Godavari throttles to 1.6 GHz after 20 seconds

[Valantar]

If I'm reading this correctly, the first step in any Kaveri build, regardless of CPU model, should be to lower CPU voltage significantly? Even the non-K models? I'm planning an A8-7600 build - does that apply to this as well?

 

[PG]

If I'm reading this correctly, the first step in any Kaveri build, regardless of CPU model, should be to lower CPU voltage significantly? Even the non-K models? I'm planning an A8-7600 build - does that apply to this as well?

I would check and then adjust accordingly. You are probably thinking, "What does that mean?

Here is what I would do:

Before you even build the new system, download the latest bios for the board you will use. Have that bios ready to go on a usb flash drive.

First boot of new system, go into the bios and check the revision. Most likely it will not be the latest revision, so update the bios. Then set to defaults, save, and reboot again just to be safe.

Now go back into the bios. All manufacturers are different, but they all have a page where you can see the voltages, temps, etc. If the core voltage is say 1.45 or higher, then I'd say it's overvolting and you should probably make an adjustment.

If you are going with that 7600, then lower power means something, so you need to lower the voltage with an offset type setting. Don't set a lower but fixed voltage. You want the power saving features on and working. My cheap Asus board can lower the voltage by an offset, and it works pretty well. You will have to experiement a little to see how far down you can go.

Also use cpu-z to watch voltages in windows. I have read that Kaveri doesn't blue screen so much, but locks up when voltages are too low or the northbridge gets too high.

That also reminds me, bump up the northbridge on a Kaveri if you can. Of course locked chips don't have that feature. 2000 works pretty well with many Kaveri apus.

 

[DrMrLordX]

There's really nothing wrong with setting a low base voltage instead of messing with offsets. With a 7600, you have to be mindful of the target TDP (45w or 65w) since that will affect clocks.

But really, you should just set the base voltage to the lowest thing that will be stable in a stress test like Prime95, IBT/Linpack, etc. If you're worried about turbo, dial in the max turbo speed and then set voltage to accommodate it. 3.8 ghz should be doable with 1.2v or less, and that's lower voltage than many board UEFIs are supplying to Kaveri chips now regadless of clockspeed.

If you really want turbo, you can lock in a base vcore for state p3 and (depending on the board) turbo will still function. You can then use a tool like amdmsrtweaker to adjust the voltages for states p0-p2 to bring those voltages down (if necessary) to accommodate turbo states.

Right now, I have my 7700k locked to 3.4 ghz with vcore of 1.075v (1.056v actual idle, 1.064v at load with Medium LLC).

 

[PG]

I can't argue with you, but I did forget to mention a big thing that I always do when overclocking Kaveri: turn off turbo

 

[DrMrLordX]

Sensible enough. There's precious little reason to mess with turbo anyway . . . may as well just lock it in to max turbo clocks, even with a locked chip like a 7600. Nearly every 7600 out there should do 3.8 at a reasonably low voltage.

 

[sm625]

If you think OEMs running CPUs at 1.5V is bad, just imagine what these clowns are doing to their notebooks. Those are probably overvolted as heck too, giving AMD an even worse name than they deserve.

 

[Shehriazad]

Turbo is so weird, anyway....on my board it was just perma turboing...so I went ahead and just OC'd it as far as I could first (4.7) but since I didn't feel like going hotheaded in that little cube...I went for as far as 1.2V could take me.

Which was 4.1.

Either way....this can't become a lasting thing. AM4 socket better has this sorted out.

 

[Valantar]

Sensible enough. There's precious little reason to mess with turbo anyway . . . may as well just lock it in to max turbo clocks, even with a locked chip like a 7600. Nearly every 7600 out there should do 3.8 at a reasonably low voltage.

Won't that completely kill off any kind of power saving features? Or will it still scale down in the same way under low loads?

 

[Dresdenboy]

2 pages of discussion wasted:

"Update (May 31): Throttling problem was fixed by upgrading to P2.60 BIOS. Thanks to Roger Harshman for the hint! The processor is stable now, and runs CPU benchmarks and games without throttling and lockups. By the way, new BIOS sets lower core voltage, close to 1.45V."

http://www.cpu-world.com/news_2015/2015052601_AMD_A10-7870K_APU_a_few_extra_details.html

 

[Vesku]

Don't see how it's wasted as pretty early on people guessed it was not having a fully Godavari aware bios. It still runs at a high default voltage which is where most of the discussion has been, guess AMD is saving quite a bit of money by having such generous binning requirements.

 

[LTC8K6]

2 pages of discussion wasted:

"Update (May 31): Throttling problem was fixed by upgrading to P2.60 BIOS. Thanks to Roger Harshman for the hint! The processor is stable now, and runs CPU benchmarks and games without throttling and lockups. By the way, new BIOS sets lower core voltage, close to 1.45V."

http://www.cpu-world.com/news_2015/2015052601_AMD_A10-7870K_APU_a_few_extra_details.html

The first few posts indicated that it was likely a BIOS problem.

Your posted is the one better labeled "wasted".

And 1.45V is still crazy high for stock...

 

[Dresdenboy]

The first few posts indicated that it was likely a BIOS problem.

Your posted is the one better labeled "wasted".

And 1.45V is still crazy high for stock...

Guessing vs. facts. This is the reason, why stock markets fluctuate that much.

Well, does the voltage still apply with newer BIOS?

 

[AtenRa]

http://anandtech.com/show/9307/the-kaveri-refresh-godavari-review-testing-amds-a10-7870k

 

[sm625]

The review just posted on AT is a disaster. It doesnt mention any of this. But you can clearly see it in the benchmarks. It regularly scores worse than a 7850k. There is only one way that can happen...

they need to set the volts to 1.3 and rerun the benchmarks. Then again at 1.25 and maybe even 1.2.

 

[LTC8K6]

The review just posted on AT is a disaster. It doesnt mention any of this. But you can clearly see it in the benchmarks. It regularly scores worse than a 7850k. There is only one way that can happen...

they need to set the volts to 1.3 and rerun the benchmarks. Then again at 1.25 and maybe even 1.2.

If Anandtech did not notice throttling, that would look bad...

 

[LTC8K6]

The benches seem to be all over the place among the AMD chips.

The A6-7400K dominates a few times.

Odd.

 

[sm625]

I understand why the A6-7400K dominates that one test, because it is a per thread type benchmark. So it gets an unfair advantage by only having 2 threads. What I dont get is why the A8-6600k would dominate any benchmark vs a kaveri A10.

 

[Dresdenboy]

You would think it was an issue that would be addressed after the blunder with the first generation of Kaveri chips and their relatively mild throttle compared way below baseclock to the 3Ghz mark. But 1.6Ghz....

Let's reconsider the 1.6GHz after we got new measurements.

The mild throttle might just be the design specific working range for heavy fp load, like this one:

http://www.anandtech.com/show/8423/intel-xeon-e5-version-3-up-to-18-haswell-ep-cores-/5

 

[LTC8K6]

I understand why the A6-7400K dominates that one test, because it is a per thread type benchmark. So it gets an unfair advantage by only having 2 threads. What I dont get is why the A8-6600k would dominate any benchmark vs a kaveri A10.

It's actually in more than one bench.

 

[Tsavo]

1.45 VOLTS? What are their chips made out of, cast iron?

 

[LTC8K6]

1.45 VOLTS? What are their chips made out of, cast iron?

Yeah, it was "Oh my god!" "1.48 volts!"

Now it's okay at 1.45 Volts?

I dunno...

I don't see any new benches showing the higher performance with the throttling gone, but I guess I will take their word for it.

 

[Dresdenboy]

1.45 VOLTS? What are their chips made out of, cast iron?

If you heat them up, a glowing inscription will appear: "5V TTL".

 

[Erenhardt]

No overclocking? my 7850K did 4.4@1.3V
This chip at 1.45V should do 5.0+

Also, my kaveri break 220W from the wall in CPU+IGP torture test at those settings
At those voltages it would probably start a chain nuclear reaction

 

[DrMrLordX]

Won't that completely kill off any kind of power saving features? Or will it still scale down in the same way under low loads?

If you lock in a clockspeed other than the normal base clock, it won't necessarily eliminate power saving clocks. That is sort of OS dependant. Linux can (and will) downclock the CPU according to the power governor in use, but Windows has problems with it.

Interestingly enough, Kaveri won't downclock during iGPU load under Linux, either.

Yeah, it was "Oh my god!" "1.48 volts!"

Now it's okay at 1.45 Volts?

I dunno...

I don't see any new benches showing the higher performance with the throttling gone, but I guess I will take their word for it.

1.45v is still stupid. My older-stepping KA-V1 chip will do 4.5 ghz @ 1.425v, and that's conservative (it might need a little less). GV-A1 should need less, not more.

No overclocking? my 7850K did 4.4@1.3V
This chip at 1.45V should do 5.0+

I can tell you from personal experience that there is a pretty hard wall around 4.7 ghz. Anything 4.8 or higher usually needs cold to get there. -60C will get you to maybe 5.2-5.6 ghz depending on the chip. Anything colder than that, and you get cb/cbb nonsense.

Also, my kaveri break 220W from the wall in CPU+IGP torture test at those settings
At those voltages it would probably start a chain nuclear reaction

FM2+ just isn't designed from the ground-up for power delivery like that. It's a wonder that anyone designed boards that can handle that kind of load at the socket.

 

[LTC8K6]

Some bits regarding binning, voltage, and power from the Anandtech review:

That last bit is particularly important, based on information I received from one of the motherboard manufacturers that specifically released new motherboards for Kaveri Refresh. The motherboard manufacturer stated that they released new motherboards because the power draw for the graphics portion of the APU is now more substantial than before, and they wanted to implement an element of security into their products. Needless to say, it allows OEMs and system builders to provide a ‘more up-to-date’ motherboard when providing systems around the A10-7870K, but AMD were quick to respond by saying words to the effect of ‘any motherboard on the A88X/A78/A68H/A58 chipsets that use an FM2/FM2+ to our specifications will work with the Kaveri Refresh line’. So there you go.

When speaking to AMD, it transpires that this boost on the Kaveri design is due to three things: an improved process, binning, and a marginal voltage increase (which is doubly effective due to the binning). The stock voltage of the processor goes up 0.050 volts compared to the 7850K, although AMD still reports a 95W TDP. Usually stock voltages are listed as a range depending on the quality of the die, but on average that window is slightly higher this time around.

At 117W in our test, it seems a little over the 95W TDP listed, especially when the 7850K achieved 85W and under its 95W TDP. There will be variation within a SKU based on the binning quality, as I doubt the average 0.050 volts added to this SKU as I was told by AMD accounts for this much. Perhaps we ended up with a good 7850K sample but a mediocre 7870K sample.

http://www.anandtech.com/show/9307/the-kaveri-refresh-godavari-review-testing-amds-a10-7870k