Moderátoři: morke, Walker1134, Wilik
To je dost odvazne tvrzeni. Na zaklade ceho ?HEAD píše:Kartu brzdi propustnost.Proto neni na 1600mhz o 50% vykonejsi jak furyx.Je to podobny jako u pascalu, ale v mnohem horsi mire pac Vega pouziva delta color kompresi z polarisu a ta stoji za hovno.
Těch do applu bude minimum, budou jen v top-end Imacu a tam sloužej jako OpenCL akcelerátory, hrát se na nich nebude (díky bohu že....).godlike píše:Zere to 2x vic nez rok stara 1080, topi vic, je draha, co dodat. Nechapu. AMD ma vubec stesti, ze asi levne dodava Applu a levny Polarisy jsou dobry na mining. Nevim s cim chteji do budoucna prorazit.
Tak už fiji škálovala s OC HBM a Vega má prostupnost stejnou. 2x takovej takt ale jen polovina stacků, a pokud nezapracovali na správě paměti tam logicky musí bejt bottleneck.HEAD píše:Kartu brzdi propustnost.Proto neni na 1600mhz o 50% vykonejsi jak furyx.Je to podobny jako u pascalu, ale v mnohem horsi mire pac Vega pouziva delta color kompresi z polarisu a ta stoji za hovno.
No pokud to dobře sleduju, tak před x mesici při prvních ukázkách. Ne ted, to si ted vymysleli fans.Srandista píše: Uz x-krat bolo spominane, ze drivre boli derivovane z Fury driverov.
https://forum.beyond3d.com/threads/amd- ... st-1990723This might just be my ignorance of how drivers work so please help me out. I'm not sure how this could simply be a case of treating the hardware as if it were a different generation. Unless the hardware is being abstracted away by some layer between the driver and the GPU, how could a "Fiji driver" even work on Vega and be expected to give the same results?
For the results to be so bafflingly identical due to the driver alone, this would indicate some form of emulation where the underlying hardware features don't matter. Results would be based on what's exposed via some sort of HAL that is exactly as performant on Vega as it is Fiji, with some of Vega's special sauce simply turned off at this stage. Given the differences between NCU vs GCN I would be astonished if that were the case. In other words, what's being rumored around the interwebs is that AMD spent years of dev time in order to create a GPU emulator that runs in fiji-mode (or whatever other driver they want to run?) on Vega instead of proper, to-the-metal drivers for the brand new hardware layout? I'm not ready to buy into that just yet.
The results make little sense either way based on predictions of improvements that were coming. I feel it's more likely that AMD pushed back the release until they were finally seeing the IPC get to the level where Fiji has been in order to release a platform that wasn't absolutely regressive. There should hopefully be quite a bit of headroom left there for software optimizations but it might be another year before we see what it's really capable of. I would love to be wrong, I do think driver versions will have everything to do with improvements, I just can't honestly believe a "wait til you see if run on actual vega drivers" type conspiracy.
RYS AMD RTG:The "Fiji fallback driver" or "Fiji drivers" meme needs to stop. That's not how it works or should be described. Otherwise you should start calling Volta drivers Pascal drivers Maxwell drivers. There is obviously commonality -- that's just how software engineering needs to work for a GPU -- but calling it a Fiji driver or Fiji fallback is wrong.
Spíš čekám, že se stane něco jako s R9 390X vůči R9 290X, trochu to poladějí. Ale též bych nečekal žádné zázrakyDOC_ZENITH píše:15-20% je moc, takový takty z toho nedsoanou, už Frontier sedí na TDP limitu a jede skoro na max. V driverech očekávám posun nula nula prd, žádnej unicorn nepřijde. Já čekám 5-10% čistě na taktech.
