AMD FX-4xxx,6xxx,8xxx Bulldozer a Vishera - infa

[flanker]

[trojan4x4]

flanker zase nezaháľal
http://pctuning.tyden.cz/component/cont ... 8350-a-ln2

flanker nikdy nezaháľa

[flanker]

Prvně jsem si pustil nový 3D Mark na svém hlavním PC

[flanker]

Vrabci na střeše říkali, že se tomuto dá věřit:
http://www.xbitlabs.com/news/cpu/displa ... ealed.html
http://www.xbitlabs.com/news/cpu/displa ... ssors.html


FX-9370 125W 4.1/4.3 GHz
FX-9370E 125/95W 4.1/4.3 GHz (konfugurovatelné TDP v BIOSu)

he new FX-series microprocessors from AMD, which are due to be formally introduced on September 1, 2014, are the FX-8370 and the FX-8370E. Both chips feature eight cores, 4.10GHz/4.30GHz clock-rates, 8MB L2 cache, 8MB L3 cache and dual-channel DDR3 memory controllers. The FX-8370 will have thermal design power of 125W, whereas the FX-8370E will feature TDP of 95W. Both microprocessors will be priced at $189 in mass quantities.

In addition, AMD will also slash prices on existing FX-8300-series chips and introduce the model FX-8320E with 95W TDP.

Tu FX-8370E chci.

[Vlko]

Fííha, to znižené TPD je super správa, ale ako to dosiahli? Zlepšeny proces výroby, či novy steping ? Inak paráda

[pala]

95 TDP dosiahli tak, ze ide o lepsi kus kremika, ktory ma pri nizsom alebo rovnakom napati vyssie takty? 4.3 GHz pri 95W je celkom slusne. V tom pripade by to malo znamenat, ze 8370E pri rovnakej spotrebe malo mat vyssie takty ako 8370 (bez E), co mi pride ako selekcia lepsich kuskov a pritom cena je rovnaka. Alebo sa mylim?

[flanker]

já si ymslím, že to bdue ajko u Kaveri, kdy v jendovláknu výkon nijak neklesne. S 95W bude pak multithread třeba na 80% výkonu FX-8370 klasické. Tedy pro většinu uživatelů OK, protože málokdo jede neustály těžký mutltihread.

[flanker]

Chew je zpět... Taktujíc FX-8370

pro neznalé, Chew je legenda AMD overclockingu, který odešel ze scény po uvedení Bulldozeru v roce 2011. Nyní se s představením FX-8370 (dočasně?) vrací....Z informací, které mám, tak FX-8370/8370E bude stát za pořízení, pokud se díváte po AM3+. Bude to nejvýhodnější CPU pro domácí užití z hlediska výkonu/efektivity/OC.

[flanker]

A je tu první WR, jak je u FX tradiční, zase frekvence . Nový FX-8370 a 8722 MHz

http://rog.asus.com/353352014/crosshair ... formula-z/

[flanker]

FX-8350 GTX970/980 SLI vs i7-4930K se stejnými GF kartami.

http://www.tweaktown.com/tweakipedia/56 ... index.html

[Vlko]

Neviem či ste par dní dozadu zachytili info o novom FX-8310, tak to sem postnem.

AMD v tichosti uvedla cenově nejdostupnější osmijádro FX-8310

8 jader architektury Piledriver, 8MB L2 a 8MB L3 cache, TDP pod sto wattů a cena $125, to je téměř pohádková kombinace, která by i přes 32nm výrobní proces jistě dokázala přesvědčit nejednoho zákazníka. Parametrově srovnatelný FX-8370E totiž stojí o více než polovinu více a lépe na tom není ani jeho přímý konkurent Intel i5-4570. Je tu snad nějaký zádrhel?

Základní takt procesoru je nastavený na 3,4 GHz, což o 200 MHz překonává nový FX-8320E a o 100 MHz další novinku, 95W FX-8370E. Problém může být v turbu. Zatímco CPU World na základě dříve dostupných specifikací hovoří o 4,3 GHz (300 MHz nad FX-8320E, stejně s FX-8370E), oficiální parametrová tabulka na webu AMD ponechává kolonku „turbo“ prázdnou.

http://diit.cz/clanek/amd-fx-8310-osmijadro
http://www.cnews.cz/amd-potichu-uvedlo- ... s-otazniky

Cena je viac ako ferová.
Zaujimavé, že vraj nemá mať Turbo, hmm je to tak? Inak toto FXko nepatrí náhodou tiež do King Vishera rodiny?

[flanker]

dle mě patří do nových CPU, ale to zjistí někdo, až to koupí....Turbo tutově má, ale jaké...? Každopádně by to mohla být velmi výhodná koupě .

[flanker]

hodně zajímavé info, včetně úprav a nových měření upravených kompilátorů (něco si tudíž můžeme ověřit i my tady)

http://www.overclock.net/t/1554359/redd ... otage-wiki

[flanker]

Stilt mluví o problémech se změnami napětí a frekvence na deskách Gigabyte (IR kontroléry) a proč se to děje:

It has nothing to do with BCLK itself. Gigabyte bioses are completely broken and the issues are preventing the system from posting beyond certain clocks when multiplier is used to increase the clocks. They are most likely doing things in the wrong order. For example if they program the VRM controller in very late phases of the POST, it can definitely cause such issues. Most likely they are trying to run the CPU at the default voltage (ignoring all user made voltage adjustments), regardless of the clocks. After the CPU no longer can complete POST at the default voltage at the new, higher clocks you´ll see behavior like that.

In every case the CPU clocks will be programmed first. This is the reason why the system powers off when you change the multiplier and save the bios settings. In the early stages of POST, the CPU will run at clocks and voltages specified in P4 PState (on all FX-8K CPUs). After all of the CPU and chipset related parameters have been configured by AGESA then the bios should execute motherboard specific code, which may program the clock generator (set BCLK) and the VRM controller. After all of the lower level stuff has been completed, the CPU frequency will (should) be set to the normal levels (as defined by the higher PStates).

Some of the newer Gigabyte boards, such as 970A-D3P, 970A-UD3P and 990FXA-UD3 R5 are equipped with IR digital VRM controller. In order to adjust the voltage offsets, loadline or any other parameters of the controller, the controller must be programmed through SMBUS by using custom commands. So until the controller has been programmed by the proprietary code from Gigabyte, the CPU will run at the default voltages.

I think the issue is that the bios is currently executing the VRM controller programming code too late in the POST, after the point where the CPU has already been ramped to the maximum clocks.

They probably execute their code in this way:

1) AGESA (CPU, NCLK, DRAM, Chipset)
2) Exit Startup Core P-State (P4) -> P0
3) Program VRM controller through SMBUS
4) Program clock generator through SMBUS
5) Program DACs (NCT3933) for minor voltages (NB, DRAM, etc) through SMBUS

When the user has changed the P0 PState from 20x (4000MHz) to 22.5x (4500MHz) for example, the POST will most likely fail in "phase 2" when the CPU frequency ramps up but the CPU is still operating at it´s default voltage.

The issue would be most likely fixed by simply changing the execution order:

1) AGESA (CPU, NCLK, DRAM, Chipset)
2) Program VRM controller through SMBUS
3) Program clock generator through SMBUS
4) Program DACs (NCT3933) for minor voltages (NB, DRAM, etc) through SMBUS
5) Exit Startup Core P-State (P4) -> P0
Edited by The Stilt - 11/15/15 at 12:25am

[flanker]

Stilt z AMD mluví o problémech se změnami napětí a frekvence na deskách Gigabyte (IR kontroléry) a proč se to děje:

It has nothing to do with BCLK itself. Gigabyte bioses are completely broken and the issues are preventing the system from posting beyond certain clocks when multiplier is used to increase the clocks. They are most likely doing things in the wrong order. For example if they program the VRM controller in very late phases of the POST, it can definitely cause such issues. Most likely they are trying to run the CPU at the default voltage (ignoring all user made voltage adjustments), regardless of the clocks. After the CPU no longer can complete POST at the default voltage at the new, higher clocks you´ll see behavior like that.

In every case the CPU clocks will be programmed first. This is the reason why the system powers off when you change the multiplier and save the bios settings. In the early stages of POST, the CPU will run at clocks and voltages specified in P4 PState (on all FX-8K CPUs). After all of the CPU and chipset related parameters have been configured by AGESA then the bios should execute motherboard specific code, which may program the clock generator (set BCLK) and the VRM controller. After all of the lower level stuff has been completed, the CPU frequency will (should) be set to the normal levels (as defined by the higher PStates).

Some of the newer Gigabyte boards, such as 970A-D3P, 970A-UD3P and 990FXA-UD3 R5 are equipped with IR digital VRM controller. In order to adjust the voltage offsets, loadline or any other parameters of the controller, the controller must be programmed through SMBUS by using custom commands. So until the controller has been programmed by the proprietary code from Gigabyte, the CPU will run at the default voltages.

I think the issue is that the bios is currently executing the VRM controller programming code too late in the POST, after the point where the CPU has already been ramped to the maximum clocks.

They probably execute their code in this way:

1) AGESA (CPU, NCLK, DRAM, Chipset)
2) Exit Startup Core P-State (P4) -> P0
3) Program VRM controller through SMBUS
4) Program clock generator through SMBUS
5) Program DACs (NCT3933) for minor voltages (NB, DRAM, etc) through SMBUS

When the user has changed the P0 PState from 20x (4000MHz) to 22.5x (4500MHz) for example, the POST will most likely fail in "phase 2" when the CPU frequency ramps up but the CPU is still operating at it´s default voltage.

The issue would be most likely fixed by simply changing the execution order:

1) AGESA (CPU, NCLK, DRAM, Chipset)
2) Program VRM controller through SMBUS
3) Program clock generator through SMBUS
4) Program DACs (NCT3933) for minor voltages (NB, DRAM, etc) through SMBUS
5) Exit Startup Core P-State (P4) -> P0
Edited by The Stilt - 11/15/15 at 12:25am