AMD novinky CPU/grafické karty -infa, recenze

[FlanK3r]

Zase něco o nových polárkách AMD GPU:
http://www.xtremesystems.org/forums/sho ... -immersion

[FlanK3r]

Příběh o novém AMD BOX chladiči :)
http://www.tomshardware.com/news/amd-wr ... 31015.html

[Rewqa]

Jak to bude se socketem AM4 a novými procesory? Něco jsem četl, ale moc toho na netu není a liší se datem vydání atd. Díky.

[Martab]

Podle všeho by AMD mělo sjednotit vše do AM4, první do něj vyndou nové APU a později pak i FX bez iGPU na bázi ZENu.

[Rewqa]

A datum vydání? Vyplatí se na to případně čekat?

[Martab]

Pokud vím, tak uvedení je plánováno na konec Q2, tedy spíš až v létě. A jestli čekat, toť otázka, záleží jak moc teď výkon omezuje...

[Rewqa]

No tak po koupi GTX 660 skoro vůbec, ale tak proč si neudělat radost Spíš jestli se vyplatí počkat a přejit pak rovnou na DDR4 paměti a nebo se na to vykašlat a koupit nějaký FX do stávajícího socketu.

[Martab]

Pokud tě teť výkon nějak extra netlačí, počkal bych, už jen kvůli docela výraznému očekávanému poklesu spotřeby.

[Rewqa]

Nene i nejnovější hry pořád dávám na pohodu. Dík.

[FlanK3r]

Stilt vydává fix pro procesory Kaveri, kdy můžou nastat situace, že se CPU při silné využití iGPU podtaktovává. Více o tom:
http://www.overclock.net/t/1588994/disa ... oller-apus


Most of Steamroller (Kaveri / Godavari) based APU owners have noticed that the CPU frequency drops immediately (below the advertized base frequency) once the iGPU is being utilized. This behavior has nothing to do with reaching or exceeding the TDP limits, but it is infact a normal feature of the power management. This feature is called "GeAPM" or "GPU Efficient Application Power Management".

While the performance impact from having GeAPM enabled is in most cases anything between negligible and non-existent, there are certain scenarios where having "GeAPM" enabled actually does deteriorate the performance. Workloads (single or multiple) which is able to fully utilize both of the CPU cores and the iGPU simultaneously usually show at least some improvement when GeAPM is turned off, since the CPU is allowed constantly to operate at it´s base frequency (unless limited by the TDP). For example A10-7850K or 7870K will operate at 3.7GHz and 3.9GHz instead of 3.0GHz when GeAPM is turned off. The largest difference I´ve personally been able to achieve was when I ran GPUPI (GPU) and Cinebench R15 simultaneously. With GeAPM enabled the Cinebench score jumped between 282 - 297 points and with GeAPM disabled the result increased to 317-326 points.

In short: It is not a big deal, but I decided to make it available in case it bothers someone.

Here are the first two bioses, for ASUS Crossblade Ranger and A88X PRO. The rest of ASUS models will follow at some point, but for now releasing the bioses for other ASUS models is impossible. ASUS bioses are delivered in a signed capsule, so when they are modified the signature will no longer match. Without the use of SPI programmer (or USB Flashback) there is no way to update the modified ROMs. These first two boards are equipped with USB Flashback, so they can be updated. Just use the normal USB Flashback procedure with the supplied file.

ASRock, Gigabyte and MSI bioses will follow once I´ve checked if there is any security measures used which I need to take into account. I´ve got ASRock and Gigabyte boards I can test with, but zero MSI boards. Therefore I will prioritize ASRock and Gigabyte initially.

To check if GeAPM is disabled, open GPU-Z and check the VBIOS version.

[FlanK3r]

AMD snad plánuje i 32 jádrový procesor pro servery s kódovým názvem Zeppelin (AMD 17h)

http://www.planet3dnow.de/cms/22279-zep ... 32-kernen/
http://dresdenboy.blogspot.cz/2016/02/a ... ed-by.html

A informace ze staršího kódu o Zen:
An older patch on the same mailing list also gives a little more info about Zen:
On AMD Fam17h systems, the last level cache is not resident in Northbridge. Therefore, we cannot assign cpu_llc_id to same value as Node ID (as we have been doing currently)
We should rather look at the ApicID bits of the core to provide us the last level cache ID info. Doing that here.
The most interesting part describes the way, how the last level cache (LLC) ID is being calculated for Zen based MPUs:
+ core_complex_id = (apicid & ((1 << c->x86_coreid_bits) - 1)) >> 3;
+ per_cpu(cpu_llc_id, cpu) = (socket_id << 3) | core_complex_id;
"Core complex" should be similar to "compute unit" and has been used in some AMD patents already. The expression marked in red means a shift right by 3, which equals a division by 8. So with two logical cores per physical core due to SMT, a core complex should contain four Zen cores and a shared LLC.
The next line shows the socket ID being shifted left by 3, leaving 3 bits for the core complex ID, which suggests a maximum number of eight core complexes per socket, or 32 physical cores. This number should first be seen as a placeholder, but we've already seen rumours mentioning that many cores.

[FlanK3r]

Jak výkonný bude Carizzo v obstarožním testu Superpi je vidět zde (možná AM4 verze na tom bdue ještě o chlub lépe díky větší L2 cache). Jasně nejvýkonnější AMD na tento benchmark.




Na frekvenci 3400 MHz s turbem (která tam není stále, ale skáče to) je to pecka. Cca jako 5100 MHz velmi vyladěná Vishera a nebo 4500 MHz vyladěný Kaveri!
1M je klasicky poměrově slabší, což je zajímavé (cca 17s), jak Kaveri, tak Carizzo mají nijak výrazný 1M, ale velmi dobrý 32M (v porovnání s Vishera)