Kupa GTX 970

[Zoltan Balaton]

chcem,ale bez zvysovania napatie..tych 300Mzh co da,ked mu zvysim napatie ma nezaujima

[scream]

No pozri, mam MSI GTX970 Gaming, nic nepiska, karta je aj v zatazi ticha a po OC je na tom vykonovo vacsinou lepsie ako pretaktovana AMD 390ka. Aj 390 aj 970 pri 1080p v zaklade podavaju skoro rovnaky vykon, cize vyberaj podla rozlisenia v ktorom hras, hier, chladenia, farby, ceny, vystupov, ceny, spotreby... ale nevyberaj podla priemerneho vykonu.

[Asnod]

Az na to ze pri DX12 to predbezne zase vyzera na fuck*d up pre Nvidiu koli skvelym recenziam a tak orezanej karte ze vdaka asynchronnym shaderom v dx12 980ti=r9 390

[loleqboleq]

bóže asnod, zase tote bludy...to je ako robiť benche hry v alpha štádiu

[Asnod]

Tu to mas aj s vysvetlenim. 2GB vram, po release maxwellu zazracny kepler low performance, memory blackhole kauza s 970 a teraz pride low Maxwell DX12 performance po release Pascalu a prvych DX12 hrach. Nuteny upgrade (vzhladom na Predaje grafických kariet dosiahli 10-ročné minimum). It is meant to be f*cked up. Pre niekoho bohuzial tu sa skryva cela zazracna efektivita Maxwell architektury.





Az na to ze DX uz v alpha stadiu nieje a je vo final verzii oficialne vonku. Tak si uzivajte svoje karty. Vies preco AMD sa neopicilo a netlacilo dole spotrebu? Vzhladom na DX12 to na sucastnom procese a GCN orezat nemozu a vzhladom na svojich zakaznikov si to ani nedovolia. Preto volilo skorsie pouzitie HBM aby v DX11 dobehlo 980ti a usetrilo par W na pamatiach pre jadro a tu kartu dotiahlo Nv a dx12 bude este pekna komedia.

DX12 PC/Xbox one api, PS4/PC pravdepodobne skonci s planovanym Vulcan api. Vsetko do seba pekne zapada. To som este nevedel po release 970 ze to nie ta karta je ojeb ale cela architektura

[loleqboleq]

akože prajem ti, aby si mal pravdu, ale bludáriš, samé domnienky, no nebude to tak

[Asnod]

Tu mas same fakty

I see what you guys are saying and I think I've figured out something new that didn't come to me before...

nVIDIA mentioned that Maxwell 2 is capable of queuing 32 Compute or 1 Graphics and 31 Compute. Everyone agrees with this. Now the question becomes... what are these queues tied too?

I assumed that Maxwell 2 had no Asynchronous Compute Engines because nVIDIA has been quite quiet about this. I assumed that they scheduled these queues from within their SMM Units (Compute Units). This would account for the performance deficit.

Now I'm thinking that something else could account for this performance deficit. That is if Maxwell 2 has only a single Asynchronous Compute Engine tied to 32 Queues.

So I began to read up on Kepler/Maxwell/2 CUDA documentation and I found what I was looking for. Basically Maxwell 2 makes use of a single ACE-like unit. nVIDIA name this unit the Grid Management Unit. The CPUs various Cores send Parallel streams to the Stream Queue Management. The Stream Queue Management sends streams to the Grid Management Unit (Parallel to Serial thus far). The Grid Management unit can then create multiple hardware work queues (1 Graphics and 31 Compute or 32 Compute) which are then sent in a Serial fashion to the Work Distributor (one after the other or in Serial based on priority) . The Work Distributor, in a Parallel fashion, assigns the work loads to the various SMMs. The SMMs then assigns the work to a specific array of CUDA cores. nVIDIA call this entire process "HyperQ".
Here's the documentation: http://docs.nvidia.com/cuda/samples/6_A ... HyperQ.pdf

GCN 1.1/1.2, on the other hand, works in a very different manner. The CPUs various Cores send Parallel streams to the Asynchronous Compute Engines various Queues (up to 64). The Asynchronous Compute Engines prioritizes the work and then sends it off, directly, to specific Compute Units based on availability. That's it.
HyperQ is thus bottlenecked at the Grid Management and then Work Distributor segments of its pipeline.

nVIDIAs HyperQ can be thus deemed "in order". HyperQ contains only a single pipeline.

AMDs Asynchronous Compute Engine implementation can be thus deemed "out of order". AMDs implementation contains eight pipelines.

AMDs implementation incurs less latency as well as being superior in terms of efficiently utilizing the available Compute resources.

This explains why Maxwell 2 (GTX 980 Ti) performs so poorly under Ashes of the Singularity. Asynchronous Shading kills its performance compared to GCN 1.1/1.2 (in this case an R9 290x) which barely loses any performance.

GCN 1.1/1.2 are clearly being limited elsewhere, and the explanation I gave (regarding the Peak Rasterization Rate or Gtris/s) still stands as the most plausible cause.

I've been away from this stuff for a few years so I'm quite rusty but Direct X 12 is getting me interested once again.

[loleqboleq]

chápem myšlienku a nemám proti tomu nič ale jednoducho som sa naučil v tejto oblasti, že ukvapené dohady (aj keď sa zdá, že sú podložené technickými faktami nemusia v konečnom dôsledku platiť (dvojsečná zbraň vskutku)

avšak ja zatiaľ čerpám napríklad z Mantle, ktorý robil to isté ako DX12 (podľa rovnako irelevantných ako tieto "aktuálne") v tých benchoch na porovnanie DX11,Mantle a DX12 a aj tak to nestačilo na maxwelly

takže ja si radšej počkám na real-world performance predtým ako si dovolím povedať, že R9 390 zhltne 980Ti jak nič

[scream]

...

ze ta to bavi... normalne mi ukaz predbezne aspon jednu hru, kde to tak je. Sak tu mame kopec alfa,beta,gama releasov...

Ano AMD v globale dokaze spracovat viac draw calls ako maxvell... avsak, pocet draw calls, ktore dokaze maxvel spracovat pod DX12 stupol voci DX11 SEDEMNASOBNE! mas pocit, ze to bude do buducna malo?

[Tracer]

Zdravím aj ja sa prikláňam k GTX 970 láka ma tento model :http://www.tichepc.sk/ASUS-GeForce-GTX- ... 1162001023 len sa bojím ako to bude chladiť ten malý FAN samozrejme AMD návrhy príjmem veľmy rád

[Asnod]

Nejde tu o drawcalls. dalsi expert sa ozval. Dalej no comment lebo to bude rovnaky fail a sama perla ako tie tvoje tvrdenia z nedavna.

Lolek, mantle tak isto nikde nebenefitovalo z asynchronnych shaderov. Islo akurat o MT rendering a upraveny memory management.

[scream]

Dal si tu jediny graf, kde su DRAW CALLS, tak si na to dostal odpoved. By si spokojne mohol ist robit PR do AMD.

Chcem vidiet tych developerov, co spravia AAA titul tak, aby siel na NVIDII horsie o 40% ako na karte AMD rovnakej triedy.

[loleqboleq]

Zdravím aj ja sa prikláňam k GTX 970 láka ma tento model :http://www.tichepc.sk/ASUS-GeForce-GTX- ... 1162001023 len sa bojím ako to bude chladiť ten malý FAN samozrejme AMD návrhy príjmem veľmy rád

čo chceš na to počuť ? je to hrôza a des....referák a za takú cenu ?

https://www.alza.sk/asus-strix-gtx970-d ... 169036.htm
o dve galaxie lepšia

alebo od GABA
http://www.tichepc.sk/ASUS-GeForce-GTX- ... DP_3_.html

[Asnod]

Dal si tu jediny graf, kde su DRAW CALLS, tak si na to dostal odpoved. By si spokojne mohol ist robit PR do AMD.

Chcem vidiet tych developerov, co spravia AAA titul tak, aby siel na NVIDII horsie o 40% ako na karte AMD rovnakej triedy.

Ospravedlnujem sa, nevedel som ze si dyslektik.

[scream]

Asi to bude tym, ze tie tvoje sracky, ktore zacinaju plackanim nad keplerom pokracuju neustalym hejtingom nvidia kariet, ktore podavaju rovnaky alebo lepsi vykon za rovnaku cenu ako AMD sa proste nedaju citat kompletne.

Ale v skratke to bolo asi takto, ze?

1.Ashes of the Singularity - hra ktora v alfa verzii vyzera ako rit a je schopna bezat na 30fps s uplne nezodpovedajucim vizualom. Hura podme teda na nej otestovat karty a urobit z toho zaver, ze do buducna je maxvel shit, odpad, podvod na zakaznikov a GTX970 bude ligat -40% oproti 390ke. Tak?

2. 3dmark - draw calls... tajomne API overhead

3. Plackanie nad architekturou maxvelu part 2

Ci sa toto niekde prejavi REALNE na vykone v hrach, nevie este ani moja vestecka gula ale ty to uz automaticky podavas ako fakty a este si k tomu neodpustis zopar fanboyovskych slovnych spojeni... fakt sa to neda citat.