Radeon Fury to be slower than GTX 980 Ti

[Xzibit]

One of the links I just showed you talks about Pascal having NVLink AND being compared to Maxwell. How would it make sense that they'd compare Pascal to Maxwell, since they explain it will take NVLink to reach that claimed 10x power?

Custom HPCs
Maxwell = PCIE
Pascal = NVLink

PC needs uni mem, consoles have had it for a while. It would be stupid to bring this only to workstation cards. And the charts clearly show multi GPU use as per SLI.

None of the charts show SLI. They show the NVLink communication. Do you see SLI connectors on any of the Pascal screenshots ?

Then there's the fact that they've said Maxwell will have the software version of NVLink via Cuda 6. You're implying they'll go from that to leaving consumer Pascal cards without ANY uni mem. Otherwise they'd have said Maxwell AND Pascal will have the Cuda 6 software version of uni mem.

CUDA 7 has been available since the start of the year. Nvidia officially launched CUDA 7 at GTC. Havent heard of this CUDA 6 NVLink your talking about.

NVLink = New connector in Mezzanine form for custom HPCs
Pascal for consumers will be in PCIE = No NVLink

 

[GreiverBlade]

390X is rebranded 290X which is far behind 980ti. Its not that it could be slower, it will be.
Fury however, yes, unknown but I don't see it being slower. And it might come bundled with BF Star Wars (Hexus article). Ooooh.

a 20ish % behind the Ti but it's the 290X we still don't know how the refined core in 390X will behave

 

[Frag_Maniac]

None of the charts show SLI. They show the NVLink communication. Do you see SLI connectors on any of the Pascal screenshots ?

You're the one implying NVLink can't even work with Pci Ex switches, yet it says this right below one of the charts in the last link I posted, and multi GPU being used.

"The scenarios that Nvidia is suggesting in the chart above for how CPU-GPU complexes will be linked with the NVLink 1.0 interconnect shows the CPUs being linked to the GPUs using PCI-Express switches."

CUDA 7 has been available since the start of the year. Nvidia officially launched CUDA 7 at GTC. Havent heard of this CUDA 6 NVLink your talking about.

NVLink = New connector in Mezzanine form for custom HPCs
Pascal for consumers will be in PCIE = No NVLink

Here it shows the very first feature listed for CUDA 6 is uni mem, which has also been described as the software version of NVLink.

http://devblogs.nvidia.com/parallelforall/powerful-new-features-cuda-6/

I think you're in denial and a bit out of the loop dude.

 

[Xzibit]

You're the one implying NVLink can't even work with Pci Ex switches, yet it says this right below one of the charts in the last link I posted, and multi GPU being used.

"The scenarios that Nvidia is suggesting in the chart above for how CPU-GPU complexes will be linked with the NVLink 1.0 interconnect shows the CPUs being linked to the GPUs using PCI-Express switches."

Here it shows the very first feature listed for CUDA 6 is uni mem, which has also been described as the software version of NVLink.

http://devblogs.nvidia.com/parallelforall/powerful-new-features-cuda-6/

I think you're in denial and a bit out of the loop dude.

From the first link you posted

Now to be clear here, CUDA 6’s unified memory system doesn’t resolve the technical limitations that require memory copies – specifically, the limited bandwidth and latency of PCIe – rather it’s a change in who’s doing the memory management. Data still needs to be copied to the GPU to be operated upon, but whereas CUDA 5 required explicit memory operations (higher level toolkits built on top of CUDA withstanding) CUDA 6 offers the ability to have CUDA do it instead, freeing the programmer from the task.

I think your confused as to what NVLink is

NVIDIA NVLINK HIGH-SPEED INTERCONNECT

NVIDIA® NVLink™ is a high-bandwidth, energy-efficient interconnect that enables ultra-fast communication between the CPU and GPU, and between GPUs. The technology allows data sharing at rates 5 to 12 times faster than the traditional PCIe Gen3 interconnect,

I cant make it any more simple then that video.

NVLink replaces PCI Express

 

[the54thvoid]

a 20ish % behind the Ti but it's the 290X we still don't know how the refined core in 390X will behave

TPU review, its 50% gap in an undoubted AMD title, DA: Inquisition at 4k.

Could pick n choose but it looks like 40% fps gap on most things averaged (this is a quick check before work at 6.45am!)

 

[irani]

390X is rebranded 290X .

it's not true

 

[Frag_Maniac]

I think your confused as to what NVLink is...

I'm well aware of what it is, but unlike you, I don't ignore that it can be used between GPUs even on a Pci Ex MB, which I just showed in my last post. The only thing currently that's keeping it from being a reality between CPU and GPU on consumer MBs is someone like Intel jumping on board.

You have no spec chart, no system info sig, so I have to wonder, are you trying to hide being an AMD fanboy or something, because this reeks of denial.

After all, you wouldn't even respond to the info I showed on Nvidia's own blog about the FIRST feature of CUDA 6 being listed as uni mem, after saying you've heard of no such thing. And that's for already existing Maxwell cards.

 

[GreiverBlade]

TPU review, its 50% gap in an undoubted AMD title, DA: Inquisition at 4k.

Could pick n choose but it looks like 40% fps gap on most things averaged (this is a quick check before work at 6.45am!)

ooohhh... one title ... (tho a 290/290X still achieve a playable framerate in any title, so : no biggies )

ok sorry ... i was talking in 1080p and averaged... my bad too bad all my friends have 970 or 980 and not a Ti otherwise i would gladly like to see that gap in real life situation just as i, well ... , don't see it.
well yes fps counter show their card is faster, but does it warrant the price difference they paid (even new cards ... not specially talking about the price i got for my card ) now if nvidia would have a more "affordable" price i wouldn't mind supporting them a bit more (than having just a Shield Tablet and a GT730... as a backup for Bios flash ) because now from a 290X to a 980Ti the price difference is around 116.7% more

oops sorry it was 29% in 1080p averaged, 33% on 1440p, 27% on 2160p and 29% in 900p so yep around 40% averaged (don't misunderstand me i find it quite impressive for the power consumption difference)

but i see what you meant no worries

it's not true

well not technically ... but yep it's the same silicon with some refinement. (they don't need to renew the whole lineup as they hold their ground quite fine ... )

 

[HumanSmoke]

it's not true

You have some proof? Because there is some pretty substantial evidence that points to the 390X being a barely warmed over 290X.
The first example would be PowerColor's 390X using the same PCB as its 290X ( LF R29FA), while XFX and Asus's leaks confirm basically the same thing. If you consider a 50MHz core bump and a 250Mhz (1G effective) memory clock increase to not be a rebrand, then that's your prerogative, but cards such as the GTX 770 are considered by many people to be rebrands - and that particular card bought a larger clockspeed increase as well as a dynamic boost algorithm.

 

[irani]

I was wrong with fury

yes 390x is re-branded of 290x

Sorry guys

 

[the54thvoid]

I want to state here that if the Fury card over clocked doesn't beat a standard over clocked 980ti, its an Nvidia win (and vice versa). I say this because Maxwell cards could all be a good 100-200mhz higher in core but Nvidia insist on being power efficient (by limiting their hardware).
I'll concede that Fury gets the popular win if at stock its faster. And if it wins the over clock condition, well, happy days for AMD!

 

[BiggieShady]

I want to state here that if the Fury card over clocked doesn't beat a standard over clocked 980ti, its an Nvidia win (and vice versa). I say this because Maxwell cards could all be a good 100-200mhz higher in core but Nvidia insist on being power efficient (by limiting their hardware).
I'll concede that Fury gets the popular win if at stock its faster. And if it wins the over clock condition, well, happy days for AMD!

I must agree, but 980 Ti needs to be OC-ed on the same liquid cooler as Fury card has (will have)
... otherwise we could already give a title to Fury because with massive shader count and memory bandwidth, it should throttle insanely for it to be slower ... and it won't because liquid cooling.

 

[RejZoR]

390X is rebranded 290X which is far behind 980ti. Its not that it could be slower, it will be.
Fury however, yes, unknown but I don't see it being slower. And it might come bundled with BF Star Wars (Hexus article). Ooooh.

Not true. R9-390X is based on R9-290X GPU core arrangement with R9-285 optimizations and possibly additional DX12 instructions. Rebrand is when you just stick a new BIOS with different name on it and maybe raise clocks a bit...

 

[the54thvoid]

Not true. R9-390X is based on R9-290X GPU core arrangement with R9-285 optimizations and possibly additional DX12 instructions. Rebrand is when you just stick a new BIOS with different name on it and maybe raise clocks a bit...

Not in my books. GPU rebrands are done using the vast majority of the chip architecture and tweaking here or there. The 390X is a spiritual 'rebrand'. Feel free to take that up with many other tech sites that call it such.

I suppose it would be normalised if it were called the 295X but that kind of got taken. I know NV do it too, 770 was a 680 on juice but either way - Fury is AMD's only real 'new' creation.

 

[TheGuruStud]

Not in my books. GPU rebrands are done using the vast majority of the chip architecture and tweaking here or there. The 390X is a spiritual 'rebrand'. Feel free to take that up with many other tech sites that call it such.

I suppose it would be normalised if it were called the 295X but that kind of got taken. I know NV do it too, 770 was a 680 on juice but either way - Fury is AMD's only real 'new' creation.

Intel is the biggest rebrander of all time. From SB to skylake lol

 

[Rivage]

AMD Radeon Fury X 3DMark performance

 

[TheGuruStud]

AMD Radeon Fury X 3DMark performance

View attachment 65590

View attachment 65591

Let the fancy faking of numbers begin!

 

[Rivage]

AMD Radeon Fury X pictured some more - Liquid Cooling, Backplate And Red LED Lit Radeon Logo

 

[the54thvoid]

What's wrong with the infill on the 'R', 'A', 'D' and 'O'? It's not shiny like the rest of the shroud.

And do i see a double BIOS switch for tampering goodness? Though I've never flashed a Radeon, will need help if I get one.

I wonder what the air cooled one looks like? Different PCB but longer cooler shroud?

Not long now......

 

[Rivage]

That soft touch backplate and plate on top looks sexy. But are they in their places? I mean, is it good for heat dissipation? Doubt..

And do i see a double BIOS switch for tampering goodness?

that one at 1 pic. Think so..

Not long now......

Yeah! just a 5 days to Go!

 

[RejZoR]

Not in my books. GPU rebrands are done using the vast majority of the chip architecture and tweaking here or there. The 390X is a spiritual 'rebrand'. Feel free to take that up with many other tech sites that call it such.

I suppose it would be normalised if it were called the 295X but that kind of got taken. I know NV do it too, 770 was a 680 on juice but either way - Fury is AMD's only real 'new' creation.

Then I don't know what books you are reading. Refresh is NOT rebranding. It even goes against the dictionary definition of it.

New architecture = entirely new chip that has very little in common with predecessor
Refresh = better process node, optimizations
Rebrand = Same everything, maybe different clocks and different BIOS with (obviously) different name.

 

[the54thvoid]

Then I don't know what books you are reading. Refresh is NOT rebranding. It even goes against the dictionary definition of it.

New architecture = entirely new chip that has very little in common with predecessor
Refresh = better process node, optimizations
Rebrand = Same everything, maybe different clocks and different BIOS with (obviously) different name.

Same process. Is it a half baked refresh?

In fact - what tweaks has it got? What GCN compared with 290X? Every site I see refers to it as a rebrand. Its got more memory and it's got higher clocks. It's a rebrand.

Final Edit from TPU article today:

The R9 390X, is expected to be a re-brand of the previous generation R9 290X, with its standard memory amount raised to 8 GB. It's based on the 28 nm "Grenada" silicon. We've seen no evidence pointing at "Grenada" being some sort of an upgrade of "Hawaii" with newer GCN 1.2 stream processors. Perhaps AMD polished its electricals to the extent it could, without changing the silicon. We'll know for sure only next week.

 

[RejZoR]

R9-290X has NO framebuffer compression. R9-390X does. R9-290 uses GCN 1.1, R9-390X uses (modified) GCN 1.2 cores. Ergo, NOT a rebrand no matter what you say. And considering the changes within, you can also expect extended DX12 support. There is no way they'll leave high end range at only base DX12 feature level...

 

[Fluffmeister]

Source: https://www.reddit.com/r/Amd/comments/399775/new_picture_of_fury_nano_w_backplate_tubing_and/

 

[R-T-B]

R9-290X has NO framebuffer compression. R9-390X does. R9-290 uses GCN 1.1, R9-390X uses (modified) GCN 1.2 cores. Ergo, NOT a rebrand no matter what you say. And considering the changes within, you can also expect extended DX12 support. There is no way they'll leave high end range at only base DX12 feature level...

TPU seems to disagree, as noted above.