So Navi has 512Kb L1? :shock:
Vega10 has 16Kb L1.
Yeah. NVs RTX Demo has some artifacts too, some objects are blurred
NAVI10LITE : GFX1000 (aka Navi 12?)
NAVI10 : GFX1010
Vega20 vs Vega10 in Compubench...
Polaris with Rapid Packed Math support?
Navi10 is added to pre-release version of HWiNFO64
Navi10 = GFX10 is also marked as NV10
Vega20 (gfx906/7) and Navi10 (gfx1000/1001) spotted in the Radeon Software Adrenalin Edition 18.1.1 Alpha Driver
And Intel KNL uses MCDRAM, which isn't a JEDEC standard.
Okay. What about HMC?
So HMC and GDDR6 aren't a JEDEC standard too.
"Nextgen Memory" is DiRAM4 Memory, maybe Navi will have this memory architecture I think
There are two new Navi chips -- 1000 and 1001 GPUs
Vega is Vega. Vega 20 is gfx9.
AMD has a Linux driver for gfx10 architecture. That's 100% Navi GPU.
Worst-Case is a scenario that is very rare.
PCPer updated: 564mm2 >> 512mm2 >> .... To be continued.
Vega GPU has Low Cost HBM by Samsung?
So Low Cost Samsung HBM2 has faster pin speed of 3 GB/s (+) that can deliver 200 GB/s compared to...
Maybe V10/64CU >> 1CU/96SP -- 6144SP*2 = 12.288, 12.288*1000MHz = 12.288.000, ~12.2 TFlops. No?
Ooops, sorry for the typo. I mean DP1.3.
Maxwell v2 -- HDMI 2.0, FL12_1
GCN Gen4 -- HDMI 2.0a, DP 1.2 and FL12_1?
Hawaii -- Graphics IP7, GCN Gen2
Tonga -- Graphics IP8, GCN Gen3
Fiji -- Graphics IP8, GCN Gen3
Polaris -- same Graphics IP8, but new GCN Gen4?
HBM DualLink = 8GB
596mm2/2.5 = 238.4mm2
Raja Koduri: Polaris is 2.5 times brighter today than when Ptolemy observed it in 169 A.D
2 Hardware Schedulers and 4 new ACE's with 128 Queues Total?
New ACE F32 with 32 Threads, 4 ACE's = 128 Threads/Queues...
Fiji has 4 ACEs and 2 HWS? What is HWS?
2.88% -- GPU Clock speed over the Tri-X
15.4% -- Memory Clock speed over the Tri-X
6GB for Fiji?
Сustom version of HBM by AMD&SK-Hynix, I think
1250 MHz HBM
The mass rebranding - it is a fraud and a scam. That's enough to take legal action against AMD.
I think that a new chip based on Tonga architecture. Just specs similar to Hawaii specs.
Fiji is GCN 1.2?
AMD published a paper last year about multi-level design in next-gen memory. It's possible they just wanted it out sooner and merely toyed with...
ACE can operate in parallel with Graphics CP, ie ACE have access to shader engine with rasterizer via Graphics CP >> GDS for mixed mode.
1 DMA for GeForce cards and 2 DMA for Quadro/Tesla cards, I think. This chart was posted on AT-forum some days ago.
"NVIDIA GeForce GTX 750 Ti"
Maximum feature level : D3D_FEATURE_LEVEL_11_0 (11.0)
ResourceBindingTier : D3D12_RESOURCE_BINDING_TIER_2 (2)
French article said GCN 1.1 has up to 8 ACE's per 8 queues = 64 + CP = 65 compute...
some slide from brazilian website
Page 8 from AMD's PDF shows 8 stacks of HBM on GPU/APU interposer.
4GB vs 12GB in this "4K-tests". And R600 was more slower than G80.
Liquid water cooler require ~20/30W
GCN 1.2 PDF...
More details here
970 faster than 290X in Luxmark :cool: