Kaveri Refresh "Godavari" performance

[Flash831]

I've been following this summer's release of several CPU's and APU's from both Intel and AMD with great interest. However, It seems to me that the former has received most of the media's attention.

The Godavari APU's (A10-7870K and A8-7670K) seem to be easy to find at most retailers. But reviews are not.

Anandtech did a review of A10-7870K.
Since then some other sites have also made reviews of it.

Reviews of A8-7670K seems to be completely MIA though. :hmm:

After reading the reviews I get the impression the Godavari is a so-so (at most) improvement to Kavari.
However, after reading some posts (by The Stilt) in the Kaveri Refresh-thread at OCN, it seems to me Godavari offers more than meets the eye, especially when overclocking/undervolting. The default voltage simply seems excessively high.

Do you have any experience of these new parts and how they stand against Kaveri?
How would an undervolted A10-7870K stand against A10-7800?
And an undervolted A8-7670K against A8-7600?

 

[ShintaiDK]

It performs exactly as Kaveri at the same speed. Because its just another name for Kaveri.

 

[Flash831]

It performs exactly as Kaveri at the same speed. Because its just another name for Kaveri.

Yes, but it requires less voltage for doing so. In a TDP-constrained case, like Streacom FC5 it should be more suitable than Kaveri (as long as your willing to change the default clocks). ^_^

 

[Abwx]

However, after reading some posts (by The Stilt) in the Kaveri Refresh-thread at OCN, it seems to me Godavari offers more than meets the eye, especially when overclocking/undervolting. The default voltage simply seems excessively high.

Do you have any experience of these new parts and how they stand against Kaveri?
How would an undervolted A10-7870K stand against A10-7800?
And an undervolted A8-7670K against A8-7600?

The numbers he posted suggest that Godavari get to 4.1GHz at 1.196V, the early 7850K was stuck at 3.7GHz at this voltage and did require 1.330V to get to 4.1Ghz.

Godavari has 20% perf/watt advantage starting from 3.7 and increasing with frequency, at lower frequencies the regular Kaveris are about as good in principle since the process yield better dies anyway, current Kaveris should be better than the ones released in early 2014.

 

[Flash831]

The numbers he posted suggest that Godavari get to 4.1GHz at 1.196V, the early 7850K was stuck at 3.7GHz at this voltage and did require 1.330V to get to 4.1Ghz.

Godavari has 20% perf/watt advantage starting from 3.7 and increasing with frequency, at lower frequencies the regular Kaveris are about as good in principle since the process yield better dies anyway, current Kaveris should be better than the ones released in early 2014.

20% better performance/watt seems nearly as the improvements Richland made over Trinity. It's sad this has not come forward through the reviews.
I'm guessing AMD keeps the 7xxx naming scheme to be able to do a quick AM4-transition early in 2016.

 

[DrMrLordX]

It hasn't come forward in reviews because OEMs are (for whatever reason) heavily overvolting almost all FM2+ chips in recent UEFI revisions. As a consequence, a stock 7870k will often run 1.45-1.48v vcore. You really have to tune the voltage by hand to reign in the stupidity, and some boards make that difficult.

 

[Abwx]

20% better performance/watt seems nearly as the improvements Richland made over Trinity. It's sad this has not come forward through the reviews.
I'm guessing AMD keeps the 7xxx naming scheme to be able to do a quick AM4-transition early in 2016.

That s indeed of the same amplitude, Richland did improve the voltages above 4GHz while extending the frequency ceiling by 10%, for Godavari it start to show its advantage over 3.5Ghz.

Planet3D got very good voltage when undervolting a A8-7650K at the 3.3 stock frequency, but overclocking at stock voltage show that Godavari is quite better as they got only 4.2Ghz at the stock 1.35V.

http://www.planet3dnow.de/cms/14312...ri/subpage-uebertakten-und-spannung-absenken/

 

[Flash831]

That s indeed of the same amplitude, Richland did improve the voltages above 4GHz while extending the frequency ceiling by 10%, for Godavari it start to show its advantage over 3.5Ghz.

Planet3D got very good voltage when undervolting a A8-7650K at the 3.3 stock frequency, but overclocking at stock voltage show that Godavari is quite better as they got only 4.2Ghz at the stock 1.35V.

http://www.planet3dnow.de/cms/14312...ri/subpage-uebertakten-und-spannung-absenken/

That's some nice numbers, and they will probably be better for A8-7670K. I hope someone makes a serious review of it soon! :biggrin:

I wonder if it would be possible to release a 4Ghz, 512sp, 65w-tdp part of Godavari :sneaky:
Now it seems like a wise choice to not release Carrizo for FM2+. Kaveri seem to have offered some low hanging fruits.

 

[Abwx]

I wonder if it would be possible to release a 4Ghz, 512sp, 65w-tdp part of Godavari :sneaky:

A10 7890K was finaly released but it s logicaly a 95W SKU...

http://www.computerbase.de/2016-01/...10-7890k-gezeigt-in-wenigen-wochen-im-handel/

 

[cbn]

Reviews of A8-7670K seems to be completely MIA though. :hmm:

Here is the Anandtech review:

http://www.anandtech.com/show/9763/the-amd-a8-7670k-apu-review-rocket-league

 

[Zodiark1593]

It hasn't come forward in reviews because OEMs are (for whatever reason) heavily overvolting almost all FM2+ chips in recent UEFI revisions. As a consequence, a stock 7870k will often run 1.45-1.48v vcore. You really have to tune the voltage by hand to reign in the stupidity, and some boards make that difficult.

The hell the OEMs do that for? Are they looking to kill the chips quickly?

 

[LTC8K6]

The hell the OEMs do that for? Are they looking to kill the chips quickly?

This issue has never been fully explained. I don't think even AMD has commented on it.

You can see it in the CPU-Z screenshot in the Anandtech review.

1.408V for 3.7Ghz which is well above the 1.25V in the specification.

http://images.anandtech.com/doci/9763/AMD A8-7670K_575px.png

http://www.cpu-world.com/CPUs/Bulldozer/AMD-A8-Series A8-7670K.html

 

[deasd]

Overvolting is one of the most critical reason why AMD chips always looks power hungry. This problem already found in old old K8(I still have one) and never be explained......

http://www.planet3dnow.de/cms/14312...ri/subpage-uebertakten-und-spannung-absenken/
Undervolting from 1.35V to 1.05V for A8-7650K has almost 46 watts down.

 

[ZGR]

Sounds like AMD, OEMs, and professional review sites don't care. Really bizarre and makes AMD look even more incompetent.

 

[The Stilt]

The default voltage rating of the chips include a hefty tolerance due safety margins and load-line. The load-line for AMD desktop parts is 1.3mOhms (AM3+, FM2/+), which means that with the maximum current draw of 89.9A (e.g. 7870K, TDC, VDDCR) the voltage can drop up to ~117mV under full load. The end-users who use aftermarket coolers and reduce the amount of vdroop by lowering the load-line, can naturally achieve the same clocks with voltage significantly below the default levels.

 

[TheELF]

This issue has never been fully explained. I don't think even AMD has commented on it.

You can see it in the CPU-Z screenshot in the Anandtech review.

1.408V for 3.7Ghz which is well above the 1.25V in the specification.

http://images.anandtech.com/doci/9763/AMD A8-7670K_575px.png

http://www.cpu-world.com/CPUs/Bulldozer/AMD-A8-Series A8-7670K.html

Cpu world shows 1.25V for Boosted P states,p state being a power saving mode and boosted being the turbo core version of this power saving feature,it's very very different from having a set frequency on all 4 cores that doesn't change and has to work even under the most demanding workloads,think CPU and iGPU at 100% + avx or aes or whatever.

AMD is very creative in these things,they do everything they can to make people believe they have high performance numbers with low TDPs...

 

[William Gaatjes]

Cpu world shows 1.25V for Boosted P states,p state being a power saving mode and boosted being the turbo core version of this power saving feature,it's very very different from having a set frequency on all 4 cores that doesn't change and has to work even under the most demanding workloads,think CPU and iGPU at 100% + avx or aes or whatever.

AMD is very creative in these things,they do everything they can to make people believe they have high performance numbers with low TDPs...

That is nothing special, when the tdp allows it (meaning your cooling solution) the boosted clock will work. It helps for single threaded performance since 3 cores are almost idle and one core fully active, . As long as the internal temperature is not becoming too high, the turbo core function stays active. If you can keep the the die cool enough, the turbo core function will stay on indefinitely. I tested it before with prime95. I have an A10-6700 running at base clock of 3.7GHz and a boosted clock of 4.3GHz.

I am still interested to see what my APU will do when i have the gfx core disabled after i bought a discrete card. I am sure that with no gfx core running side by side with the cpu modules on the same die, the turbo core function will be a lot more effective.

Before you ask why i did not bought a discrete card right away, is because i had no need for it. Now that the upcoming energy efficient mid end cards are becoming cheap and do not have any problem with 1080 resolution. I see a reason to buy a discrete card. In the end, i will have to upgrade the cpu/MB/memory again, but that will not be interesting until 2017/2018.

EDIT:

I should mention that i have Scythe shuriken revb cooler.

 

[Abwx]

Cpu world shows 1.25V for Boosted P states,p state being a power saving mode and boosted being the turbo core version of this power saving feature,it's very very different from having a set frequency on all 4 cores that doesn't change and has to work even under the most demanding workloads,think CPU and iGPU at 100% + avx or aes or whatever.

AMD is very creative in these things,they do everything they can to make people believe they have high performance numbers with low TDPs...

There was two batches of the first 7850K with one using a higher supply voltage and this latter model was tested by Hardware.fr, at 1.352V and 3.7GHz stock settings CPU TDP under Prime 95, that is with AVX and FMA, was 67W, that was already a quite good number.

So no, AMD has nothing to hide TDP wise, it s just that high voltage margins allow very good yields as any chip that has the full 512 SPs working can be sold as a 7800/7850K rather than being downgraded with the GPU being partialy disabled.

Other than that the 7890K benefit from a new cooler wich is very good according to Charlie at SA and wich will be also used for some FXs.

https://semiaccurate.com/forums/showpost.php?p=251972&postcount=14

 

[Phynaz]

Is that doctored photo an AMD press photo?

 

[Zodiark1593]

Cpu world shows 1.25V for Boosted P states,p state being a power saving mode and boosted being the turbo core version of this power saving feature,it's very very different from having a set frequency on all 4 cores that doesn't change and has to work even under the most demanding workloads,think CPU and iGPU at 100% + avx or aes or whatever.

AMD is very creative in these things,they do everything they can to make people believe they have high performance numbers with low TDPs...

Could also be OEMs getting lesser quality chips in exchange for a lower price.

 

[frozentundra123456]

Yea, that is my guess. They can get away with much less careful binning.

 

[LTC8K6]

Cpu world shows 1.25V for Boosted P states,p state being a power saving mode and boosted being the turbo core version of this power saving feature,it's very very different from having a set frequency on all 4 cores that doesn't change and has to work even under the most demanding workloads,think CPU and iGPU at 100% + avx or aes or whatever.

AMD is very creative in these things,they do everything they can to make people believe they have high performance numbers with low TDPs...

I used those numbers because the chip was running at 3700 in the screen shot.

Yes, the stock 3600 speed should require a voltage lower than 1.25V, but the boards will not go anywhere near that low, for whatever reason.

You can manually get the chip to run at the specified voltages with no problem.

If you let the board handle the voltages, you get more than necessary, resulting in unnecessarily high power consumption.

It's a strange situation and it's puzzling to have little info on it.

What's most puzzling is AMD's silence about it. You'd think they'd do something about it since it makes the chips look worse than they really are regarding heat and power.

 

[escrow4]

The parts place close to me where I buy at least 90% of my parts has 1 FM2+ board - 1. The rest are a mix of ultra low end FM2 A68 mobos and dustbin AM3+ mobos. The entire range fits on barely 2 pages. There are only 12 AMD processors remaining. Now this is one of the major stores over here, its name always comes up when you decide where to go. Not a good look for AMD, but indeed, no one cares.

 

[DrMrLordX]

Not a good look for AMD, but indeed, no one cares.

Actually if you consider the fact that FM2+ is effectively EoLed, that's not a bad thing. Carrying a ton of excess FM2+ inventory into what is going to be a major release year for AMD would be bad for them.

They need AM3+ and FM2+ gone and AM4 pushed to the fore, as quickly as possible.

 

[rancherlee]

I've build a 7850k rig 2 months ago with a recent build week chip and its at 1.35V @ 3.7 and 1.40v turbo. It only manages 4.1 @ stock voltage with a 4.2 turbo. Just some reference points from a recent 7850k