately, and the 15.20 drivers are coming a few weeks later. The Catalyst 15.15 drivers bring new features like:
Pretty sure the rest of the cards here would also put out horrible power consumptions numbers in furmark. As to why only Fury CF has furmark numbers, I don't know. No one should fall for that.
When comparing the relevant numbers, Fury consumes less than 290/390x while performing much better, up to a 980Ti, sometimes beating a TX depending on what review you see. Can't see anything wrong with that.
Another aspect that damages the credibility of the eTeknix tests is the absence of 980ti SLI, which has been found to have less than optimal scaling due to heat issues.
And what is up with doing Furmark on only one card?
Here’s where things get a little interesting. Running our standard benchmark stress test, which is enough to simulate gaming; produced just 378W. When I checked the cards for usage, the first card was maxed out, but the second was only in about 20% use (judging by the LED indicators by the power connections). I then decided to turn to ol’ faithful and run Furmark. Even though Furmark shows 841W, your day to day use will likely be hugely less.
Ultimately, the test is without real value.
I don't know if it's been said already...
Somewhere I saw the Fury specs mentioning 4 triangles set up per clock, while the top Maxwells do 6 per clock. Digging further I found this nice discussion:
http://www.rage3d.com/board/showthread.php?t=34022743
So it looks like the geometry engine is the bottleneck at lower resolutions. Higher resolutions (esp. 4K) hit the pixel fill rate bottleneck (also lower by 1/3 vs. 980 Ti). Finally some fps graphs had some spikes and most frames longer than 50ms were on the Fury X. This likely means that it has to refetch lots of textures into its RAM stacks. A driver update might help here.
They are not wrong. That's just the difference between measured values and theoretical peak values (ROPs * clock frequency). What we see here is just a different efficiency in reaching that rate (Fury X 59% vs. Titan X 45%).The datas are wrong.
http://www.anandtech.com/bench/GPU15/1242
Here’s where things get a little interesting. Running our standard benchmark stress test, which is enough to simulate gaming; produced just 378W. When I checked the cards for usage, the first card was maxed out, but the second was only in about 20% use (judging by the LED indicators by the power connections). I then decided to turn to ol’ faithful and run Furmark. Even though Furmark shows 841W, your day to day use will likely be hugely less.
Seems that whatever their standard test is, crossfire is broken in it. Where's the mystery?
You're kinda late.http://www.3dmark.com/3dm/7597009 - HBM stock.
http://www.3dmark.com/3dm/7597144 - HBM 585 MHz.
Found out them here.
http://semiaccurate.com/forums/showpost.php?p=240079&postcount=23
Looks like HBM OC exists.
You're kinda late.
OMG!! & they are still using those unrefined drivers? i think the tdp and noise level can be curbed a little with better drivers.
I didn't think that we'd see this type of performance, with all of our games playable at 4K with over 100FPS average, except for Metro: Last Light. Battlefield 4 at 4K at 118FPS average is just amazing... something that the GTX Titan X offers, for $2000 compared to around $1300.
Read more at http://www.tweaktown.com/articles/7226/amd-radeon-r9-fury-video-cards-crossfire/index8.html
High-end gamers are going to want to examine more forthcoming articles on Fury X Crossfire. We know for example that the card's close proximity is not an issue, yet Ryan saw some signs of just one card being CPU bound in some of the testing; this would be worse with two cards. Maybe with an OCed X99 hexa or octa, such a setup would be fully unleashed. I believe Tweaktown should have OCed thier 5820K for the tests, at stock a 4790K might have actually posted better numbers.
Yeah, it has the potential to reduce the CPU bottleneck and it seems like this will especially benefit AMD on both CPU and GPU.Really interested to see what DX12 will do for multi GPU setups.