AMD FX "Vishera" reviews info (again, after mod mistake)

[splmann]

First Mem Test with my Corsairs Platinum 2800 :

[tbone8ty]

I need to pick from the 1229PGN batch

someone should start a batch thread for vishera cpus.... wasn't there one for bulldozer/zambezi?

[FlanK3r]

was...Beep created last year this thread

[tbone8ty]

so what are the best batches so far?

what week are we up to now?

[FlanK3r]

from my point of view 1229, 1237, 1242/43. But of course batch =//= 100% sure about quality of chip. Only in 1229 (example) you can find more good pieces than in 1232 batch.

[FlanK3r]

http://ewh.ieee.org/r5/denver/sscs/P...2_05_Sathe.pdf

Resonant Clock Design at Piledriver core info

[ryboto]

Resonant mesh must be for updated PD, right? Maybe we'll see it in 2013?

[The Stilt]

1.325V required for 4.5G Prime95 LargeFFT stability on AMD FX watercooling (40% fan DC).

[FlanK3r]

you kidding me Stilt? Where did you found this chip? Only 1.85V!!!!!!!!!!!!OMG. Can you try PiFast? In PIFast is Vishera "strong". And wprime maybe too. This is your "Cinebench R11.5" chip?

[tbone8ty]

What's batch number?

[Mechanical Man]

1.325V required for 4.5G Prime95 LargeFFT stability on AMD FX watercooling (40% fan DC).

The real wtf is 4GHz NB.

Edit. OH. it was LN2. 8GHz screenie.

[The Stilt]

The batch is 1244PGN.

The chip seems to be able to handle some serious under voltage so I did some extended testing.
At the default base frequency (4.0GHz) the chip is Prime95 stable with just 1.1875V

Crosshair V Formula-Z (LLC matched to leakage = Ultra High, OCP 100%, Phase Control = Standard, VRM Fsw 300kHz, VRM DC = T.Probe, VRM Response = Regular)
VRM input current = clamp on a EPS12V
VRM input voltage = Fluke DMM
VRM output current = CHIL DCR
VRM output voltage = Fluke DMM & CHIL Vsen
Workload: Pov-Ray 3.7 RC6 (all core benchmark).





The VRM efficiency peaks at 77.2% on Crosshair V Formula-Z, which is decent but not the best.
The VRM on CVF-Z is built to withstand some severe molesting. It can take anything you can throw at it, however this will hurt the efficiency a bit.

Lowering the core voltage from 1.3250V to 1.1875V decreased the power consumption by 42.8% (VRM Input) or 39.9% (VRM Output).
Now it fits into the 95W TDP envelope too

Since Prime95 is no considered as "normal use" (for CPU power measurement) I made separate tests with it.
Settings: Custom 512K - 512K (LargeFFT, In Place)

Core VDD: 1.3250V
VRM Input Current: 14.98A
VRM Input Voltage: 12.13V
VRM Input Power: 181.70W
VRM Output Current: 103A
VRM Output Current: 1.316V
VRM Output Power: 135.54W


Core VDD: 1.1875V
VRM Input Current: 10.52A
VRM Input Voltage: 12.13V
VRM Input Power: 127.60W
VRM Output Current: 77A
VRM Output Current: 1.184V
VRM Output Power: 91.16W

[tbone8ty]

This is awesome. Imagine if Amd sold fx-8350 with 95w tdp they be in a little better shape.

Hey stilt what's max stable 24/7 for yur 8350 1244pgn chip on Amd fx watercooler? What r temps?

I want a 8350 24/7 stable @ at least 5ghz....just curious if Its worth it to get one if these all in one systems like a corsair h100 or a antec kuhler 920 would do the job... Assuming I get a good batch number. Maybe ill get lucky like u guys

Gonna pluck this into my Crosshair 5

[Alex-Ro]

Ain't gonna happen,at least 5 ghz daily requires a beastly cooling and i have serious doubts that h100 can handle that.

[FlanK3r]

I have one 5000 MHz stable, but this chip, what i saw in others forum, is better than average 8350s...So if will be lucky, after you can hit 5 GHz stable (or very close to 5 GHz) with Antec or H80 etc.
I think, the chip of Stilt is more than good . Maybe with 1.55V he culd get 5.1 GHz stable.

[The Stilt]

This is awesome. Imagine if Amd sold fx-8350 with 95w tdp they be in a little better shape.

Hey stilt what's max stable 24/7 for yur 8350 1244pgn chip on Amd fx watercooler? What r temps?

I want a 8350 24/7 stable @ at least 5ghz....just curious if Its worth it to get one if these all in one systems like a corsair h100 or a antec kuhler 920 would do the job... Assuming I get a good batch number. Maybe ill get lucky like u guys

Gonna pluck this into my Crosshair 5

My 24/7 system is using a A10-5800K so I don?t really have 24/7 clocks for the FXs

Nowdays I'll always try to find the optimal balance between the efficiency and performance, not the absolute performance at all cost.

An example: FX-8350 requires 1.3750V for Prime95 stability at 4500MHz and 1.4400V at 4700MHz.

4500MHz
1.3750V
Estimated Pmax: 145.9W

vs.

4700MHz
1.4400V
Estimated Pmax: 167.1W

Difference in performance: 4.4%
Difference in power consumption: 14.5%

To maximize the balance of performance and efficiency I want the "golden ratio" to be =<2.0.
In the example above the power limit would be 158.7W, which means the chip must be able to do the 4.7GHz with 1.40V voltage.
In that case the increase in power consumption could be justified (difference in power 8.8%, difference in performance 4.4%).

Also keep in mind that a small 20W increase in CPU power consumption will increase the actual power consumption much more than that.
CPU VRM efficiency: 75% -> 26.7W
PSU efficiency: 90% -> 29.7W

I'll start to sound like a greenpeace rep now so better stop...


I have not tried the maximum Prime95 stable frequency yet on this chip, however I recon it is not that good.
This is an ultra high leakage part so the heat will become a issue very quickly.

Achieving 24/7 stability at 5.0GHz+ is not matter of how good your chip is, let alone the batch code of the CPU.
Each and every single one Vishera CPU ever manufactured will reach 5.0GHz 24/7... with sufficient cooling.

The term sufficient varies. It is anything between 10C liquid to LN2.

Only one thing is certain: 24/7 stable 5.0GHz or higher cannot currently be done on normal air or watercooling.
Because of heat <=> power consumption (motherboard).

A rough estimate what the 24/7 stable 5.0GHz would require:

• 190W+ cooling capacity at <40C CPU die temperature
• 60W+ cooling capacity for CPU VRM
• 130A+ constant CPU VRM current capability

The highest 24/7 Prime95 stable frequency I have reached on 8-core Vishera is 4.9GHz.
Cooled with AMD FX watercooler, fans running at 100% DC (ear protection required) and in 21C ambient.
The CPU *might* have been a "airbender" special (with special seasoning)

[Alex-Ro]

That's what i tought too,having a 5 ghz 24/7 would require a chip that has veeery low leakage and can keep the temperatures and power comsumption in somewhat normal range.

[FlanK3r]

Il can try it around New Year with AMD cooling setup...I believe, the 5700+ MHz chip can do this

[vitchilo]

So according to the reviews, the i7 3770 is still kicking butts compared to the fx-8350... still the 8350 is $100+ cheaper.

[tbone8ty]

Not in all cases maybe 1 or 2 multitasking cases where 8350 beats the 3770k

http://techreport.com/review/23750/a...or-reviewed/10

http://www.anandtech.com/show/6396/t...x4300-tested/4

http://www.anandtech.com/show/6396/t...x4300-tested/3

But basically yes the 3770k handles the 8350

Hopefully this doesn't derail this thread I apologize in advance.

[Olivon]

Thanks a lot The Stilt, your comments are always very interesting

[Final8ty]

That Vcore has got to be wrong, if not then lucky man.
http://valid.canardpc.com/2625134

http://www.overclock.net/t/1318995/o...ners-club/6060

[Maxforces]

measurements of programs are wrong. I have two matherboards Gigabyte 970 and MSI 990fxa GD80 and two different readings of vCore on the same CPU. I checked Gigabyte motherbord and 1.4V is real 1.3V.
Since the first FX measurements are great unknown and dependent of motherboard and bios.This is the same with temperature.

[tbone8ty]

check coretemp is should say what your specific chips VID is...

[The Stilt]

Santa played cheapskate this year and brought me some FX-4300s instead of those FX-8350s Ive asked



The actual VDD is 1.489V.

AMD FX watercooler with Enermax Magma 120mm (18dB) intake & Scythe S-Flex 120mm (20.1dB) exhaust fans for reduced noise levels.
25?C ambient.

Prime95 LargeFFT

DCR Imax: 91.0A & DCR Vsen 1.482V == 134.86W (Pmax)

X264 R2230 10-bit RAW BD 1080p -> mkv (preset = slow, tune = film, bitrate = 9000, output = .mkv)

DCR Imax: 72.25A & DCR Vsen 1.482V == 107.1W (Pmax)