je to jako vzdy free preklad, takze chyb tam je plno
je to vytrzeny z kontextu tak vi buh... no-x to jiste doladi a upresni a berte to s rezervou, pac tomu panovi nerozumim
no-X píše:
Q: Speaking about better library designs, do you think it would benefit you to move to AMD's tools/libraries/processes later on?
A: I won't comment on future projects, but that is now a real possibility.
Mluvite o lepsich navrzich knihoven, myslite ze to pro vas bude vyhodou presunout je do amd nastroju/knihoven/procesu pozdeji?
Nebudu komentovat projekty, ale nyni je to skutecne mozne.
---
Q: ...the merger with AMD presented situations that might have in some ways impacted the roll out of R600...
A: When we heard about it last summer, the chip was pretty much finished and on its way to manufacturing, or even back. ... We do interact with a lot of new engineers now ... From a time to launch, marketing, PR standpoint -- It's possible those were affected...
... je na snade, ze spojeni s amd ovlivnilo v urcitych smerech vydani r600...
Kdyz jsme o tom dozvedeli minule leto, chip byl vesmes hotov a byl na sve ceste do vyroby, mozna uz i na ceste zpet... Jsme ted ovlivneni mnoha novymi inzenyry... od casu k vydani, marketingem, z hlediska PR... je mozne, ze se to na tom projevi...
---
Q: Is there a difference in those considerations when moving from say, 90->80nm or 90->65nm?
A: Well, all physical aspects depend on the technology. This includes the elements that make up ASICs (stdcell, memories, macros, etc...). We need to re-characterize and update all libraries and models associated with those. This can mean new synthesis, new P&R rules, new macro designs, etc... This affects everything from the netlist on down, but it can even have ramifications up through the architecture, as you might need to change some functionality, re-pipeline, change your memory usage, for example. It's a lot of work. - 1/2 nodes are a little easier, as most things can be simply optically shrunk -- You have to design with that in mind, but it can be easier. Famous last words.
Je zde nejaky rozdil v uvahach/navrzich kdyz se prechazi z 90->80nm nebo z 90->65nm?
Hmm, vsechno zalezi na technologii. Zahrnuje to prvky ktere tvori ASICs(stdcell, pameti, makra, atd.) Musime znovu navrhnout a predelat vsechny knihovny a modely spojene s timto. To muze znamenat nova slouceni, nova P&R pravidla, novy navrh maker atd... Ovlivni to vsechno od "netlist on down?"(tak to fakt netusim), ale muze to take vest ke zmenam architektury, predelani piplines, vyuziti pameti.... Spousta prace. polovicni nodes(uzly - osvetli asi no-x)jsou o neco "lehci" a stejne tak spousta veci muze byt jednoduse opticky zestihlena. (famous last words? no, kdybych tomu rozumnel tak mozna jo)
---
Q: Do they have a group within ATI strictly working on future technologies?
A: We sort of did at the time we started R600, but we clearly have a group for that now.
---
Maji(spis mate - asi myslel AMD) nejakou skupinu uvnitr ATI primo urcenou pro budouci technologie?
jakousi skupinu jsme meli uz kdyz jsme zacinali s r600... ted jiz mame pro to skupinu(mlzi
)
---
Q: Which team at ATI creates the requirements and how soon in the development process are they created? 1 year? 2 years? 3 years?!?
A: A new architecture like R6xx took over 3 years of design work. 36 months for a new design is probably a minimum, at this time, considering the shear number of transistors (and associated design) involved. Follow on products, typically follow within 6 months or so, and don't need a huge amount of redesign. In the last year of design, very little change occurs -- Configs, features, etc... are all locked down. Only clock can change a little afterwards.
Ktery team v ATI stvoril pozadavky a jak brzo v procesu vyvoje se tvorili? Rok, dva, tri?
Nova architektura jako r600 zabere kolem 3 let navrharske prace. 36 mesicu pro novy navrh je pravdepodobne minimum, v soucasnosti s prihlednutim na mnozstvi tranzistoru a prislusneho navrhu. (ted fakt netusim co chtel rict... Nasledujici produkty (revize?!?), obvykle nasleduji po 6 mesicich nebo tak a nepotrebuji zadne velke predelavky.) V poslednim roce navrhu se daji delat jiz jen velmi male zmeny - configy, vychytavky, atd... vsechno je uzamceno. Pouze takty se daji menit trochu pozdeji.
---
Q: A13 revision was fabricated in January, but released in May...
A: Most of the work was on the drivers -- 2 OS's, new arch, DX9, DX10 and OGL. Work is still ongoing. I'm not sure on the silicon dates, off hand. But at some point, it just made more sense to launch all at the same time.
Revize a13 byla vyrobena v lednu, ale vysla az v kvetnu....
Bylo spousty prace na ovladacich - 2 operacni systemy, nova architektura, dx9, dx10 a opengl. Stale na tom pracujeme. (??? Ale nekde to dava vetsi smysl vydavat vse najednou...)
---
Q: So, can we conclude, that resolving AA samples in INT type (non-HDR type) of surfaces is actually under-utilizing overall throughput capacity of moving data, inside the chip?
A: In a way, yes. But it's not simple -- Some busses are under utilized, but that doesn't mean you could easily go faster. For example, we only filter in float; but having int doesn't make it go faster, since the number of units doesn't change (we actually promote most things to float).
* NO-X parketa pro tebe....AA, samples, typ int a pozdeji i float, surface - go on....
Muzeme soudit, ze "resolving AA vzorky typu INT (ne typu hdr) povrchu" momentalne nedostatecne vyuzivaji celkovou vykonovou kapacitu prenosu dat uvnitr chipu?
Svym zpusobem ano. Ale neni to jednoduchy. Nektere sbernice jsou nevytizene, ale to neznamena, ze je muzete lehce donutit pracovat rychleji. Priklad - my pouze filtrujeme ve float, ale ikdyby to byl int, nebude to rychlejsi, dokud se nezmeni pocet jednotek(momentalne prosazujeme vice float)
---
Q: I saw that in tech specs for HD2900 there is listed "bicubic fitlering"...
A: Bicubic is supported, and was intended for some OS font features (I think). It's not exposed at this time. I don't believe that DX allows it to be exposed. It does not operate at the same rate as bilinear. Might show up in OGL.
Vsiml jsem si v technicke specifikaci k HD2900 ze je vyjmenovan bikubicky filtr...
ano je podporovan a byl zamyslen pro nejakou OS "vychytavku" fontu(aspon myslim). nyni neni pristupny a neverim, ze dx dovoli aby byl pouzit. nepracuje stejne "vykonne" jako bilinearni. Mohl by predvest v OGL.??
---
Q: AF drop is very bad...
A: The AF hit can be more significant on R600 at this time. It should show higher quality in exchange (not on the LOD isotropy front, but on the mipmap transition front). I believe some of that will be addressed, in part, in future drivers. I can't promise a % improvement at this time.
tohle nechapu, pac on mu oznamuje, ze AF propad je velmi spatnej a Eric odpovida, ze AF "hit"(nechapu co to ma bejt) muze byt jeste dulezitejsi. Muze ukazat vetsi kvalitu ve vymene(ne v LOD isotropy fron, ale v mipmap transition front???) Veri, ze to bude adresovano?/zahrnotu v budoucich ovladacich, nemuze slibit % narust...
---
Q: why does the MSAA algortihm cost so much frames? The memory interface is much wider than R580, the compression algoritm are better and the ROPs can do more color/z compares than R580.
A: The actual number of units working on AA wrt to R580 isn't substantially different - thought the new ones are focused on 64b pixels. There are cases where moving the resolve to the shader has hurt performance -- That's typically where the frame rate is very high (i.e. > 100fps). Those are cases we were aware of and only really show up in benchmarks, not so much in real game play. It sort of caps the max fps to 200~300 or so, depending on resolution. Again, not something we felt would seriously impact real game play. It can also be mitigated by selecting the proper AA modes.
As for workloads in various apps, yes, it can be improved in some cases. Without specifics, it's hard to say. Overall, I expect the *scaling* of AA to be similar to slightly worst than R580, for "regular" pixels, but still to be higher than R580 in absolute terms.
But anytime 64b pixels will be invoked (i.e. HDR games), I expect performance of R600 to be substantially higher than R580.
proc algoritmus MSAA "stoji" tolik framu? pametove rozhrani(sbernice?) je mnohem sirsi nez u r580, kompresni algoritmy jsou lepsi a rops umi vice coloz/z porovnani nez r580
aktualni pocet jednotek pracujicich na AA vuci r580 neni nijak podstatne rozdilny - mysleno tak, ze nove jsou zamereny na 64b pixly. nastavaji tak pripady kde prenost vysledku k shaderum srazi vykon. to je typicke zejmena tam kde je fps vysoke(vetsi nez 100). ty jsou vsak pripady, ktere se projevi jen v benchmarcich, ne tak v realnem hrani. ??? to je jakysi strop max fps 200-300 v zavislosti na rozliseni ??? znovu rika, ze to skutecne nema dopad na hrani a takjako tak se da jeste snizit spravnou volbou aa modu.
A ohledne vytizeni v ruznych aplikacich, ano, muze byt vylepseno v nekterych pripadech. Bez specifikace je tezke rict. Celkove, ocekavam, ze *stupnovani* aa bude podobne nebo mirne horsi nez u r580, pro *regularni* pixly, ale porad bude vyzsi nez u r580 v absolutnich podminkach/clenech.
ale kdykoliv se budou pozadovat 64b pixly(hdr hry), ocekavam vykon r600 podstatne vyzsi nez u r580
---
Q: However, you can't deny that R5xx/RV5xx are a perfect example of what you were talking about with features taking die space and going unused.
A: Nope, it's generally simply being late to the game. By then, the channel is full and the OEMs have their deals. Getting deals is simply not possible (most OEMs for mainstream and below) or very difficult (above mainstream, you need to show a huge delta). It's not a competition when you are late. R5xx was fine and price/performance was quite excellent. It was, at least, 4 months late.
Nicmene, nemuze poprit ze r5xx/rv5xx jsou skvelym prikladem toho o cem jste mluvil, vlastnostech "zabirajicich misto(na chipu asi) a zustanou nevyuzity".
ne, pouze proste pozde vstoupi do hry. ???? zatim, "kanal" je plny a oems maji sve dohody/casti, ziskat je neni mozne(mnoho oems pro mainstream a pod) nebo velmi tezke(nad mainstreamem, musite ukazat velkou deltu)
sorry totalne mi unika co chtel rict. Neni motivace/soutezivost kdyz uz je pozde. r5xx byla dobra a cena/vykon byl pomerne excelentni. To bylo, nejmene, 4 mesice pozde/zpozdeno - god knows....
---
Q: The implication is that R600's design and hardware attributes are fantastic and the drivers are poo-pooing all over performance.
A: I did not mean to imply that. It's a good and reasonable design. The drivers are very stable and deliver the expected performance in many cases. There are certainly quite a few left to address. If someone is expecting a 2x boost in all apps, well, I'm sorry, but it won't happen. There's a couple of places where that kind of delta is possible, and a lot where 10 ~30% is possible too. But we are shipping a good product at an awesome price. From a functional aspect, there's more to do too (such as more custom filters, fixing some of the video stuff, etc...)
???
muze byt, ze r600 navrh a hardwarove parametry jsou fantasticke a ovladace "hazej bobek/s*rou - slusne recenou" na vykon...
nic takoveho jsem nenaznacil. je to dobry a rozumny navrh. ovladace jsou velmi stabilni a prinasi ocekavany vykon v mnoha pripadech. ??? Jestli nekdo ocekava dvounasobny narust ve vsech aplikaci, tak sorry, ale to se nestane. Je jeste par mist kde je to takovy narust mozny a spousta jinych kde je mozny narust 10-30%. ale my "nabizime(?)" dobry produkt za strasnou cenu(?). z funkcniho hlediska, je jeste hodne co delat, jako jsou uzivatelske filtry, oprava nejakeho video materialu, atd...????
---
Q: Where has R6xx demonstrated that its ALU:TEX ratios are more optimal than the competition's?
A: In yesterday's games, and even a lot of today's game, the ALU:TEX ratio selected is probably overkill. But when I look at some of the newest games, or some coming down the pipe, or even some of our demoes, a ratio of 4:1 seems pale -- Shaders are coming with 10:1, 20:1 ratios -- Even with the most advance filters, those apps are still ALU bound. That is more where we shot for.
---
Kde predvedl r6xxx ze ALU:TEX pomery/prevody jsou optimalnejsi nez u konkurence...
ve starsich hrach, i ve spouste soucasnych her, alu:tex pomer je pravdepodobne znicujici. Kdyz se ale podivam na nektere novejsi hry, nebo nektere "jdouci dolu rourou" (boze co je zase tohle) nebo na nektera nase dema, pomer 4:1 se zda mdly - shadery prichazi s pomery 10:1, 20:1. stejne i s mnoha pokrocilymi filtry, tyto aplikace jsou stale ALU svazane.
Q: Was there ever a set of simulations run comparing R600 + R580 RBEs vs the final design, and if so what were the interesting datapoints that came out?
A: Yes, there were hundreds. Yes, a few things came out in the final design, but, in general, it was as expected. -- Resolve was moved into the shader, so that can't really get compared.
Byla vubec nejaka sada simulaci, ktera by porovnala r600 + r580 rbes s finalnim navrhem a jestli ano, jaka zajimava zjisteni z nich vysla...
Ano bylo jich stovky. Neco malo veci vyslo ve finalnim navrhu, ale v zasade, to bylo tak ocekavano. rozhodnuti padlo na shadery a to se nedalo porovnat....
... srry za veskery errory, ale tohle bylo na me moc. kdyz uz sem zacal, tak jsem to dojel, ale never ever.
no-xe, protoze potvrzoval no-xovy myslenky a to no-x neni na vyplatni pasce ATI a nema pristup k internim materialum
]