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

[informal]

Very nice chip Stilt. It's surprising(to me) that it draws "only" 107W in real world threaded workload at that clock and voltage. At 4.9Ghz this thing is everything but slow,even with it's "lowly" 2 modules .

[The Stilt]

Very nice chip Stilt. It's surprising(to me) that it draws "only" 107W in real world threaded workload at that clock and voltage. At 4.9Ghz this thing is everything but slow,even with it's "lowly" 2 modules .

At stock frequency, undervolted to 1.1625V the Pmax during P95 LargeFFT is less than 55W
With 6W BaseTDP and all of the other voltages included it is still easily under 65W.
In real world applications around 55W or so.

[tbone8ty]

AMD FX-8300 95w

same C0 stepping

3.3 ghz base clock
4.2 ghz turbo

http://techreport.com/news/24111/rum...ater-this-week

http://www.cpu-world.com/CPUs/Bulldo...20FX-8300.html

i wonder if these r cherry picked low leakage parts?

probably not but we can hope

[The Stilt]

FX-8300 should be exactly the same 'bin' as FX-8350
Vishera can operate at very low voltage after even a minor drop in frequency.

I recon the FX-8300s have 1.225V - 1.275V VID fuse for the base frequency, depending on leakage.
FX-8350s usually have 1.325V - 1.375V.

[tbone8ty]

so most FX-8350 could be downvolted and fit under the 95w power envelop (at stock 8350 clock settings)??

so like ~1.2v could probably get u 95w at 4ghz?

[The Stilt]

so most FX-8350 could be downvolted and fit under the 95w power envelop (at stock 8350 clock settings)??

so like ~1.2v could probably get u 95w at 4ghz?

Most of them if not all.

To fit in 95W TDP slot a FX-8350 requires around 0.1V undervoltage (vs. PS0 Fuse VID).
The PS0 Fuse VID already contains a hefty margin as the loadline specification allows massive undervoltage.
The allowed amount of undervoltage depends on the leakage as the loadline slope is in relation to current draw.
With a high leaking part I have seen up to 60mV undervoltage with the default slope.

On average with FX-8xx0 -series CPU:

At 3500-4000MHz - 100MHz increase or decrease in frequency (with the same voltage) is worth of 2W in Pmax.
At 4100-4600MHz - 100MHz increase or decrease in frequency (with the same voltage) is worth of 2.5W in Pmax
At 4700-5000MHz - 100MHz increase or decrease in frequency (with the same voltage) is worth of 3W in Pmax

The leakage is thermally induced so it increases (or decreases) along with the temperature.

[massman]

Had about an hour or two fun with Piledriver for Country Cup. Just boot 'n' bench, nothing special

[tbone8ty]

Most of them if not all.

To fit in 95W TDP slot a FX-8350 requires around 0.1V undervoltage (vs. PS0 Fuse VID).
The PS0 Fuse VID already contains a hefty margin as the loadline specification allows massive undervoltage.
The allowed amount of undervoltage depends on the leakage as the loadline slope is in relation to current draw.
With a high leaking part I have seen up to 60mV undervoltage with the default slope.

On average with FX-8xx0 -series CPU:

At 3500-4000MHz - 100MHz increase or decrease in frequency (with the same voltage) is worth of 2W in Pmax.
At 4100-4600MHz - 100MHz increase or decrease in frequency (with the same voltage) is worth of 2.5W in Pmax
At 4700-5000MHz - 100MHz increase or decrease in frequency (with the same voltage) is worth of 3W in Pmax

The leakage is thermally induced so it increases (or decreases) along with the temperature.

question is what volts/clock speed would it would take to reach 77w?

[The Stilt]

question is what volts/clock speed would it would take to reach 77w?

These numbers are recorded with the same chip I posted the previous charts with.

FX-8350
Node TDP = 125W (in real world applications, TDP will be exceeded in Prime95 or similar software. On both AMD and Intel).
Base TDP = 6.75W (Core Northbridge + Logics).
Core TDP = 118.25W (Node TDP - Base TDP).

Target Node TDP = 77W
Target Base TDP = 6.75W
Target Core TDP = 70.25W

At stock frequency (4.0GHz) and with the rated (1.3250V) voltage, this chip exceed the rated Core TDP by 15.5% in Prime95 LargeFFT (136.66W DCR Pmax). Therefore the Target Core TDP with this margin included is 81.1W (70.25*1.155).

Result: 3600MHz / 1.1250V
DCR Pmax: 79.6W (72A @ 1.105V)
Total Node TDP: 86.3W

In a real world application (tested with Pov-Ray 3.7 RC6), at stock frequency (4.0GHz) and with the rated (1.3250V) voltage the actual DCR Pmax was 1.57% below the rated Core TDP value of 118.25W (= 116.42W). The Target Core TDP with this margin included is 69.2W (70.25/1.0157).

Result: 3500MHz / 1.0875V
DCR Pmax: 66.2W (61A @ 1.085V)
Total Node TDP: 72.95W

The underclocked frequencies are Prime95 LFFT stable for 20 minutes or more.

ps. At 3.3GHz / 1.05V the Node TDP was 63.8W

Anyone seen a FX doing 5.1GHz Prime95 LargeFFT with aircooling and with 4 or more cores enabled?
I have

[massman]

Also had one with high htt and high mem:

[The Stilt]

Quite an FX-4300 this would be...

The CU0/CU1 are superior to anything else Ive seen in the past.
CU2/CU3 are nothing special after 4.7GHz mark.





Thermalright UEX 120 + a single Everflow 1225SU (from AMD FX watercooler).

[tbone8ty]

So why isn't Amd making a 77w fx-8300 part?

im sure people would buy an 8 core 77w FX chip with slower clock speeds

i guess thats what opterons for

[FlanK3r]

Massman: test PiFast, in superpi is Vishera lazy ....

Stilt: do you think, next year we will see higher models than FX-8350? Or some new revision C1/C2?

[massman]

I had about 17.41s in PiFast, but the memory was not optimised much. Next week, I will re-try (also 32M). I improved my max 1M speed by about 60MHz using PSCheck. I'm quite confident I can do a 8GHz 1M run, but I need to do some tricks to make sure the efficiency is alright .

More next week. I'll do a write-up with tips if I get 8GHz stable



link: http://hwbot.org/submission/2340068_...50_11sec_281ms

[The Stilt]

Massman: test PiFast, in superpi is Vishera lazy ....

Stilt: do you think, next year we will see higher models than FX-8350? Or some new revision C1/C2?

Not sure about a "major" stepping change however I would imagine the node is changing quite frequently, as usual.

IMO currently the biggest issue with Vishera is inequality between the compute units (CU2/3), just like it was on Bulldozer.
On average it is limiting the maximum frequency on FX-8xx0 series CPUs by 200MHz.
Ultimately heat will limit the maximum frequency of course.

A high leakage CPU with with completely equal CUs (0-3) would make a stable 5GHz on a proper motherboard and the very best air cooling.
These chips are more rare than the ones which can do 8.5G on LN2.

[Zeus]

Also had one with high htt and high mem:

Good to see high ram clocks are possible on Vishera looks on par with Ivy. How come everybody is running such conservative ram speeds om AMD set ups here?

I would love to give my Sammy based Trident-X rams a run for their money on an AMD setup but i haven't got one (yet? )

[FlanK3r]

Vishera looks better for LN2 OC too. Looks easy to get 8 GHz than Zambezi Bulldozer. I hope, someone will broke record of FX-8150 with new Vishera chips in freqency...I wish really some very great low VID chip for it

[massman]

Good to see high ram clocks are possible on Vishera looks on par with Ivy. How come everybody is running such conservative ram speeds om AMD set ups here?

I would love to give my Sammy based Trident-X rams a run for their money on an AMD setup but i haven't got one (yet? )

No idea why everyone is being so conservative. I've not had the slightest issue pushing the memory over DDR3-2600 and performance is pretty okay too ... if you use high HTT frequency (same issue on Bulldozer). I'm going to do a couple of 32M runs next week - seeing how high I can go with mem/nb - and probably a pifast too. Piledriver is a lot of fun .

By the way, since the key to both Bulldozer and Piledriver performance seems to be memory divider (see: http://hwbot.org/forum/showpost.php?...7&postcount=37), I think I need a different board. UD7 is limited to ~340MHz HTT (tested different samples of motherboards and CPUs), but it seems the UD3 might be pulling a little better (300+ on air compared to 275 max of UD7). Hope to get my hands on a UD3 next week .

Vishera looks better for LN2 OC too. Looks easy to get 8 GHz than Zambezi Bulldozer. I hope, someone will broke record of FX-8150 with new Vishera chips in freqency...I wish really some very great low VID chip for it

Not sure if it's a lot easier - tested a total of 3 CPUs so far and the two others did around 7.8GHz boot into OS whereas this one goes to boot around 7.91MHz easily. A lot better than some Bulldozer CPUs, though

[FlanK3r]

I had no luck with Bulldozers...the best one was 8320 and hit about 7700 MHz, but OBR kill my chip with high voltage after ...ANd others 8350s/8320s had only limit around 7200-7400 MHz...
Vishera? My first Vishera easy 8 GHz, others looking similar (others I compared only with AMD liquid cooling)

[docwhom]

The underclocked frequencies are Prime95 LFFT stable for 20 minutes or more.

ps. At 3.3GHz / 1.05V the Node TDP was 63.8W

This is what caught my interest, as low as 63.8W........?

So could you use it as the ultimate 24/7 Utorrent setup? Because according to your findings it literally means this cpu is now in the low-voltage category like an i7 2600S or an i7 3770S.

[Panther_Seraphi]

This is what gets me about AMD.

Everyone complains about how power hungry they are. Well I have my 1090t sitting at 800mhz with CnQ on with 0.8 volts going through it. I could probably go lower but just aint tried.

[mockingbird]

Why does it lag so much in SuperPi compared to the i7? 14 minutes more to complete the 32m calculation?

[Zeus]

^^It's all about older instruction sets legacy. Bulldozer was awfull at SuperPi but Piledriver is even worse! Apart from that performance isn't that bad at all.

[docwhom]

This is what gets me about AMD.

Everyone complains about how power hungry they are. Well I have my 1090t sitting at 800mhz with CnQ on with 0.8 volts going through it. I could probably go lower but just aint tried.

Exactly, everyone has been duped by Intel with their so-called "low-power usage" with all of their product range. And they really do sell heaps of their product to everyone, everywhere. Yet AMD has been painted as the old-style villain unable to compete with Intel at all. That still burns me up........

Yet all of AMD cpu's are able to coast at 800mhz when idle. I still have a 910E that still sits on 800mhz when ever it's bored. Hence for me I really still like AMD, their video cards and their cpu's.

Just evaluating which cpu to drop into my now ancient 990FXA-UD7 board. Want to use it as a 24/7 server base.

[FlanK3r]

http://www.xtremesystems.org/forums/...-of-Andre-Yang!