Monday, February 9th 2015
Radeon R9 380X Based on "Grenada," a Refined "Hawaii"
AMD's upcoming Radeon R9 380X and R9 380 graphics cards, with which it wants to immediately address the GTX 980 and GTX 970, will be based on a "new" silicon codenamed "Grenada." Built on the 28 nm silicon fab process, Grenada will be a refined variant of "Hawaii," much in the same way as "Curacao" was of "Pitcairn," in the previous generation.
The Grenada silicon will have the same specs as Hawaii - 2,816 GCN stream processors, 176 TMUs, 64 ROPs, and a 512-bit wide GDDR5 memory interface, holding 4 GB memory. Refinements in the silicon over Hawaii could allow AMD to increase clock speeds, to outperform the GTX 980 and GTX 970. We don't expect the chip to be any more energy efficient at its final clocks, than Hawaii. AMD's design focus appears to be performance. AMD could save itself the embarrassment of a loud reference design cooler, by throwing the chip up for quiet custom-design cooling solutions from AIB (add-in board) partners from day-one.
In other news, the "Tonga" silicon, which made its debut with the performance-segment Radeon R9 285, could form the foundation of Radeon R9 370 series, consisting of the R9 370X, and the R9 370. Tonga physically features 2,048 stream processors based on the more advanced GCN 1.3 architecture, 128 TMUs, 32 ROPs, and a 384-bit wide GDDR5 memory interface. Both the R9 370 and R9 370X could feature 3 GB of standard memory amount.
The only truly new silicon with the R9 300 series, is "Fiji." This chip will be designed to drive AMD's high-end single- and dual-GPU graphics cards, and will be built to compete with the GM200 silicon from NVIDIA, and the GeForce GTX TITAN-X it will debut with. This chip features 4,096 stream processors based on the GCN 1.3 architecture - double that of "Tonga," 256 TMUs, 128 ROPs, and a 1024-bit wide HBM memory interface, offering 640 GB/s of memory bandwidth. 4 GB could be the standard memory amount. The three cards AMD will carve out of this silicon, are the R9 390, the R9 390X, and the R9 390X2.
Source:
3DCenter.org
The Grenada silicon will have the same specs as Hawaii - 2,816 GCN stream processors, 176 TMUs, 64 ROPs, and a 512-bit wide GDDR5 memory interface, holding 4 GB memory. Refinements in the silicon over Hawaii could allow AMD to increase clock speeds, to outperform the GTX 980 and GTX 970. We don't expect the chip to be any more energy efficient at its final clocks, than Hawaii. AMD's design focus appears to be performance. AMD could save itself the embarrassment of a loud reference design cooler, by throwing the chip up for quiet custom-design cooling solutions from AIB (add-in board) partners from day-one.
The only truly new silicon with the R9 300 series, is "Fiji." This chip will be designed to drive AMD's high-end single- and dual-GPU graphics cards, and will be built to compete with the GM200 silicon from NVIDIA, and the GeForce GTX TITAN-X it will debut with. This chip features 4,096 stream processors based on the GCN 1.3 architecture - double that of "Tonga," 256 TMUs, 128 ROPs, and a 1024-bit wide HBM memory interface, offering 640 GB/s of memory bandwidth. 4 GB could be the standard memory amount. The three cards AMD will carve out of this silicon, are the R9 390, the R9 390X, and the R9 390X2.
156 Comments on Radeon R9 380X Based on "Grenada," a Refined "Hawaii"
Efficiency at the high end really isn't that pressing a concern - Nvidia built markets share fielding the GT200 and GF100 - although might be a consideration in non-gaming scenarios.
Where you should care about efficiency is that the architecture scales efficiently with the smaller GPUs. As I said earlier, AMD aren't competitive in discrete mobile, and that also reflects on low/mid range OEM builds, where the vast majority of discrete sales happen.
You can say that efficiency doesn't matter (to you), but it just means one more strike against sales and revenue, which means less R&D funding, which means that AMD may not be able to field a top-to-bottom GPU refresh(!). So many things don't seem to matter with AMD - GPU efficiency, enthusiast desktop, the x86 server market....sooner or later the sum total of these "doesn't matter" must indeed matter.
So you all argue about power consumption with 800w+ PSUs? Really, chill.. Power consumption is overrated. :p
Also there is no TITAN-X or what ever you all like about Titan, GM200 won't have DP so no Titan variant, just Geforce (google Nvidia Japan conference 30-12-2014)..
Talk about Wccf spreading this false Titan-X hype to the max..:shadedshu:
Agreed those reference coolers were awful just like the GTX 480's cooler.
I don't see the big deal :wtf: If the 380x was going to be the 4096 monster, pretty sure the 390x would of been the dual GPU config. Just different labels :wtf:
You're comparing apples to oranges here, for two different reasons.
1. You're comparing CPU architecture to GPU, which are very different in design.
2. You're comparing a chip produced by Intel's fabs to that of one designed by AMD, but produced at either TSMC or Global Foundries.
What's more, nvidia said @ that Japan tech conference there won't be any DP gpu with Maxwell, only with Pascal and then we will see new Tesla/Titans again.
www.kitguru.net/components/graphic-cards/anton-shilov/nvidia-to-speed-up-development-of-graphics-processing-architectures/
Besides these GM200 Geforce 1080gtx? will feature 6Gb vram anyway, so Titan name is irrelevant now if that extra 3gb vram buffer by GK110 made them a little more special.
Anyway OT, can't wait for this FijiXT, really interested in that 3D memory, should make a big change or two at higher resolutions.
Personally, I'd like to see the name changed to Zeus (son of Titans) if only to troll AMD rumour lovers, btarunr, and RCoon ;)That is incorrect. GM 200 likely has double precision at the same rate of GM 204 ( 1:32 ). Insufficient for Tesla duties, but that is why the rep said that Kepler will continue to be the Tesla option - simply because GK 210's developmenthas been in tandem with GM 200.
Current Titans cost so much because of FP64 DP, not because of extra 3GB vram.Exactly, its improved FP32, but its not FP64. So it can't be used like with GK110 Titan @ FP64 mode.
Btw that GK210 is 2x improved and energy efficient GK110 each with 2496cores.. So its not really Maxwell either :)
What Im also trying to say is, all this means no "absurd" prices for us end-users, from both camps AMD FijiXT and NV GM200, the usual 550-650$/€.
Oh and for general public information, AMD's TressFX works on ALL graphic cards, not just AMD, because unlike NVIDIA's proprietary crap, TressFX works through DirectCompute, which means support on all modern graphic cards.
GTX 980 8 W;
GTX 970 9 W;
R9 290 16 W;
R9 290X 17 W.
www.techpowerup.com/reviews/Gigabyte/GTX_960_G1_Gaming/26.html
That is exactly double the power consumption and the question is principled....
In multi-display it is even worse.
GTX 980 9 W;
GTX 970 10 W;
R9 290 51 W;
R9 290X 54 W.
:(You are mistaken. It actually shows that something is not working properly
FUN FACT #2: Nvidia offered AMD a PhysX licence (after paying the $150 million asking fee to buy AGEIA), but AMD decided to go with HavokFX, because OpenCL gaming was the next big thing. This is the same Havok that required a licensing fee and was supported by exactly zero games.
FUN FACT #3: When the PhysX hack for AMD cards arrived, it was AMD who threw up the roadblock.
So, ATI/AMD couldn't be bothered buying PhysX, couldn't be bothered licensing it once Nvidia purchased it, and actively blocked the development of a workaround that would allow the AMD community from using it. If you have an Nvidia card you can use it. If you have an AMD card, why should you care? AMD certainly don't.
Nvidia are such crybaby bitches that they actively block their cards from using physx when made secondary to an AMD card. You made that point...and it goes against your propaganda!
Obviously, you have no rebuttal against TressFX LOL. Nvidia won't have ANY of this open standard stuff. They'll bankrupt the company before they let it happen. That's how arrogant and greedy they are.
And I know you fanboys are INCREDIBLY butt hurt about Mantle and Freesync. Let me see those tears, baby!
Last i checked Freesync and Mantle are Proprietary CLOSED software for AMD. SO tell us another AMD fanboy blind lie.
And so what if TressFX uis limited only to hair. It does work on ANY graphic card with DirectCompute support. You can't even have PhysX hardware accelerated hair if you just happent o have Radeon...