Moderátoři: Walker1134, Wilik
Ale GP100 ma HW TMU a je docela dobre mozne ze ma i Rops ....DOC_ZENITH píše:Jasné jak jsem řekl od GP100, čip kterej byl jasnej že nemá HW TMU a Rops....
https://images.nvidia.com/content/pdf/t ... epaper.pdfEach GPC inside GP100 has ten SMs. Each SM has 64 CUDA Cores and four texture units. With 60 SMs, GP100 has a total of 3840 single precision CUDA Cores and 240 texture units
You cannot directly map GPU Boost 3.0 overvoltage % to millivolts. % maps to different offset in millivolts depending on the card and even on the same system and card 100% overvoltage value may be scaled down to lower offset in mv depending on thermal conditions.
100% value is the difference between the maximum safe voltage (which is defined by BIOS) and GPU operating voltage (which depends on silicon and which is different for each card and defined by GPU leakage level (ASIC quality)). And as I said above maximum voltage can be also dynamically scaled down depending on thermal conditions.
For example on my 1070 card operating voltage is about 1040mV and with GPU Boost 3.0 overvoltage set to 100% it can reach up to 1090mV.
The best way to know your operating and maximum voltages would be setting power and thermal limits to maximum, setting fan speed to maximum to exclude 100% overvoltage downscaling, then running some not too heavy 3D application (which should NOT be power limited). This will cause GPU Boost 3.0 dynamic clock control to be limited by operating voltage (Voltage limit graph in monitoring window should report "1" in this case). Then you can set overvoltage to 100% and see how much voltage will be added.
According to Steve Keckler, the director of architecture research at Nvidia, the Echelon design incorporates a large number (~1024) of stream cores and a smaller (~8) number of latency-optimized CPU-like cores on a single chip, sharing a common memory system.Considerint the 25 times difference in performance, it is highly likely that the Echelon will employ post-Maxwell (~2013 ~ 2014) Nvidia GPU design.
Výše vyobrazená verze by byla vyráběná 10nm procesem, dosahovala velikosti 17×17 milimetrů (tedy zhruba 290 mm²). Bill Dally se podělil i o odhady vlastností 10nm procesu…
co to jen připomíná ?To nám dává 32 stream-procesorů na každý výpočetní blok. Těchto bloků nese celý čip 64 (~256 SM), což odpovídá 2048 stream-procesorům celkem. Na rozdíl od stávajících stream-procesorů (které zvládají 2 FLOPS za takt) budou navíc podporovat 4 FLOPS v taktu. Pokud by běžely na uváděných 2,5 GHz, dosahoval by celý čip výkonu 20,48 TFLOPS (single-precision)
a kam se poděl Einstein ? 
To znamená že čip podporuje double precision stejně jako GP100The P6000 cannot be a P100 GPU for several reasons:
#1 - GDDR5X memory
#2 - 24GB
Reason #1 the memory controller on the P100 is not configurable between HBM2 and DDR5X
Reason #2 24GB fits a 384 bit bus
Both of these point to the GP102 not the GP100
A na to si prišiel ako? Podľa dostupných informácií má Quadro P6000 tiež len smiešny výkon v FP64... takže ešte nič nie je isté.del42sa píše:To znamená že čip podporuje double precision stejně jako GP100
NVIDIA has given us a few answers to the question above. We have confirmation that the FP64 and FP16 rates are identical to GP104, which is to say very slow, and primarily there for compatibility/debug purposes. With the exception of INT8 support, this is a bigger GP104 throughout.
Meanwhile we have a die size for GP102: 471mm2, which is 139mm2 smaller than GP100. Given that both (presumably) have the same number of FP32 cores, the die space savings and implications are significant. This is as best of an example as we're ever going to get on the die space cost of the HPC features limited to GP100: NVLInk, fast FP64/FP16 support, larger register files, etc. By splitting HPC and graphics/inference into two GPUs, NVIDIA can produce GP102 at what should be a significantly lower price (and higher yield), something they couldn't do until the market for compute products based on GP100 was self-sustaining.
Finally, NVIDIA has clarified the branding a bit. Despite GeForce.com labeling it "the world’s ultimate graphics card," NVIDIA this morning has stated that the primary market is FP32 and INT8 compute, not gaming. Though gaming is certainly possible - and I fully expect they'll be happy to sell you $1200 gaming cards - the tables have essentially been flipped from the past Titan cards, where they were treated as gaming first and compute second. This of course opens the door to a proper GeForce branded GP102 card later on, possibly with neutered INT8 support to enforce the market segmentation.
ještě zbývá otázka uspořádání SM, respektive počet cuda cores/SM. Takže až Nvidia uvolní schéma čipu...to znamená že GP100 má ROP´s. Další verze Titanu nebo Quadro karty v budoucnu ?Ryan Smith píše:"OTOH, assuming it has all the circuitry needed to function as a GPU (I've seen rumors claiming otherwise but nothing official either way)"
I can confirm that GP100 is fully graphics capable. NVIDIA has told me as much back at GTC.
Uvidíme ještě jaké bude uspořádání SM. 
