{"id":122838,"date":"2025-04-02T12:55:26","date_gmt":"2025-04-02T10:55:26","guid":{"rendered":"https:\/\/prohoster.info\/?p=122838"},"modified":"2025-04-02T12:55:26","modified_gmt":"2025-04-02T10:55:26","slug":"novaya-statya-radeon-rx-9070-xt-protiv-geforce-rtx-5070-ti","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/novaya-statya-radeon-rx-9070-xt-protiv-geforce-rtx-5070-ti","title":{"rendered":"Nov\u00fd \u010dl\u00e1nek: Radeon RX 9070 XT vs GeForce RTX 5070 Ti","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Vlastnosti GPU rodiny Blackwell a testy GeForce RTX 5080<\/a><\/noindex> nazna\u010duj\u00ed, \u017ee v\u0161echny modely 50. s\u00e9rie, krom\u011b GeForce RTX 5090, p\u0159edstavuj\u00ed jen m\u00edrn\u00fd upgrade odpov\u00eddaj\u00edc\u00edch model\u016f minul\u00e9ho generace. Situaci zhor\u0161ily nedostatky nov\u00fdch grafick\u00fdch karet a n\u00e1sledn\u011b astronomick\u00e9 ceny.<\/p>\n

V\u011bt\u0161ina kupuj\u00edc\u00edch za\u0159\u00edzen\u00ed vy\u0161\u0161\u00ed kategorie, nad kter\u00fdmi m\u00e1 monopol NVIDIA, vzbuzuje pouze teoretick\u00fd z\u00e1jem. Na jedn\u00e9 \u00farovni n\u00ed\u017e v\u0161ak technologick\u00e1 stagnace \u201ezelen\u00e9\u201c grafiky vytvo\u0159ila situaci, kdy by AMD mohla jedn\u00edm skokem dohnat zpo\u017ed\u011bn\u00ed v problematick\u00fdch aspektech architektury RDNA, jako je sledov\u00e1n\u00ed paprsk\u016f a upscale. Nicm\u00e9n\u011b, proroctv\u00ed, \u017ee AMD obnov\u00ed paritu s konkurentem (i kdy\u017e bez n\u00e1rok\u016f na vlajkov\u00fd sektor), sl\u00fdch\u00e1me ji\u017e dlouho, a poka\u017ed\u00e9 zn\u00ed st\u00e1le m\u00e9n\u011b p\u0159esv\u011bd\u010div\u011b, proto ozn\u00e1men\u00ed grafick\u00fdch karet Radeon RX 9000 na pozad\u00ed bou\u0159liv\u00e9 \u010dinnosti NVIDIA nep\u0159il\u00e1kalo velkou pozornost. P\u0159itom AMD provedla rozs\u00e1hlou pr\u00e1ci na odstran\u011bn\u00ed chyb a pokud je neodstranila, alespo\u0148 kompenzovala nedostatky minul\u00fdch GPU.<\/p>\n

Pohled na Radeon RX 9070 XT p\u0159edstavujeme v neobvykl\u00e9m form\u00e1tu srovn\u00e1n\u00ed pro 3DNews, proto\u017ee \u201e\u010derven\u00e1\u201c novinka a jej\u00ed hlavn\u00ed konkurent \u2013 GeForce RTX 5070 Ti \u2013 se dostaly do na\u0161ich rukou z\u00e1rove\u0148, a jedno za\u0159\u00edzen\u00ed nelze posuzovat bez druh\u00e9ho, nebo\u0165 v Rusku se prod\u00e1vaj\u00ed za p\u0159ibli\u017en\u011b stejn\u00e9 pen\u00edze.<\/p>\n\n\n\n
\"Nov\u00fd <\/td>\n <\/td>\n \"Nov\u00fd <\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

<\/center> \t\t\t\t <\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex> Grafick\u00fd procesor Navi 48 a architektura RDNA 4<\/strong><\/h2>\n

O tom, \u017ee se NVIDIA stala monopolistou na trhu vysoce v\u00fdkonn\u00fdch GPU a chov\u00e1 se tomuto odpov\u00eddaj\u00edc\u00edm zp\u016fsobem, ne\u0159ekl jen lenoch. M\u00e9n\u011b se zam\u011b\u0159uje na skute\u010dnost, \u017ee v\u0161echny spole\u010dnosti, kter\u00e9 navrhuj\u00ed velk\u00e9 procesory, jsou nuceny vyu\u017e\u00edvat slu\u017eby jednoho dodavatele \u2013 tajwansk\u00e9 TSMC. Popt\u00e1vka po kapacit\u00e1ch TSMC se jist\u011b stala jedn\u00edm z d\u016fvod\u016f, pro\u010d AMD zastavila v\u00fdvoj velk\u00fdch spot\u0159ebitelsk\u00fdch GPU, a v \u0159ad\u011b \u201e\u010derven\u00fdch\u201c grafick\u00fdch procesor\u016f nov\u00e9 generace je zat\u00edm pouze jeden \u010dip \u2013 Navi 48.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
V\u00fdrobce<\/th>\nAMD<\/th>\n<\/tr>\n
N\u00e1zev<\/td>\nNavi 32<\/td>\nNavi 31<\/td>\nNavi 48<\/td>\n<\/tr>\n
Kde se pou\u017e\u00edv\u00e1<\/td>\n\n

Radeon RX 7700 XT;<\/p>\n

Radeon RX 7800 XT<\/p>\n<\/td>\n

\n

Radeon RX 7900 GRE;<\/p>\n

Radeon RX 7900 XT;<\/p>\n

Radeon RX 7900 XTX<\/p>\n<\/td>\n

\n

Radeon RX 9070;<\/p>\n

Radeon RX 9070 XTX<\/p>\n<\/td>\n<\/tr>\n

Architektura<\/td>\nRDNA 3<\/td>\nRDNA 3<\/td>\nRDNA 4<\/td>\n<\/tr>\n
Technologick\u00fd proces, nm<\/td>\nTSMC N5\/N6<\/td>\nTSMC N5\/N6<\/td>\nTSMC N4P<\/td>\n<\/tr>\n
Po\u010det tranzistor\u016f, mld<\/td>\n28,1<\/td>\n57,7<\/td>\n53,9<\/td>\n<\/tr>\n
Plocha \u010dipu, mm2<\/td>\n346<\/td>\n522<\/td>\n357<\/td>\n<\/tr>\n
Po\u010det CU\/WGP\/SE<\/td>\n<\/tr>\n
V\u00fdpo\u010detn\u00ed jednotky (CU)<\/td>\n60<\/td>\n96<\/td>\n64<\/td>\n<\/tr>\n
Pracovn\u00ed procesory (WGP)<\/td>\n30<\/td>\n48<\/td>\n32<\/td>\n<\/tr>\n
Shaderov\u00e9 pole (SA)<\/td>\nNe<\/td>\nNe<\/td>\nNe<\/td>\n<\/tr>\n
Shaderov\u00e9 motory (SE)<\/td>\n4?<\/td>\n6<\/td>\n4<\/td>\n<\/tr>\n
Konfigurace v\u00fdpo\u010detn\u00ed jednotky<\/td>\n<\/tr>\n
Vektorov\u00e9 ALU (FP32)<\/td>\n2 \u00d7 32<\/td>\n2 \u00d7 32<\/td>\n2 \u00d7 32<\/td>\n<\/tr>\n
Vektorov\u00e9 ALU (FP32\/INT32)<\/td>\n2 \u00d7 32<\/td>\n2 \u00d7 32<\/td>\n2 \u00d7 32<\/td>\n<\/tr>\n
Skal\u00e1rn\u00ed ALU<\/td>\n2<\/td>\n2<\/td>\n2<\/td>\n<\/tr>\n
ALU speci\u00e1ln\u00edho ur\u010den\u00ed (SFU)<\/td>\n2 \u00d7 8<\/td>\n2 \u00d7 8<\/td>\n2 \u00d7 8<\/td>\n<\/tr>\n
Jednotky sledov\u00e1n\u00ed paprsk\u016f (Ray Accelerators)<\/td>\n1<\/td>\n1<\/td>\n1<\/td>\n<\/tr>\n
Jednotky p\u0159ekr\u00fdv\u00e1n\u00ed textur (TMU)<\/td>\n4<\/td>\n4<\/td>\n4<\/td>\n<\/tr>\n
Vektorov\u00e9\/skl\u00e1rn\u00ed registry, KB<\/td>\n384\/10<\/td>\n384\/10<\/td>\n384\/16<\/td>\n<\/tr>\n
Velikost cache L0, KB<\/td>\n32<\/td>\n32<\/td>\n32<\/td>\n<\/tr>\n
Konfigurace Workgroup Processor (WGP)<\/td>\n<\/tr>\n
Velikost Shared Memory, KB<\/td>\n128<\/td>\n128<\/td>\n128<\/td>\n<\/tr>\n
Velikost cache instrukc\u00ed, KB<\/td>\n32<\/td>\n32<\/td>\n32<\/td>\n<\/tr>\n
Velikost skal\u00e1rn\u00ed cache, KB<\/td>\n16<\/td>\n16<\/td>\n16<\/td>\n<\/tr>\n
V\u00fdpo\u010detn\u00ed bloky GPU<\/td>\n<\/tr>\n
Vektorov\u00e9 ALU (FP32)<\/td>\n3 840 (7 680)<\/td>\n6 144 (12 288)<\/td>\n4 096 (8 192)<\/td>\n<\/tr>\n
Jednotky p\u0159ekr\u00fdv\u00e1n\u00ed textur (TMU)<\/td>\n240<\/td>\n384<\/td>\n256<\/td>\n<\/tr>\n
Jednotky rasterizace (ROP)<\/td>\n96<\/td>\n192<\/td>\n128<\/td>\n<\/tr>\n
Jednotky sledov\u00e1n\u00ed paprsk\u016f (Ray Accelerators)<\/td>\n60<\/td>\n96<\/td>\n64<\/td>\n<\/tr>\n
Konfigurace pam\u011bti<\/td>\n<\/tr>\n
Velikost cache L1, KB<\/td>\n2 048<\/td>\n3 072<\/td>\n2 048<\/td>\n<\/tr>\n
Velikost cache L2, MB<\/td>\n4<\/td>\n6<\/td>\n8<\/td>\n<\/tr>\n
Velikost Infinity Cache, MB<\/td>\n64<\/td>\n96<\/td>\n64<\/td>\n<\/tr>\n
\u0160\u00ed\u0159ka sb\u011brnice VRAM, bit<\/td>\n256<\/td>\n384<\/td>\n256<\/td>\n<\/tr>\n
Typ \u010dip\u016f VRAM<\/td>\nGDDR6<\/td>\nGDDR6<\/td>\nGDDR6<\/td>\n<\/tr>\n
<\/td>\n<\/tr>\n
Rozhran\u00ed PCI Express<\/td>\n4.0 x16<\/td>\n4.0 x16<\/td>\n5.0 x16<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Mezi star\u0161\u00edmi \u010dipy se nejv\u00edce podob\u00e1 modelu druh\u00e9 \u00farovn\u011b, Navi 32, podle z\u00e1kladn\u00edch charakteristik, ale obsahuje t\u00e9m\u011b\u0159 tolik tranzistor\u016f jako Navi 31 (53,9 a 57,7 miliard). Navi 48 je vyrobena technologi\u00ed TSMC N4P, kter\u00e1 pat\u0159\u00ed k t\u00e9\u017ee fotolitografick\u00e9 technologii 5 nm, kter\u00e1 byla zvl\u00e1dnuta ji\u017e p\u0159ed dv\u011bma lety. Nicm\u00e9n\u011b Navi 48 zab\u00edr\u00e1 pouze 68 % plochy Navi 31 d\u00edky tomu, \u017ee AMD se vzdala \u010dipletov\u00e9ho designu, proto\u017ee MCD \u010diplety, kter\u00e9 obsahuj\u00ed cache t\u0159et\u00ed \u00farovn\u011b a \u0159adi\u010de pam\u011bti, jsou vyr\u00e1b\u011bny ve hrub\u0161\u00edm technologick\u00e9m procesu N6.<\/p>\n

\"Nov\u00fd <\/p>\n

Architektura nov\u00e9ho grafick\u00e9ho procesoru, RDNA 4, se struktur\u00e1ln\u011b neli\u0161\u00ed od p\u0159edchoz\u00edch iterac\u00ed. Navi 48 obsahuje 64 v\u00fdpo\u010detn\u00edch jednotek, tedy 4 096 shaderov\u00fdch ALU kompatibiln\u00edch s FP32, co\u017e je zcela typick\u00e1 konfigurace pro GPU druh\u00e9ho tieru \u2013 stejn\u011b jako 256bitov\u00e1 sb\u011brnice videopam\u011bti. Nicm\u00e9n\u011b AMD v hodnocen\u00ed v\u00fdkonu stav\u00ed Navi 48 proti vlajkov\u00e9mu modelu Navi 31, kter\u00fd m\u00e1 6 144 shaderov\u00fdch ALU standardn\u00ed p\u0159esnosti. Rozd\u00edl by m\u011bl b\u00fdt v\u00edce ne\u017e kompenzov\u00e1n mnoha vylep\u0161en\u00edmi logiky, a pro jejich pochopen\u00ed doporu\u010dujeme osv\u011b\u017eit pam\u011b\u0165 ohledn\u011b architektury RDNA 3, kterou jsme podrobn\u011b rozeb\u00edrali v recenzi. Radeon RX 7900 XTX<\/a><\/noindex>.<\/p>\n

Ve srovn\u00e1n\u00ed s \"zelenou\" architekturou Blackwell<\/a><\/noindex> RDNA 4 p\u0159edstavuje rozs\u00e1hlej\u0161\u00ed a hloubkovou aktualizaci p\u0159edchoz\u00edch prac\u00ed. Av\u0161ak popis v\u0161ech novinek se p\u0159ekvapiv\u011b skl\u00e1d\u00e1 do kr\u00e1tk\u00e9ho seznamu, proto\u017ee NVIDIA tradi\u010dn\u011b v\u011bnuje zv\u00fd\u0161enou pozornost slo\u017eit\u00fdm softwarov\u00fdm nebo poloprogramov\u00fdm \u0159e\u0161en\u00edm, kter\u00e1 vy\u017eaduj\u00ed podrobn\u00e9 koment\u00e1\u0159e, zat\u00edmco inovace AMD obvykle za\u010d\u00ednaj\u00ed a kon\u010d\u00ed v hardwaru.<\/p>\n

\"Nov\u00fd <\/p>\n

Fundament\u00e1ln\u00ed zm\u011bny se odehr\u00e1ly v konstrukci v\u00fdpo\u010detn\u00ed jednotky \u2013 ned\u011bliteln\u00fd v\u00fdpo\u010detn\u00ed blok, kter\u00fd je analogem SM u NVIDIA a jader CPU. Pl\u00e1nova\u010d instrukc\u00ed nyn\u00ed podporuje tzv. d\u011blen\u00e9 bari\u00e9ry, kter\u00e9 umo\u017e\u0148uj\u00ed p\u0159ed\u00e1vat instrukce k vykon\u00e1n\u00ed b\u011bhem jejich platnosti. D\u00e1le byla optimalizov\u00e1na p\u0159edb\u011b\u017en\u00e1 volba instrukc\u00ed z cache a skalarov\u00e9 ALU, kter\u00e9 se pou\u017e\u00edvaj\u00ed pro \u00fasporn\u00e9 vykon\u00e1v\u00e1n\u00ed operac\u00ed podm\u00edn\u011bn\u00e9ho skoku a v\u011btven\u00ed, p\u0159evzaly schopnost pracovat s re\u00e1ln\u00fdmi \u010d\u00edsly z mobiln\u00ed architektury RDNA 3.5. Sou\u010dasn\u011b se tak\u00e9 zv\u00fd\u0161il objem skalarov\u00fdch registr\u016f: z 10 na 16 KB.<\/p>\n

\"Nov\u00fd <\/p>\n

Jedn\u00edm z nejd\u016fle\u017eit\u011bj\u0161\u00edch novinek RDNA 4 je zdvojen\u00e9 tempo vykon\u00e1v\u00e1n\u00ed maticov\u00fdch instrukc\u00ed WMMA, kter\u00e9 se pou\u017e\u00edvaj\u00ed p\u0159i \u00faloh\u00e1ch strojov\u00e9ho u\u010den\u00ed a zpracov\u00e1n\u00ed dat pomoc\u00ed neuronov\u00fdch s\u00edt\u00ed. Krom\u011b toho sada instrukc\u00ed RDNA 4 obsahuje p\u0159\u00edkazy pro v\u00fdpo\u010dty nad strukturovan\u011b \u0159\u00eddk\u00fdmi maticemi a 8bitov\u00fdmi daty s plovouc\u00ed desetinnou \u010d\u00e1rkou (FP8 a BF8).<\/p>\n

\"Nov\u00fd <\/p>\n

Nicm\u00e9n\u011b propustnost 1 024 FLOP za takt znamen\u00e1, \u017ee \u010dipy Navi pouze dos\u00e1hly \u00farovn\u011b minul\u00e9 \"zelen\u00e9\" architektury \u2013 Ampere. Nav\u00edc, zat\u00edmco \u010dipy NVIDIA a Intel pou\u017e\u00edvaj\u00ed pro maticov\u00e9 operace samostatn\u00e9 pole ALU, v architektu\u0159e RNDA 4 tato pr\u00e1ce nad\u00e1le le\u017e\u00ed na shaderov\u00fdch SIMD.<\/p>\n

Je v\u0161ak t\u0159eba m\u00edt na pam\u011bti, \u017ee v re\u00e1ln\u00fdch aplikac\u00edch se grafick\u00e9 procesory zalo\u017een\u00e9 na RDNA 3 nikdy nedostanou ani k pln\u00e9mu vyt\u00ed\u017een\u00ed. Uv\u00e1d\u011bn\u00e9 teraflopy v FP32 v\u00fdpo\u010dtech obecn\u00e9ho ur\u010den\u00ed lze dos\u00e1hnout pouze za p\u0159edpokladu, \u017ee se pou\u017e\u00edv\u00e1 form\u00e1t VOPD (Vector Operation Dual), kter\u00fd umo\u017e\u0148uje zabalit dv\u011b nez\u00e1visl\u00e9 vektorov\u00e9 instrukce dohromady. A ten s\u00e1m o sob\u011b ukl\u00e1d\u00e1 \u0159adu omezen\u00ed, proto polovina 32-bitov\u00fdch ALU uvnit\u0159 Compute Unit \u010dasto z\u016fst\u00e1v\u00e1 nevyu\u017eita. P\u0159i sm\u00ed\u0161en\u00e9 z\u00e1t\u011b\u017ei (nap\u0159. hra s upscalingem) matri\u010dn\u00ed operace samoz\u0159ejm\u011b ukroj\u00ed \u010d\u00e1st v\u00fdpo\u010detn\u00edch zdroj\u016f u shader\u016f, ale tyto ztr\u00e1ty nebudou tak velk\u00e9, jak se na prvn\u00ed pohled zd\u00e1.<\/p>\n\n\n\n\n\n\n
<\/th>\nCompute Unit (AMD RDNA 3)<\/th>\nCompute Unit (AMD RDNA 4)<\/th>\nStreaming Multiprocessor (NVIDIA Ada Lovelace)<\/th>\nStreaming Multiprocessor (NVIDIA Blackwell)<\/th>\nXe-core (Intel Xe-HPG)<\/th>\nXe-core (Intel Xe2)<\/th>\n<\/tr>\n
V\u00fdkonov\u00e9 bloky<\/td>\n\n

2 \u00d7 SIMD32 (FP32\/INT32);<\/p>\n

2 \u00d7 SIMD32 (FP32);<\/p>\n

2 \u00d7 SIMD2 (FP64);<\/p>\n

2 \u00d7 SIMD8 (SFU);<\/p>\n

2 \u00d7 skal\u00e1rn\u00edch ALU<\/p>\n<\/td>\n

\n

2 \u00d7 SIMD32 (FP32\/INT32);<\/p>\n

2 \u00d7 SIMD32 (FP32);<\/p>\n

2 \u00d7 SIMD2 (FP64);<\/p>\n

2 \u00d7 SIMD8 (SFU);<\/p>\n

2 \u00d7 skal\u00e1rn\u00edch ALU<\/p>\n<\/td>\n

\n

4 \u00d7 SIMD16 (FP32\/INT32);<\/p>\n

4 \u00d7 SIMD16 (FP32);<\/p>\n

2 \u00d7 SISD? (FP64);<\/p>\n

4 \u00d7 SIMD4 (SFU);<\/p>\n

4 \u00d7 skal\u00e1rn\u00edch ALU;<\/p>\n

4 \u00d7 tenzorov\u00fdch jader<\/p>\n<\/td>\n

\n

8 \u00d7 SIMD16 (FP32\/INT32);<\/p>\n

2 \u00d7 SISD? (FP64);<\/p>\n

4 \u00d7 SIMD4 (SFU);<\/p>\n

4 \u00d7 skal\u00e1rn\u00edch ALU;<\/p>\n

4 \u00d7 tenzorov\u00fdch jader<\/p>\n<\/td>\n

\n

16 \u00d7 SIMD8 (FP32);<\/p>\n

16 \u00d7 SIMD8 (INT32);<\/p>\n

16 \u00d7 SISD (FP64);<\/p>\n

16 \u00d7 SIMD2 (SFU);<\/p>\n

16 \u00d7 XMX<\/p>\n<\/td>\n

\n

8 \u00d7 SIMD16 (FP32);<\/p>\n

8 \u00d7 SIMD16 (INT32);<\/p>\n

8 \u00d7 SIMD2 (FP64);<\/p>\n

8 \u00d7 SIMD4 (SFU);<\/p>\n

8 \u00d7 XMX<\/p>\n<\/td>\n<\/tr>\n

Operace na lince SIMD za takt<\/td>\n\n

128 \u00d7 FP32;<\/p>\n

64 \u00d7 INT32;<\/p>\n

256 \u00d7 FP16;<\/p>\n

4 \u00d7 FP64;<\/p>\n

16 \u00d7 transcen-dentn\u00ed funkce<\/p>\n<\/td>\n

\n

128 \u00d7 FP32;<\/p>\n

64 \u00d7 INT32;<\/p>\n

256 \u00d7 FP16;<\/p>\n

4 \u00d7 FP64;<\/p>\n

16 \u00d7 transcen-dentn\u00ed funkce<\/p>\n<\/td>\n

\n

128 \u00d7 FP32;<\/p>\n

64 \u00d7 INT32;<\/p>\n

128 \u00d7 FP16;<\/p>\n

2 \u00d7 FP64;<\/p>\n

16 \u00d7 transcen-dentn\u00ed funkce<\/p>\n<\/td>\n

\n

128 \u00d7 FP32;<\/p>\n

128 \u00d7 INT32;<\/p>\n

128 \u00d7 FP16;<\/p>\n

2 \u00d7 FP64;<\/p>\n

16 \u00d7 transcen-dentn\u00ed funkce<\/p>\n<\/td>\n

\n

128 \u00d7 FP32;<\/p>\n

128 \u00d7 INT32;<\/p>\n

256 \u00d7 FP16;<\/p>\n

16 \u00d7 FP64;<\/p>\n

32 \u00d7 transcen-dentn\u00ed funkce<\/p>\n<\/td>\n

\n

128 \u00d7 FP32;<\/p>\n

128 \u00d7 INT32;<\/p>\n

256 \u00d7 FP16;<\/p>\n

16 \u00d7 FP64;<\/p>\n

32 \u00d7 transcen-dentn\u00ed funkce<\/p>\n<\/td>\n<\/tr>\n

Maticov\u00e9 operace, FLOP za takt (FP16)<\/td>\n512<\/td>\n1 024<\/td>\n2 048<\/td>\n2 048<\/td>\n2 048<\/td>\n2 048<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Zvy\u0161ov\u00e1n\u00ed sn\u00edmk\u016f pomoc\u00ed neuronov\u00e9 s\u00edt\u011b \u2014 u\u017e to nen\u00ed teoretick\u00e1 situace pro akceler\u00e1tory Radeon. Nov\u00e1, \u010dtvrt\u00e1 verze FSR kone\u010dn\u011b spol\u00e9h\u00e1 na strojov\u00e9 u\u010den\u00ed, v\u010detn\u011b generov\u00e1n\u00ed sn\u00edmk\u016f, a vyu\u017e\u00edvan\u00e9 modely p\u0159edstavuj\u00ed hybridn\u00ed variantu konvolu\u010dn\u00edch s\u00edt\u00ed a transformer\u016f. Bohu\u017eel na rozd\u00edl od p\u0159edchoz\u00edch verz\u00ed upscaleru, kter\u00e9 pracovaly prakticky na jak\u00e9mkoli hardwaru, FSR 4 podporuj\u00ed pouze grafick\u00e9 procesory architektury RDNA 4 (v souvislosti s roz\u0161\u00ed\u0159en\u00edm sady instrukc\u00ed WMMA pro data FP8). Univerz\u00e1ln\u00ed re\u017eim se zjednodu\u0161en\u00fdmi v\u00fdpo\u010detn\u00edmi j\u00e1dry, jako v Intel XeSS, nen\u00ed k dispozici.<\/p>\n

\"Nov\u00fd <\/p>\n

Dal\u0161\u00edm sm\u011brem reformy architektury RDNA je ray tracing. Rychlost testov\u00e1n\u00ed pr\u016fse\u010d\u00edku paprsku s BVH boxem a troj\u00faheln\u00edkem se zdvojn\u00e1sobila na jednoho Ray Accelerator (a tedy i jednu Compute Unit): z 4 a 1 na takt na 8 a 2. Nicm\u00e9n\u011b RT blok \u010dip\u016f Intel Battlemage pracuje rychlost\u00ed 18 a 2, zat\u00edmco grafick\u00e9 procesory Blackwell nach\u00e1zej\u00ed hned \u010dty\u0159i pr\u016fse\u010d\u00edky paprsku s troj\u00faheln\u00edkem na takt a nezve\u0159ejn\u011bn\u00fd po\u010det pr\u016fse\u010d\u00edk\u016f paprsku s BVH boxem. U konkurent\u016f se testy dvou typ\u016f (proti BVH boxu a troj\u00faheln\u00edku) prov\u00e1d\u011bj\u00ed sou\u010dasn\u011b, zat\u00edmco AMD st\u00e1le ne. Do jist\u00e9 m\u00edry je paralelismus mo\u017en\u00fd, proto\u017ee ka\u017ed\u00fd Ray Accelerator nyn\u00ed obsahuje dva nez\u00e1visl\u00e9 \u201etester\u201dy, ale ne na pln\u00e9 rychlosti.<\/p>\n

\"Nov\u00fd <\/p>\n

Krom\u011b toho Ray Accelerator z\u00edskal mo\u017enost zpracov\u00e1n\u00ed paprsk\u016f mimo po\u0159ad\u00ed. Paprsky, kter\u00e9 nevy\u017eaduj\u00ed p\u0159\u00edstup k pomal\u00fdm \u00farovn\u00edm z\u00e1sobn\u00edku pam\u011bti, se zpracov\u00e1vaj\u00ed jako prvn\u00ed, a v\u00fdsledky testov\u00e1n\u00ed se ukl\u00e1daj\u00ed, aby shaderov\u00fd program je obdr\u017eel ve o\u010dek\u00e1van\u00e9m po\u0159ad\u00ed.<\/p>\n

Logika raytracingu nyn\u00ed pou\u017e\u00edv\u00e1 hust\u0161\u00ed strukturu BVH8 nam\u00edsto BVH4 (osm nam\u00edsto \u010dty\u0159 uzl\u016f na ka\u017edou v\u011btev hierarchie objem\u016f), co\u017e zvy\u0161uje rychlost pr\u016fchodu a sou\u010dasn\u011b sni\u017euje tlak na videopam\u011b\u0165. Velikost BVH v pam\u011bti se d\u00e1le sn\u00ed\u017eila d\u00edky nov\u00e9mu algoritmu komprese primitiv. Kone\u010dn\u011b, AMD nab\u00edz\u00ed odd\u011blit orientaci BVH box\u016f od absolutn\u00edch sou\u0159adnic sc\u00e9ny. Ot\u00e1\u010den\u00ed box\u016f podle geometrie objekt\u016f umo\u017e\u0148uje sn\u00ed\u017eit po\u010det krok\u016f p\u0159i vyhled\u00e1v\u00e1n\u00ed.<\/p>\n

\"Nov\u00fd <\/p>\n

Je \u0161koda, \u017ee, navzdory v\u0161em uveden\u00fdm zm\u011bn\u00e1m, RDNA 4 st\u00e1le prov\u00e1d\u00ed proch\u00e1zen\u00ed BVH na shaderov\u00fdch ALU, a nikoli za pomoc\u00ed specializovan\u00e9 logiky. Podle klasifikace hardwarov\u00e9ho ray tracingu navr\u017een\u00e9 Imagination Technologies se \u0159e\u0161en\u00ed AMD nach\u00e1zej\u00ed na druh\u00e9 \u00farovni, grafick\u00e9 procesory Intel na t\u0159et\u00ed, zat\u00edmco hardware NVIDIA ji\u017e postupuje od \u00farovn\u011b 3 k \u00farovni 4 d\u00edky funkci SER \u2014 dynamick\u00e9mu seskupen\u00ed instruk\u010dn\u00edho proudu za \u00fa\u010delem zv\u00fd\u0161en\u00ed koherenc\u00ed pam\u011b\u0165ov\u00fdch p\u0159\u00edstup\u016f a t\u00edm sn\u00ed\u017eit latenci.<\/p>\n

\"Nov\u00fd <\/p>\n

Analog SER v architektu\u0159e RDNA 4 tak\u00e9 nen\u00ed, ale objevila se funkce Out of Order Memory, kter\u00e1 jinak pom\u00e1h\u00e1 vy\u0159e\u0161it stejn\u00fd \u00fakol. P\u0159edchoz\u00ed verze RDNA, pracuj\u00edc\u00ed s n\u011bkolika skupinami instrukc\u00ed (wavefronty v terminologii AMD), vrac\u00ed data z pam\u011bti striktn\u011b ve stejn\u00e9m po\u0159ad\u00ed, v jak\u00e9m byly po\u017eadavky u\u010din\u011bny. Kdy\u017e jsou data po\u017eadovan\u00e1 shaderem v rychl\u00e9 cache, n\u00e1vrat m\u016f\u017ee b\u00fdt p\u0159esto zpo\u017ed\u011bn, pokud d\u0159\u00edve sv\u016fj po\u017eadavek u\u010dinil jin\u00fd shader, jeho\u017e data le\u017e\u00ed v relativn\u011b pomal\u00e9 VRAM. Pravda je, \u017ee toto omezen\u00ed se zd\u00e1 b\u00fdt naprosto bezv\u00fdznamn\u00e9 v kontextu v\u00edcestup\u0148ov\u00e9ho v\u00fdpo\u010dtu, ale ur\u010dit\u011b vzniklo s c\u00edlem zjednodu\u0161it organizaci GPU a do\u010dasn\u011b (\u010dti: v rastrov\u00fdch hr\u00e1ch) z\u016fst\u00e1valo rozumn\u00fdm kompromisem. RDNA 4 umo\u017e\u0148uje skupin\u00e1m instrukc\u00ed vracet data nez\u00e1visle na po\u0159ad\u00ed po\u017eadavk\u016f.<\/p>\n

\"Nov\u00fd <\/p>\n

Out of Order Memory \u2014 jednou z zm\u011bn v architektu\u0159e RDNA 4. P\u0159id\u011blov\u00e1n\u00ed registr\u016f Compute Unitov\u00fdm shaderov\u00fdm program\u016fm d\u0159\u00edve prob\u00edhalo staticky, s ohledem na maxim\u00e1ln\u00ed spot\u0159ebu b\u011bhem \u017eivotnosti shaderu, co\u017e znamen\u00e1, \u017ee ur\u010dit\u00e1 \u010d\u00e1st registru z\u016fst\u00e1vala pr\u00e1zdn\u00e1. RDNA 4 umo\u017e\u0148uje shader\u016fm \u017e\u00e1dat a uvol\u0148ovat registry dynamicky. Takto se vyu\u017eit\u00ed objemu registru st\u00e1v\u00e1 ekonomi\u010dt\u011bj\u0161\u00ed.<\/p>\n

Zaj\u00edmav\u00e9 je, \u017ee Shader Engine \u2014 velk\u00e1 struktura GPU, kter\u00e1 kombinuje pole CU, rasteriz\u00e1tor, blok primitiv a bloky operac\u00ed rasterizace \u2014 v RDNA 4 p\u0159i\u0161el o vlastn\u00ed prvn\u00ed \u00farove\u0148 cache. Nicm\u00e9n\u011b ka\u017ed\u00fd segment cache L2 obsahuje dvakr\u00e1t v\u00edce bank. T\u00edm se zv\u00fd\u0161il nejen jej\u00ed objem, ale i propustnost, a to v obou sm\u011brech \u2014 k Shader Engine a k Infinity Cache (jin\u00fdmi slovy, L3). Objem samotn\u00e9ho Infinity Cache z\u016fstal stejn\u00fd jako u Navi 32 \u2014 64 MB.<\/p>\n

Navi 48 m\u00e1 16 linek sb\u011brnice PCI Express p\u00e1t\u00e9 generace, ale v hern\u00edch \u00faloh\u00e1ch nem\u016f\u017ee zcela vyu\u017e\u00edt tak rychl\u00fd rozhran\u00ed.<\/p>\n

\"Nov\u00fd <\/p>\n

Kone\u010dn\u011b, AMD oznamuje, \u017ee hardwarov\u00fd video kodek v nov\u00e9m GPU zaji\u0161\u0165uje vy\u0161\u0161\u00ed kvalitu k\u00f3dov\u00e1n\u00ed s obl\u00edben\u00fdm pro streamov\u00e1n\u00ed n\u00edzk\u00fdm bitrate, ale nic v\u00edc. To znamen\u00e1, \u017ee se st\u00e1le neobjevila podpora barevn\u00e9 subdiscretizace YUV 4:2:2, kterou maj\u00ed grafick\u00e9 procesory NVIDIA a Intel.<\/p>\n

Pokud jde o v\u00fdstupn\u00ed za\u0159\u00edzen\u00ed, \u0159adi\u010d displeje Navi 48 je kompatibiln\u00ed s nejnov\u011bj\u0161\u00edmi verzemi rozhran\u00ed DisplayPort a HDMI: 2.1a a 2.1b.<\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>Technick\u00e9 specifikace, ceny<\/strong><\/h2>\n

Radeon RX 9070 XT je vybaven pln\u011b funk\u010dn\u00edm \u010dipem Navi 48 se 64 Compute Units, co\u017e znamen\u00e1 4 096 plovouc\u00edch ALU standardn\u00ed p\u0159esnosti. T\u011b\u017eko naj\u00edt form\u00e1ln\u00edho p\u0159edch\u016fdce novinky mezi \"\u010derven\u00fdmi\" akceler\u00e1tory minul\u00fdch generac\u00ed, proto\u017ee se zm\u011bnila jak modelov\u00e1 \u0159ada, tak princip \u010d\u00edslov\u00e1n\u00ed GPU (\u010dipy Navi d\u0159\u00edve nikdy nedosahovaly osmi\u010dky). Podle vzorce v\u00fdpo\u010detn\u00edch blok\u016f a charakteristik pam\u011b\u0165ov\u00e9ho z\u00e1sobn\u00edku je Radeon RX 9070 XT podobn\u00fd modelu Radeon RX 7800 XT, ale AMD ho propaguje jako n\u00e1hradu za Radeon RX 7900 XTX.<\/p>\n

Frekvence Game Clock a Boost Clock (c\u00edlov\u00e1 taktovac\u00ed frekvence v hr\u00e1ch a limit taktovac\u00ed frekvence) grafick\u00e9ho procesoru v\u00fdrazn\u011b vzrostly ve srovn\u00e1n\u00ed s parametry vlajkov\u00e9 lodi p\u0159edchoz\u00ed generace a \u010din\u00ed 2,4 a 2,97 GHz. Teoretick\u00fd v\u00fdkon Radeon RX 7900 XTX p\u0159i maxim\u00e1ln\u00ed taktovac\u00ed frekvenci je o 26 % vy\u0161\u0161\u00ed, ale AMD m\u00e1 v \u00famyslu tento rozd\u00edl kompenzovat vylep\u0161en\u00edmi architektury RDNA 4.<\/p>\n

Radeon RX 9070 XT obdr\u017eel 16 GB pam\u011bti GDDR6 s propustnost\u00ed 20 Gb\/s na 256bitov\u00e9 sb\u011brnici. Podle zpr\u00e1v AMD zva\u017eovalo zv\u00fd\u0161en\u00ed objemu VRAM na 32 GB, ale nakonec pova\u017eovalo takovou konfiguraci za nadbyte\u010dnou pro spot\u0159ebitelsk\u00e1 \u0159e\u0161en\u00ed. \u0160koda, proto\u017ee n\u00e1roky her neust\u00e1le rostou, a pr\u00e1v\u011b objem grafick\u00e9 pam\u011bti byl d\u016fle\u017eitou konkuren\u010dn\u00ed v\u00fdhodou mnoha p\u0159edchoz\u00edch za\u0159\u00edzen\u00ed AMD.<\/p>\n

Referen\u010dn\u00ed energetick\u00e1 spot\u0159eba Radeon RX 9070 XT je 304 W, doporu\u010den\u00e1 cena \u2014 599 $ (o 100 $ v\u00edce ne\u017e u Radeon RX 7800 XT).<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
V\u00fdrobce<\/th>\nAMD<\/th>\nNVIDIA<\/th>\n<\/tr>\n
Model<\/td>\nRadeon RX 7900 XT<\/td>\nRadeon RX 7900 XTX<\/td>\nRadeon RX 9070<\/td>\nRadeon RX 9070 XT<\/td>\nGeForce RTX 4070 Ti SUPER<\/td>\nGeForce RTX 4080 SUPER<\/td>\nGeForce RTX 5070 Ti<\/td>\nGeForce RTX 5080<\/td>\n<\/tr>\n
Grafick\u00fd procesor<\/td>\n<\/tr>\n
N\u00e1zev<\/td>\nNavi 31 XT<\/td>\nNavi 31 XTX<\/td>\nNavi 48 XT<\/td>\nNavi 48 XTX<\/td>\nAD103<\/td>\nAD103<\/td>\nGB203<\/td>\nGB203<\/td>\n<\/tr>\n
Architektura<\/td>\nRDNA 3<\/td>\nRDNA 3<\/td>\nRDNA 4<\/td>\nRDNA 4<\/td>\nAda Lovelace<\/td>\nAda Lovelace<\/td>\nBlackwell<\/td>\nBlackwell<\/td>\n<\/tr>\n
Technologick\u00fd proces<\/td>\nTSMC N5\/N6<\/td>\nTSMC N5\/N6<\/td>\nTSMC N4C<\/td>\nTSMC N4C<\/td>\nTSMC 4N<\/td>\nTSMC 4N<\/td>\nTSMC 4NP<\/td>\nTSMC 4NP<\/td>\n<\/tr>\n
Po\u010det tranzistor\u016f, mld<\/td>\n57,7<\/td>\n57,7<\/td>\n53,9<\/td>\n53,9<\/td>\n45,9<\/td>\n45,9<\/td>\n45,6<\/td>\n45,6<\/td>\n<\/tr>\n
Taktovac\u00ed frekvence (Base Clock\/Game Clock\/Boost Clock), MHz<\/td>\n1 500 \/2 025\/2 394<\/td>\n1 855 \/2 269\/2 499<\/td>\n1 330 \/2 070\/2 520<\/td>\n1 660 \/2 400\/2 970<\/td>\n2 310\/2 610<\/td>\n2 205\/2 550<\/td>\n2 295\/2 452<\/td>\n2 295\/2 617<\/td>\n<\/tr>\n
Shader ALU (FP32)<\/td>\n5 376 (10 752)<\/td>\n6 144 (12 228)<\/td>\n3 584 (7 168)<\/td>\n4 096 (8 192)<\/td>\n8 448<\/td>\n10 240<\/td>\n8 960<\/td>\n10 752<\/td>\n<\/tr>\n
Jednotky p\u0159ekr\u00fdv\u00e1n\u00ed textur (TMU)<\/td>\n336<\/td>\n384<\/td>\n224<\/td>\n256<\/td>\n264<\/td>\n320<\/td>\n280<\/td>\n336<\/td>\n<\/tr>\n
Jednotky rasterizace (ROP)<\/td>\n192<\/td>\n192<\/td>\n128<\/td>\n128<\/td>\n112<\/td>\n112<\/td>\n96<\/td>\n112<\/td>\n<\/tr>\n
Tensorov\u00e1 j\u00e1dra<\/td>\nNe<\/td>\nNe<\/td>\nNe<\/td>\nNe<\/td>\n264<\/td>\n320<\/td>\n280<\/td>\n336<\/td>\n<\/tr>\n
RT-j\u00e1dra<\/td>\n84<\/td>\n96<\/td>\n112<\/td>\n128<\/td>\n66<\/td>\n80<\/td>\n70<\/td>\n84<\/td>\n<\/tr>\n
Objem posledn\u00ed \u00farovn\u011b cache, MB<\/td>\n80<\/td>\n96<\/td>\n64<\/td>\n64<\/td>\n48<\/td>\n64<\/td>\n64<\/td>\n64<\/td>\n<\/tr>\n
Opera\u010dn\u00ed pam\u011b\u0165<\/td>\n<\/tr>\n
\u0160\u00ed\u0159ka sb\u011brnice, bit<\/td>\n320<\/td>\n384<\/td>\n256<\/td>\n256<\/td>\n256<\/td>\n256<\/td>\n256<\/td>\n256<\/td>\n<\/tr>\n
Typ \u010dipu<\/td>\nGDDR6 SGRAM<\/td>\nGDDR6 SGRAM<\/td>\nGDDR6 SGRAM<\/td>\nGDDR6 SGRAM<\/td>\nGDDR6X SGRAM<\/td>\nGDDR6X SGRAM<\/td>\nGDDR7 SGRAM<\/td>\nGDDR7 SGRAM<\/td>\n<\/tr>\n
Propustnost na kontaktn\u00ed pin, Gb\/s<\/td>\n20<\/td>\n20<\/td>\n20<\/td>\n20<\/td>\n21<\/td>\n23<\/td>\n28<\/td>\n30<\/td>\n<\/tr>\n
Celkov\u00e1 propustnost, GB\/s<\/td>\n800<\/td>\n960<\/td>\n640<\/td>\n640<\/td>\n672<\/td>\n736<\/td>\n896<\/td>\n960<\/td>\n<\/tr>\n
Objem, GB<\/td>\n20<\/td>\n24<\/td>\n16<\/td>\n16<\/td>\n16<\/td>\n16<\/td>\n16<\/td>\n16<\/td>\n<\/tr>\n
V\u00fdkon<\/td>\n<\/tr>\n
Maxim\u00e1ln\u00ed v\u00fdkon FP32, TFLOPS<\/td>\n25,7 (51,5)<\/td>\n30,7 (61,1)<\/td>\n18,1 (36,1)<\/td>\n24,3 (48,7)<\/td>\n44,1<\/td>\n52,2<\/td>\n43,9<\/td>\n56,3<\/td>\n<\/tr>\n
V\u00fdkon FP64\/FP32<\/td>\n1\/32 (1\/64)<\/td>\n1\/32 (1\/64)<\/td>\n1\/32 (1\/64)<\/td>\n1\/32 (1\/64)<\/td>\n1\/64<\/td>\n1\/64<\/td>\n1\/64<\/td>\n1\/64<\/td>\n<\/tr>\n
V\u00fdkon FP16\/FP32<\/td>\n2\/1<\/td>\n2\/1<\/td>\n2\/1<\/td>\n2\/1<\/td>\n1\/1<\/td>\n1\/1<\/td>\n1\/1<\/td>\n1\/1<\/td>\n<\/tr>\n
Ostatn\u00ed<\/td>\n<\/tr>\n
PCI Express sb\u011brnice<\/td>\nPCI Express 4.0 x16<\/td>\nPCI Express 4.0 x16<\/td>\nPCI Express 5.0 x16<\/td>\nPCI Express 5.0 x16<\/td>\nPCI Express 4.0 x16<\/td>\nPCI Express 4.0 x16<\/td>\n5.0 x16<\/td>\n5.0 x16<\/td>\n<\/tr>\n
Rozhran\u00ed v\u00fdstupu obrazu<\/td>\nDisplayPort 2.1, HDMI 2.1a<\/td>\nDisplayPort 2.1, HDMI 2.1a<\/td>\nDisplayPort 2.1a, HDMI 2.1b<\/td>\nDisplayPort 2.1a, HDMI 2.1b<\/td>\nDisplayPort 1.4a, HDMI 2.1<\/td>\nDisplayPort 1.4a, HDMI 2.1<\/td>\nDisplayPort 2.1b, HDMI 2.1b<\/td>\nDisplayPort 2.1b, HDMI 2.1b<\/td>\n<\/tr>\n
TDP\/TBP, W<\/td>\n315<\/td>\n335<\/td>\n220<\/td>\n304<\/td>\n285<\/td>\n320<\/td>\n300<\/td>\n360<\/td>\n<\/tr>\n
Doporu\u010den\u00e1 cena (USA), $<\/td>\n899 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n999 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n549 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n599 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n799 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n999 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n749 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n999 (doporu\u010den\u00e1 p\u0159i uveden\u00ed na trh)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Na druhou stranu, GeForce RTX 5070 Ti je zalo\u017eena na stejn\u00e9m grafick\u00e9m procesoru GB203, kter\u00fd se pou\u017e\u00edv\u00e1 v GeForce RTX 5080, ale mlad\u0161\u00ed model p\u0159i\u0161el o 14 streamov\u00fdch multiprocesor\u016f, co\u017e znamen\u00e1 1 792 FP32-kompatibiln\u00edch shaderov\u00fdch ALU, a hern\u00ed taktov\u00e1 frekvence byla sn\u00ed\u017eena. V d\u016fsledku toho d\u011bl\u00ed GeForce RTX 5080 od RTX 5070 Ti 27 % teoretick\u00e9ho v\u00fdkonu, a sama GeForce RTX 5070 Ti se v tomto parametru neli\u0161\u00ed od RTX 4070 Ti SUPER. Stejn\u011b jako v p\u0159\u00edpad\u011b Radeon RX 9070 XT m\u00e1 nedostatek surov\u00e9 v\u00fdpo\u010detn\u00ed s\u00edly kompenzovat nov\u00e1 architektura GPU, p\u0159i\u010dem\u017e zejm\u00e9na funkce generov\u00e1n\u00ed v\u00edcero sn\u00edmk\u016f pomoc\u00ed DLSS (a\u017e t\u0159i vygenerovan\u00e9 sn\u00edmky mezi dv\u011bma \"skute\u010dn\u00fdmi\"). <\/p>\n

GeForce RTX 5070 Ti m\u00e1 spot\u0159ebu 300 W a s doporu\u010denou cenou 749 USD je o 50 USD levn\u011bj\u0161\u00ed ne\u017e GeForce RTX 4070 Ti SUPER.<\/p>\n

Rozd\u00edl ve $150 doporu\u010den\u00e9 m\u011bnov\u00e9 hodnot\u011b mezi Radeon RX 9070 XT a GeForce RTX 5070 Ti nen\u00ed patrn\u00fd podle aktu\u00e1ln\u00edch rusk\u00fdch cen: nejlevn\u011bj\u0161\u00ed partnersk\u00e9 modely jsme na\u0161li v DNS za 99 999 a 104 999 rubl\u016f. Pr\u00e1v\u011b tyto budeme testovat v benchmarkech. Radeon RX 9070 XT p\u0159edstavuje grafick\u00e1 karta GIGABYTE GAMING OC s tov\u00e1rn\u00edm p\u0159etaktov\u00e1n\u00edm GPU a\u017e na 2,52\/3,06 GHz, zat\u00edmco GeForce RTX 5070 Ti \u2013 akceler\u00e1tor Palit GamingPro s referen\u010dn\u00edmi specifikacemi.<\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>GIGABYTE Radeon RX 9070 XT GAMING OC: konstrukce<\/strong><\/h2>\n

Radeon RX 9070 XT v proveden\u00ed GIGABYTE m\u00e1 pom\u011brn\u011b skromn\u00e9 rozm\u011bry pro grafickou kartu se spot\u0159ebou minim\u00e1ln\u011b (a jak uk\u00e1\u017eou testy, mnohem v\u011bt\u0161\u00ed) 300 W: 288 mm na d\u00e9lku a t\u0159i sloty na \u0161\u00ed\u0159ku. Jedin\u00fdm prvkem LED podsv\u00edcen\u00ed je kr\u00e1tk\u00fd pruh na boku. \u0160t\u00edtek s logem v\u00fdrobce je pohybliv\u00fd: pokud se posune na LED pruh, logo za\u010dne sv\u00edtit.<\/p>\n

\"Nov\u00fd <\/p>\n

Chladic\u00ed syst\u00e9m zaji\u0161\u0165uj\u00ed t\u0159i ventil\u00e1tory s pr\u016fm\u011brem lopatek 87 mm. Vzduch proch\u00e1z\u00ed malou sekc\u00ed chladi\u010de skrze otvor v zadn\u00edm panelu.<\/p>\n

\"Nov\u00fd <\/p>\n

Grafick\u00fd procesor a \u010dipy video pam\u011bti odv\u00e1d\u011bj\u00ed teplo do odpa\u0159ovac\u00ed komory, ale prvn\u00ed p\u0159\u00edmo, zat\u00edmco posledn\u00ed skrze P-typovou kovovou podlo\u017eku. K lamel\u00e1m chladi\u010de jsou p\u0159ipevn\u011bny samostatn\u00e9 desky, kter\u00e9 slou\u017e\u00ed jako teplosm\u011bnn\u00e9 prvky VRM komponent. Teplovodn\u00edch trubi\u010dek je celkem sedm, jejich pr\u016fm\u011br je 6 mm. Upozor\u0148ujeme, \u017ee GIGABYTE d\u00e1v\u00e1 p\u0159ednost tekut\u00fdm tepeln\u00fdm past\u00e1m p\u0159ed b\u011b\u017en\u00fdmi.<\/p>\n

\"Nov\u00fd <\/p>\n

Zadn\u00ed deska se pod\u00edl\u00ed na odv\u00e1d\u011bn\u00ed tepla z ti\u0161t\u011bn\u00e9ho obvodu pomoc\u00ed n\u011bkolika termopad\u016f (tady u\u017e b\u011b\u017en\u00fdch, nikoli kapalinov\u00fdch).<\/p>\n

\"Nov\u00fd <\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>GIGABYTE Radeon RX 9070 XT GAMING OC: ti\u0161t\u011bn\u00fd obvod<\/strong><\/h2>\n

N\u011bkter\u00e9 varianty Radeon RX 9070 XT p\u0159ij\u00edmaj\u00ed nap\u00e1jen\u00ed p\u0159es konektor 12V-2\u00d76, ale GIGABYTE dala p\u0159ednost t\u0159em univerz\u00e1ln\u00edm osmikan\u00e1lov\u00fdm konektor\u016fm. Vzhledem k vysok\u00e9 spot\u0159eb\u011b energie grafick\u00e9 karty a neust\u00e1l\u00fdm st\u00ed\u017enostem na nov\u00fd konektor je to spolehliv\u011bj\u0161\u00ed, i kdy\u017e ne tak pohodln\u00e9 a estetick\u00e9 \u0159e\u0161en\u00ed. Z 17 f\u00e1z\u00ed nap\u00e1jec\u00edho syst\u00e9mu je 14, \u0159\u00edzen\u00fdch PWM kontrol\u00e9rem Monolithic Power Systems MP2868A, vyhrazeno pro grafick\u00fd procesor, zb\u00fdvaj\u00edc\u00ed t\u0159i pro video pam\u011b\u0165. Oba VRM jsou vybaveny nap\u00e1jec\u00edmi stupni MP87993 s p\u0159edpokl\u00e1danou jmenovitou hodnotou proudu 90A.<\/p>\n

VRAM je slo\u017eena z \u010dip\u016f SK hynix, v jejich ozna\u010den\u00ed H56G42AS8DX-014 je zak\u00f3dov\u00e1na p\u0159enosov\u00e1 kapacita 20 Gb\/s.<\/p>\n

Krom\u011b kopie BIOSu, kter\u00e1 je ve v\u00fdchoz\u00edm nastaven\u00ed aktivn\u00ed, m\u00e1 grafick\u00e1 karta rezervn\u00ed \"tichou\" verzi firmwaru.<\/p>\n

\"Nov\u00fd \"Nov\u00fd <\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>Palit GeForce RTX 5070 Ti GamingPro: konstrukce<\/strong><\/h2>\n

Palit GeForce RTX 5070 Ti GamingPro m\u011b\u0159\u00ed 331,9 mm na d\u00e9lku, ale zauj\u00edm\u00e1 pouze t\u0159i sloty pro roz\u0161\u00ed\u0159en\u00ed. Pod logem GamingPro a vedlej\u0161\u00edm segmentem \u010deln\u00edho panelu se nach\u00e1zej\u00ed RGB LED diody. Pr\u00e1ci LED lze synchronizovat s z\u00e1kladn\u00ed deskou prost\u0159ednictv\u00edm standardn\u00edho ARGB konektoru um\u00edst\u011bn\u00e9ho vedle 12V-2\u00d76 nap\u00e1jec\u00edho vstupu.<\/p>\n

\"Nov\u00fd <\/p>\n

Bo\u010dn\u00ed plochy sk\u0159\u00edn\u011b vytv\u00e1\u0159\u00ed lit\u00fd hlin\u00edkov\u00fd r\u00e1m s ventila\u010dn\u00edmi otvory na dlouh\u00fdch stran\u00e1ch. Na ochrann\u00e9 desce na zadn\u00ed stran\u011b PCB je obvykl\u00e1 m\u0159\u00ed\u017eka pro pr\u016fchoz\u00ed ventilaci chladi\u010de. Lopatky ventil\u00e1tor\u016f maj\u00ed pr\u016fm\u011br 94 mm.<\/p>\n

\"Nov\u00fd <\/p>\n

Na dn\u011b chladi\u010de je um\u00edst\u011bna odpa\u0159ovac\u00ed komora dostate\u010dn\u00e9 plochy, kter\u00e1 pokr\u00fdv\u00e1 j\u00e1dro GPU a okoln\u00ed \u010dipy videopam\u011bti, p\u0159i\u010dem\u017e pro komponenty VRM jsou ur\u010deny samostatn\u00e9 ploch\u00e9 tepeln\u011b vodiv\u00e9 desky. Lamely radi\u00e1toru jsou pro\u0159\u00edznuty na osmi tepeln\u00e9 trubky o pr\u016fm\u011bru 6 mm.<\/p>\n

\"Nov\u00fd <\/p>\n

Kovov\u00fd backplate pln\u00ed pouze ochrannou a dekorativn\u00ed funkci, proto\u017ee mezi n\u00edm a PCB nen\u00ed ani jedna teplovodiv\u00e1 podlo\u017eka.<\/p>\n

\"Nov\u00fd <\/p>\n

Grafick\u00e1 karta je dod\u00e1v\u00e1na s adapt\u00e9rem z dvou osmipinov\u00fdch nap\u00e1jec\u00edch konektor\u016f na 12V-2\u00d76 a mal\u00fdm l\u00e1tkov\u00fdm podlo\u017ekou pod my\u0161.<\/p>\n

\"Nov\u00fd <\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>Palit GeForce RTX 5070 Ti GamingPro: ti\u0161t\u011bn\u00fd obvod<\/strong><\/h2>\n

Na rozd\u00edl od grafick\u00e9 karty Palit GameRock<\/a><\/noindex>, kterou jsme prozkoumali v ned\u00e1vn\u00e9m p\u0159ehledu GeForce RTX 5080, je dal\u0161\u00ed nejvy\u0161\u0161\u00ed model \u0159ady postaven\u00fd na prostorn\u00e9 PCB, ale nap\u00e1jec\u00ed syst\u00e9m je jednodu\u0161\u0161\u00ed. Grafick\u00e9mu procesoru pat\u0159\u00ed 15f\u00e1zov\u00fd VRM \u0159\u00edzen\u00fd PWM regul\u00e1torem Monolithic Power Systems MP29816. Nap\u00e1jen\u00ed pam\u011b\u0165ov\u00fdch \u010dip\u016f GDDR7 je t\u0159\u00edf\u00e1zov\u00e9, na b\u00e1zi regul\u00e1toru MP2988. Oba regul\u00e1tory jsou vybaveny tranzistorov\u00fdmi sestavami MP87993 s p\u0159edpokl\u00e1dan\u00fdm jmenovit\u00fdm proudem 90 A.<\/p>\n

Ozna\u010den\u00ed \u010dip\u016f videopam\u011bti Samsung K4VAF325ZC-SC28 nazna\u010duje propustnost 28 Gbit\/s.<\/p>\n

Stejn\u011b jako Radeon RX 9070 XT v proveden\u00ed GIGABYTE, m\u00e1 akceler\u00e1tor Palit dva varianty BIOS \u2013 \"v\u00fdkonn\u00fd\" a \"tich\u00fd\".<\/p>\n

\"Nov\u00fd \"Nov\u00fd <\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>Testovac\u00ed stanice, metodika testov\u00e1n\u00ed<\/strong><\/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Testovac\u00ed stanice<\/th>\n<\/tr>\n
CPU<\/td>\nAMD Ryzen 9 7950X3D (PBO +150 MHz, CU -20)<\/td>\n<\/tr>\n
Z\u00e1kladn\u00ed deska<\/td>\nASUS ROG Crosshair X670E Hero<\/td>\n<\/tr>\n
Opera\u010dn\u00ed pam\u011b\u0165<\/td>\nG.Skill Trident Z5 Neo RGB (F5-6000J3040G32GX2-TZ5NR), 2 \u00d7 32 GB (6200 MT\/s, CL30)<\/td>\n<\/tr>\n
SSD<\/td>\nSolidigm P44 Pro, 2 TB<\/td>\n<\/tr>\n
Nap\u00e1jec\u00ed zdroj<\/td>\nCorsair AX1600i, 1600 W<\/td>\n<\/tr>\n
Syst\u00e9m chlazen\u00ed CPU<\/td>\nVlastn\u00ed vodn\u00ed chlazen\u00ed (EK-Quantum Velocity\u00b2 DDC 4.2 PWM D-RGB + EK-Quantum Surface X280M)<\/td>\n<\/tr>\n
Korpus<\/td>\nOtev\u0159en\u00fd testovac\u00ed stojan<\/td>\n<\/tr>\n
Opera\u010dn\u00ed syst\u00e9m<\/td>\nWindows 11 Pro<\/td>\n<\/tr>\n
Software pro GPU AMD<\/td>\n<\/tr>\n
V\u0161echny grafick\u00e9 karty<\/td>\nAMD Software Adrenalin Edition 25.2.1\/25.3.1<\/td>\n<\/tr>\n
Software pro GPU NVIDIA<\/td>\n<\/tr>\n
V\u0161echny grafick\u00e9 karty<\/td>\nNVIDIA GeForce Game Ready Driver 532.60\/532.70<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Hry bez ray tracingu<\/th>\n<\/tr>\n
Hra<\/td>\nAPI<\/td>\nMetoda testov\u00e1n\u00ed<\/td>\nNastaven\u00ed grafiky<\/td>\n<\/tr>\n
Alan Wake 2<\/td>\nDirectX 12<\/td>\nOCAT, lokalita Bright Falls<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Black Myth: Wukong<\/td>\nDirectX 12<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Cyberpunk 2077<\/td>\nDirectX 12<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
F1 24<\/td>\nDirectX 12<\/td>\nIntegrovan\u00fd benchmark, tra\u0165 Monaco (d\u00e9\u0161\u0165)<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Hogwarts Legacy<\/td>\nDirectX 12<\/td>\nOCAT, j\u00edzda na voz\u00edku v Path to Hogwarts<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Horizon Zero Dawn Remastered<\/td>\nDirectX 12<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Metro Exodus<\/td>\nDirectX 12<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita; Shading Rate: 100%<\/td>\n<\/tr>\n
Red Dead Redemption 2<\/td>\nVulkan<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Returnal<\/td>\nDirectX 12<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n
Total War: WARHAMMER III<\/td>\nDirectX 11<\/td>\nIntegrovan\u00fd benchmark (Mirrors of Madness Benchmark)<\/td>\nMax. grafick\u00e1 kvalita<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Hry s ray tracingem<\/th>\n<\/tr>\n
Hra<\/td>\nAPI<\/td>\nMetoda testov\u00e1n\u00ed<\/td>\nNastaven\u00ed grafiky<\/td>\nSkalov\u00e1n\u00ed sn\u00edmk\u016f<\/td>\n<\/tr>\n
AMD<\/td>\nIntel<\/td>\nNVIDIA<\/td>\n<\/tr>\n
Alan Wake 2<\/td>\nDirectX 12<\/td>\nOCAT, lokalita Bright Falls<\/td>\nMax. grafick\u00e1 kvalita, vysok\u00e1 kvalita ray tracingu<\/td>\nFSR Balanced<\/td>\nFSR Balanced<\/td>\nDLSS Balanced + Ray Reconstruction (+ Frame Generation)<\/td>\n<\/tr>\n
Black Myth: Wukong<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. grafick\u00e1 kvalita a ray tracing (Path Tracing)<\/td>\nFSR Balanced (+ Frame Generation)<\/td>\nXeSS Balanced\/FSR Balanced + Frame Generation<\/td>\nDLSS Vyv\u00e1\u017een\u00fd (+ generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\n<\/tr>\n
Cyberpunk 2077<\/td>\nVestav\u011bn\u00fd benchmark (OCAT pro generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\nMax. grafick\u00e1 kvalita a ray tracing (Path Tracing)<\/td>\nFSR Balanced (+ Frame Generation)<\/td>\nXeSS Balanced\/FSR Balanced + Frame Generation<\/td>\nDLSS Vyv\u00e1\u017een\u00fd (Transformer model) + Ray Reconstruction (+ generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\n<\/tr>\n
F1 24<\/td>\nIntegrovan\u00fd benchmark, tra\u0165 Monaco (d\u00e9\u0161\u0165)<\/td>\nMax. kvalita grafiky a sledov\u00e1n\u00ed paprsk\u016f<\/td>\nFSR Balanced (+ Frame Generation)<\/td>\nXeSS Balanced\/FSR Balanced + Frame Generation<\/td>\nDLSS Vyv\u00e1\u017een\u00fd (+ generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\n<\/tr>\n
Hogwarts Legacy<\/td>\nOCAT, j\u00edzda na voz\u00edku v Path to Hogwarts<\/td>\nMax. kvalita grafiky a sledov\u00e1n\u00ed paprsk\u016f<\/td>\nFSR Balanced<\/td>\nXeSS Vyv\u00e1\u017een\u00fd<\/td>\nDLSS Balanced + Ray Reconstruction (+ Frame Generation)<\/td>\n<\/tr>\n
Indiana Jones a Velk\u00fd kruh<\/td>\nOCAT, lokalita Sukhothai<\/td>\nMax. kvalita grafiky, vysok\u00e1 kvalita sledov\u00e1n\u00ed paprsk\u016f (Path Tracing); Ray Traced Shadows: Pouze slune\u010dn\u00ed sv\u011btlo<\/td>\nFSR Balanced (+ Frame Generation)<\/td>\nXeSS Balanced\/FSR Balanced + Frame Generation<\/td>\nDLSS Vyv\u00e1\u017een\u00fd (Transformer model) + Ray Reconstruction (+ generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\n<\/tr>\n
Metro Exodus Enhanced Edition<\/td>\nIntegrovan\u00fd benchmark<\/td>\nMax. kvalita grafiky a sledov\u00e1n\u00ed paprsk\u016f<\/td>\nN\/D<\/td>\nN\/D<\/td>\nDLSS Vyv\u00e1\u017een\u00fd<\/td>\n<\/tr>\n
Returnal<\/td>\nVestav\u011bn\u00fd benchmark (OCAT pro generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\nMax. kvalita grafiky a sledov\u00e1n\u00ed paprsk\u016f<\/td>\nFSR Balanced (+ Frame Generation)<\/td>\nXeSS Balanced\/FSR Balanced + Frame Generation<\/td>\nDLSS Vyv\u00e1\u017een\u00fd (+ generov\u00e1n\u00ed sn\u00edmk\u016f)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Ve v\u011bt\u0161in\u011b her se pr\u016fm\u011brn\u00e9 a minim\u00e1ln\u00ed (uv\u00e1d\u00edme 1. percentil rozd\u011blen\u00ed) sn\u00edmkov\u00e9 frekvence zobrazuj\u00ed z pole \u010dasu renderov\u00e1n\u00ed jednotliv\u00fdch sn\u00edmk\u016f nebo okam\u017eit\u00e9 sn\u00edmkov\u00e9 frekvence z\u00edskan\u00e9 pomoc\u00ed vestav\u011bn\u00e9ho benchmarku. V\u00fdjimkou jsou hry, kter\u00e9 nemaj\u00ed vestav\u011bn\u00fd benchmark, a testy zahrnuj\u00edc\u00ed generov\u00e1n\u00ed sn\u00edmk\u016f: v t\u011bchto p\u0159\u00edpadech pro zachycen\u00ed mezisn\u00edmov\u00fdch interval\u016f pou\u017e\u00edv\u00e1me program OCAT.<\/p>\n\n\n\n\n\n\n\n\n
Pracovn\u00ed aplikace<\/th>\n<\/tr>\n
Aplikace<\/td>\nBenchmark<\/td>\nNastaven\u00ed<\/td>\n<\/tr>\n
Adobe Premiere Pro 25.x<\/td>\nPugetBench pro Premiere Pro 1.1.1 (sestava test\u016f podle odkazu)<\/a><\/noindex><\/td>\nStandard (4K)<\/td>\n<\/tr>\n
Blender 4.x<\/td>\nDemo Agent 327 Barbershop z webu Blender<\/td>\nRenderer Cycles<\/td>\n<\/tr>\n
Blackmagic Design DaVinci Resolve Studio 19.x<\/td>\nPugetBench pro DaVinci Resolve 1.1.1 (sestava test\u016f podle odkazu)<\/a><\/noindex><\/td>\nStandard (4K); H.264\/HEVC re\u017eim k\u00f3dov\u00e1n\u00ed: Automaticky<\/td>\n<\/tr>\n
CAD aplikace<\/td>\nSPECviewperf 2020 v3.1<\/td>\nRozli\u0161en\u00ed obrazovky: 3840 \u00d7 2160<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Dek\u00f3dov\u00e1n\u00ed videa (ffmpeg 5.x)<\/th>\n<\/tr>\n
Form\u00e1t<\/td>\nRozli\u0161en\u00ed<\/td>\nParametry k\u00f3dov\u00e1n\u00ed<\/td>\nAPI<\/td>\n<\/tr>\n
H.264 (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\nHigh Profile, L4.1<\/td>\nD3D11VA<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\nHigh Profile, L5.1<\/td>\n<\/tr>\n
HEVC (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\nMain Profile, L4.0<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\nMain Profile, L5.0<\/td>\n<\/tr>\n
7680 \u00d7 4320<\/td>\nMain Profile, L6.0<\/td>\n<\/tr>\n
VP9 (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\nN\/D<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\n<\/tr>\n
7680 \u00d7 4320<\/td>\n<\/tr>\n
AV1 (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\nMain Profile, L4.0<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\nMain Profile, L5.0<\/td>\n<\/tr>\n
7680 \u00d7 4320<\/td>\nMain Profile, L6.0<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n\n\n\n\n\n\n\n\n\n\n\n\n
K\u00f3dov\u00e1n\u00ed videa (ffmpeg 5.x)<\/th>\n<\/tr>\n
Form\u00e1t<\/td>\nRozli\u0161en\u00ed<\/td>\nParametry k\u00f3dov\u00e1n\u00ed<\/td>\nAPI<\/td>\n<\/tr>\n
AMD<\/td>\nIntel<\/td>\nNVIDIA<\/td>\nAMD<\/td>\nIntel<\/td>\nNVIDIA<\/td>\n<\/tr>\n
H.264 (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\n-c:v h264_amf -quality speed -coder cabac -refs 1 -b:v 3M<\/td>\n-c:v h264_qsv -preset veryfast -profile:v main -level 4.1 -b:v 3M<\/td>\n-c:v h264_nvenc -preset fast -coder cabac -refs 1 -b:v 3M<\/td>\nAMF<\/td>\noneVPL<\/td>\nNVENC<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\n-c:v h264_amf -quality speed -coder cabac -refs 1 -b:v 7.5M<\/td>\n-c:v h264_qsv -preset veryfast -profile:v main -level 5.1 -b:v 7.5M<\/td>\n-c:v h264_nvenc -preset fast -coder cabac -refs 1 -b:v 7.5M<\/td>\n<\/tr>\n
HEVC (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\n-c:v hevc_amf -quality speed -b:v 3M<\/td>\n-c:v hevc_qsv -preset veryfast -tier main -b:v 3M<\/td>\n-c:v hevc_nvenc -preset fast -b:v 3M<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\n-c:v hevc_amf -quality speed -b:v 7.5M<\/td>\n-c:v hevc_qsv -preset veryfast -tier main -b:v 7.5M<\/td>\n-c:v hevc_nvenc -preset fast -b:v 7.5M<\/td>\n<\/tr>\n
7680 \u00d7 4320<\/td>\n-c:v hevc_amf -quality speed -b:v 20M<\/td>\n-c:v hevc_qsv -preset veryfast -tier main -b:v 20M<\/td>\n-c:v hevc_nvenc -preset fast -b:v 20M<\/td>\n<\/tr>\n
AV1 (YUV 4:2:0, 8 bit\u016f\/kan\u00e1l)<\/td>\n1920 \u00d7 1080<\/td>\n-c:v hevc_amf -quality speed -b:v 3M<\/td>\n-c:v av1_qsv -preset veryfast -profile main -b:v 3M<\/td>\n-c:v hevc_nvenc -preset fast -b:v 3M<\/td>\n<\/tr>\n
3840 \u00d7 2160<\/td>\n-c:v hevc_amf -quality speed -b:v 7.5M<\/td>\n-c:v av1_qsv -preset veryfast -profile main -b:v 7.5M<\/td>\n-c:v hevc_nvenc -preset fast -b:v 7.5M<\/td>\n<\/tr>\n
7680 \u00d7 4320<\/td>\n-c:v hevc_amf -quality speed -b:v 20M<\/td>\n-c:v av1_qsv -preset veryfast -profile main -b:v 20M<\/td>\n-c:v hevc_nvenc -preset fast -b:v 20M<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

V\u00fdkon grafick\u00fdch karet se zaznamen\u00e1v\u00e1 odd\u011blen\u011b od CPU a ostatn\u00edch komponent\u016f po\u010d\u00edta\u010de pomoc\u00ed za\u0159\u00edzen\u00ed NVIDIA PCAT. Jako z\u00e1t\u011b\u017e pro testy v\u00fdkonu a hladiny hluku se pou\u017e\u00edv\u00e1 hra Cyberpunk 2077 p\u0159i rozli\u0161en\u00ed 3840 \u00d7 2160 a maxim\u00e1ln\u00edch parametrech kvality grafiky (bez ray tracingu), stejn\u011b jako stresov\u00fd test FurMark s nejagresivn\u011bj\u0161\u00edmi nastaven\u00edmi (rozli\u0161en\u00ed 3840 \u00d7 2160, MSAA 8x). M\u011b\u0159en\u00ed v\u0161ech parametr\u016f se prov\u00e1d\u00ed po zah\u0159\u00e1t\u00ed grafick\u00e9 karty, kdy teplota GPU a taktovac\u00ed frekvence stabilizuj\u00ed.<\/p>\n

\u2b07<\/a><\/noindex>#<\/a><\/noindex>\u00da\u010dastn\u00edci testov\u00e1n\u00ed<\/strong><\/h2>\n

V testov\u00e1n\u00ed v\u00fdkonu se zapojily n\u00e1sleduj\u00edc\u00ed grafick\u00e9 karty:<\/p>\n