Not all ACE's are equal.
AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread
- Thread starter iMacmatician
- Start date
-
- Tags
- amd
Some are more equal than others?
8 ACEs good, 4 ACEs bad?
8 ACEs good, 4 ACEs bad?
Interesting, hmm so Nano will be using a modified/ improved/ different version Fiji GPU?
spades, hearts, clubs, and diamonds?
That was with a blower fan.
Asked the same thing this afternoon, no answer yet.
So, one HSW is worth 2 ACEs?
I suspect we're overthinking it and they just redid the diagram for Hotchips.
interested nobody talked for me the two more interesting bits seen in HotChips slide namely:
SR-IOV virtualization: this is like vGPU Nvidia introduced in GK110 ability to partition a GPU to various subGPUs with VM seeing as a HW GPU? hope AMD can left enabled for Radeon series unlike Nvidia shipping only on GRID GPUs so users finally can use a single GPU to multiple VMs without server grade hardware.. It's already exposed on Fury X?
Dispatch draw: that may be ability to launch draw calls from GPU i.e. from graphics/compute kernels?! this is new if I understand correctly as GPUs supporting OpenCL 2.0 and NV dynamic parallelism (GK110+) have ability to dispatch compute kernels from GPU not draw calls!
interesting if ever and how this will be exposend on graphics APIs (Vulkan? OGL? via DX extension?)
hope people correct me if wrong or shed more light on this..
SR-IOV virtualization: this is like vGPU Nvidia introduced in GK110 ability to partition a GPU to various subGPUs with VM seeing as a HW GPU? hope AMD can left enabled for Radeon series unlike Nvidia shipping only on GRID GPUs so users finally can use a single GPU to multiple VMs without server grade hardware.. It's already exposed on Fury X?
Dispatch draw: that may be ability to launch draw calls from GPU i.e. from graphics/compute kernels?! this is new if I understand correctly as GPUs supporting OpenCL 2.0 and NV dynamic parallelism (GK110+) have ability to dispatch compute kernels from GPU not draw calls!
interesting if ever and how this will be exposend on graphics APIs (Vulkan? OGL? via DX extension?)
hope people correct me if wrong or shed more light on this..
More details: http://www.techpowerup.com/215529/amd-radeon-r9-nano-core-configuration-detailed.html
No disabled units and 1GHz base clock in a 175W envelope. There must be some quite aggressive power throttling.
No disabled units and 1GHz base clock in a 175W envelope. There must be some quite aggressive power throttling.
For what it's worth, the information AMD gave about the silicon in the Radeon 260 isn't consistent with the Radeon 7790.
The specifications say "Up to 1000 MHz".
D
Deleted member 13524
Guest
Based on AMD's own comparison to the FuryX I'd guess it's running at around 800Mhz on average.
Yes, your point being...?
D
Deleted member 13524
Guest
Vapor chamber DOUBLE CONFIRMED:
They're claiming a performance advantage of 30% over the GTX970 mITX version from Gigabyte and Asus, which is a rather odd comparison because these cards are selling for $300 and AMD seems to be pricing the Nano towards about twice that value. Unless the Nano is coming below $500, I'm not sure this comparison is that much of a selling point.
.
They're claiming a performance advantage of 30% over the GTX970 mITX version from Gigabyte and Asus, which is a rather odd comparison because these cards are selling for $300 and AMD seems to be pricing the Nano towards about twice that value. Unless the Nano is coming below $500, I'm not sure this comparison is that much of a selling point.
.
The one you already grasped (but didn't realize...?)Yes, your point being...?
„(other than lowering the cooler's efficiency, since there's little/no air flow under the hub itself)“
VCs are basically very short but very wide Heatpipes. The upper end being the one where the heat is transferred to and from which it needs to be transported away. Thus, VCs until now had their surface soldered over and over with fins while the fan was blowing through them from the side (radial design).
That's why a VC would IMHO not be optimal for this small an area, compared to cards like the one HD 4870 X2 you mentioned - where the fan did indeed blow from the side.
--
But seeing the alleged slides, it seems I'm wrong anyway and there's indeed a vapor chamber in the base of the cooler.
Odd, yes, but the point IMHO is the maximum performance in that small form factor.
It's no different for a vapor chamber-based cooler or a heatpipe-based cooler; the fan hub is there regardless. Difference is, the vapor chamber has a higher capacity for heat dispersion, and thus more efficient overall.
Maybe the HWS ACEs support "Compute Wave Switch" as mentioned on the slide, or in other words support for preemptive scheduling of compute shaders.
HWS could just be another name for the compute wave switch feature (Hardware Wave Switcher?).