AMD Radeon R9 Fury X Launches Today, Initial Results A Bit Of A Let Down

[artivision]

Seriously? I can easily find block diagrams and describe them for you, but do I have to? You should look at several block diagrams for yourself. My objective is usually to find the most detailed one I can. My advice is that you should do that too.

Pretty much yes. Only Kepler has dedicated 64bit units that are capable to execute one 32bit in just one circle instead of four. Fermi has shared 64bit ones with 32 bit ones.

[smitty3268]

Radeon with 4000 shaders vs Geforce with 3000 shaders at the same frequency and Geforce wins.
...
So Geforce 3000 shaders = 4500 normal shaders.

That's idiotic - even more so than comparing a x86, ARM, and Power cpu at the same frequency and claiming one is better architecturally based off that, because at least there you are looking at a one-to-one comparison.

That has nothing to do with drivers.

Well, obviously. The hardware design is completely different, and not comparable in any way.

[artivision]

That's idiotic - even more so than comparing a x86, ARM, and Power cpu at the same frequency and claiming one is better architecturally based off that, because at least there you are looking at a one-to-one comparison.


Well, obviously. The hardware design is completely different, and not comparable in any way.

I just said the same thing, example:

Random Radeon GPU: 300gflops SP-32bit, 150gflops DP-32bit or much less.

Fermi GPU: 300glfops SP-32bit, 150gflops DP-"""64bit"""

Example 2: 1.6tflops Fermi beats the same generation 2.7tflops HD5000 vastly even today. Not driver related. Just different arch. Because Fermi where measured in FMAC, has extra pipeline for 64bit, extra Load/store, extra SFUs.

So i just conclude to an analogy: 1x Nvidia shader = 1.5x AMD shader at the same frequency.

[xeekei]

I just said the same thing, example:

Random Radeon GPU: 300gflops SP-32bit, 150gflops DP-32bit or much less.

Fermi GPU: 300glfops SP-32bit, 150gflops DP-"""64bit"""

Example 2: 1.6tflops Fermi beats the same generation 2.7tflops HD5000 vastly even today. Not driver related. Just different arch. Because Fermi where measured in FMAC, has extra pipeline for 64bit, extra Load/store, extra SFUs.

So i just conclude to an analogy: 1x Nvidia shader = 1.5x AMD shader at the same frequency.

By your logic, CPUs are superior to GPUs.

[bridgman]

Random Radeon GPU: 300gflops SP-32bit, 150gflops DP-32bit or much less.

Fermi GPU: 300glfops SP-32bit, 150gflops DP-"""64bit"""

What the heck is DP-32 bit ?

SP is 32-bit, DP is 64-bit, FMAC usually has some additional bits of precision between the M and the AC.

[artivision]

By your logic, CPUs are superior to GPUs.

I respect you all and your different opinions. I just wanted to state that Fury_X will always lose to Titan_X, so is marketing trick to convince people to bay and wait for the miracle (magical driver update). If you want to bay Radeon, do it for the correct reasons (Gallium Nine), as me. And don't bay the strongest ones, they will never get filled with SM3 material.

[johnc]

Personally, I really like the idea of 52c. Nice! Thank You AMD. Fantastic. Gaming PC's don't need to be ovens and attitudes toward thermals in the past have been abysmal. I'm so glad that it's regulated to something reasonable.

I like water cooling too because it cuts down on noise big time, but the temp readings are lower only because the cooler is dissipating the heat more efficiently. In terms of heat generated the power draw is going to be a better indicator.

[liam]

Sweclockers did: http://www.sweclockers.com/test/2073...y-x/16#content
King of OpenCL, but not really a surprise.

It's faster in luxmark but slower infacedetect.t

[charlie]

I woke up early and f5'ed hard everywhere in my country and still couldn't find a Fury X to buy. I'm not impressed by the negative forum impressions of this card. I still need/want 4k+ now and as quiet as possible. I'm disappointed by no 2.0HDMI, but screw tv's then--I'll save up for future 5k displayport monitors. Did you guys not see this card is capable of running 12k resolutions? Choke on that naysayers (naysayers - a person who habitually expresses negative or pessimistic views [because it is the majorities' 'in' thing]). For the near term I am looking forward to seeing if this card can scale 5k to a UHD resolution easily. AMD forever.

[wizard69]

The overclock potential is ~5% (meh), apparently there is pump noise, however to be fair my water pumps on my CPU had a break in time where they were noisy too. The problem the sapphire card I wanted has a bit of a waiting list =(. Maybe by mid July when more stock is in the cards will come down to $600us. It competes closely graphical wise with Titan TI with offering a monstrous amount more of compute.

One real negativity though, apparently the pixel fill rate on these card are low for some reason (lower than 290x even), not sure what's up with that.

Driver bugs maybe? I find it silly people compare the performance of a new device with a GPU that has been on the market for months with tons of driver updates under its belt. I would expect that at least one more driver update is needed to start to realize the new architecture performance promise.