Kaveri refresh name "Godavari" official?

Shehriazad

Senior member
Nov 3, 2014
555
2
46
I haven't seen anything about it lately until I saw a post about this link here: Click

Not sure if I should hope for anything on this "refresh". I still hope that more happened than AMD just taking their Kaveri chips and renaming them with slightly altered clocks. Someone else got some other news on the chip? I remember seeing some slides 2 months ago...back then everyone was doubting them simply because the name was weird.


Edit: Here a screenshot for people that hate clicking on links.
jxgvqszo.jpg
 
Last edited:

podspi

Golden Member
Jan 11, 2011
1,982
102
106
wow, guess so. I figured there was veey little chance that would be the actual name.

Here's hoping, but I'm not expecting much from a refresh... But OEMs like to have their new models every year...
 

Essence_of_War

Platinum Member
Feb 21, 2013
2,650
4
81
So we have a kaveri refresh, but still no indication of when, if ever, we'll see Berlin and ECC support?
 

DrMrLordX

Lifer
Apr 27, 2000
22,761
12,772
136
Heck if I know. Probably the same reason why the name Godavari wound up on that screenshot people have been passing around ever since wccftech or sweclockers or whoever turned it up first started spreading it.

Someone, somewhere got the idea that the chips would have a new part name, and they used the iGPU name from Mullins for whatever bizarre reason. If it sticks in enough places, we'll be calling it Godavari for now until the end of time despite the fact that AMD never intended for that to be the internal name for the refresh part.

But, truth is, Godavari has already been assigned to an element of an existing part. I'd be surprised if AMD recycled the name for Kaveri refresh.

edit: How odd. It looks like Mr. Wei has been working on UEFI support for "Godavari". Maybe they really did recycle the name.

So we have a kaveri refresh, but still no indication of when, if ever, we'll see Berlin and ECC support?

AMD demoed Berlin back in April of 2014. If we were going to see that part sold through normal channels alongside other Opterons, we'd probably have seen it by now. I'm guessing it's been relegated to various custom orders for clients looking to experiment with clusters of HSA hardware. Carrizo (or an Opteron variant thereof) will probably be more desirable for that sort of thing anyway.
 
Last edited:

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
AMD demoed Berlin back in April of 2014. If we were going to see that part sold through normal channels alongside other Opterons, we'd probably have seen it by now. I'm guessing it's been relegated to various custom orders for clients looking to experiment with clusters of HSA hardware.

Berlin was a huge flop for AMD, we still don't see it because nobody bought it. And nobody running a real application on servers is remotely interested in buying hardware to use with an still unreleased set of development tools, especially from a company which the history of poor software support like AMD.
 

DrMrLordX

Lifer
Apr 27, 2000
22,761
12,772
136
Berlin was a huge flop for AMD, we still don't see it because nobody bought it. And nobody running a real application on servers is remotely interested in buying hardware to use with an still unreleased set of development tools, especially from a company which the history of poor software support like AMD.

Nobody bought it 'cuz AMD didn't sell it (except maybe through ODMs or other small custom orders, the details of which are often obscured or simply unavailable). Someone got smart and realized that the use case for Berlin put it in too small a niche for AMD to bother cluttering up the channel with more parts that won't sell in volume. It would still be somewhat useful for people running OpenCL software where turnaround time can be affected by latency, but for most current OpenCL applications, dGPUs are useful and therefore probably more cost-effective than APUs.

And of course, HSA was not exactly useful back in mid 2014 when they could have/should have launched the chips, so that was the final nail in the coffin.

Being more desirable than something that looks to be vaporware isn't a terribly difficult hurdle to clear. :(

Ha ha! Ouch for AMD. Seriously though, Berlin would have been 65-95W TDP steamroller with what, 3.0-3.4 ghz and unknown turbo, and some GCN cores, with a . . . let's just say "difficult to use" HSA toolset and little compelling argument for use in OpenCL apps vs dGPUs.

Carrizo, or some Opteron derivative thereof, should improve the perf/watt significantly which will make it more useful for cluster-style setups. And with the HSA software stack finally shaping up, the HPC world will probably start wanting hardware like this. As for anyone else, well, we'll see.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Nobody bought it 'cuz AMD didn't sell it (except maybe through ODMs or other small custom orders, the details of which are often obscured or simply unavailable).

I'm not sure what you are trying to say here. Are you implying that AMD didn't *want* to sell Berlin to its customers? Given the state of their financials, I don't think they could afford to not to sell Berlin had *anybody* minimally relevant wanted it.

What I think it happened is that it offered Berlin to its Microservers and Opteron customers and the product was decisively shunned by these customers, to the point that AMD didn't even bother to market it at all. 0, or close to 0 sales.
 

Abwx

Lifer
Apr 2, 2011
11,855
4,832
136
Guess the SKU list can be posted again as a refresher for those that forgot:

AMD-Godavari-APU-Lineup.jpg

Frequencies are not right in this slide it seems, word is from an SA member that has some insider knowledge of AMD future products caracteristics.
 

DrMrLordX

Lifer
Apr 27, 2000
22,761
12,772
136
Are you implying that AMD didn't *want* to sell Berlin to its customers?

Maybe they got to the offer stage and maybe they didn't, but I think Berlin went the way of Larrabee. An interesting product that got pulled 'cuz nobody was going to buy enough of them to make it worth the trouble to sell them.
 

DrMrLordX

Lifer
Apr 27, 2000
22,761
12,772
136
Your words, not mine. Besides, Xeon Phi would like to take issue with the notion that Larrabee was a flop sir! But that is besides the point, and off the topic.

So has anyone heard any news about when we should expect Kaveri refresh in stores? I noticed people are already buying 7650ks (which, admittedly, were announced before Kaveri refresh) and they seem pretty good for the price.
 

Shehriazad

Senior member
Nov 3, 2014
555
2
46
I'm still suspicious about the Athlon parts. The 870K is way lower clocked than the 860K...at least according to that picture.


Since I don't expect AMD to actually have changed a whole lot...I'd guess that this CPU would end up being the counterpart to the 8850K...and thus boost to 4.1Ghz.

Clocking the Athlon "top" chip 300 Mhz slower than the previous stepping but Clocking the APU chip (that needs more power and has more heat etc) 100 Mhz above the previous top Kaveri? Yea...no.




That said...I am still curious if ANYTHING changed on the technical side or if they simply clocked their stuff a few percent higher. There are certainly 1-2 things that could need some fixing on the technical side.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
22,761
12,772
136
I would not hold out too much hope that notable core revisions went into Kaveri refresh. My guess is we'll see some nicer clockspeeds post-overclocking thanks to maturation of 28nm planar, though even that is not guaranteed.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
I would not hold out too much hope that notable core revisions went into Kaveri refresh. My guess is we'll see some nicer clockspeeds post-overclocking thanks to maturation of 28nm planar, though even that is not guaranteed.

How much more mature can 28nm get? :p
 

DrMrLordX

Lifer
Apr 27, 2000
22,761
12,772
136
Ask a GF process tech. 28nm planar didn't show up all that well for Kaveri chips dating back to early 2014. Judging by recent improvements in FX performance/watt, there may also be some process optimizations and/or core optimizations that can be done to make Kaveri a little less volt-hungry past the 4.0-4.2 ghz threshold. Whether or not that will yield better clockspeeds or just better efficiency is anyone's guess.

But just about every process node can be subject to improvements with sufficient iteration, to a point. It's probable that early 28nm products (initial Kaveri chips) were not on a process at that point.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
So we have a kaveri refresh, but still no indication of when, if ever, we'll see Berlin and ECC support?

The AM3+ and AM1 consumer processors have ECC capability enabled. For those processors is up to the mobo manufacturer to provide ECC support on the board (which some do).

Hopefully AMD has decided to enable ECC on the consumer Kaveri Refresh processors this round (so we can have consumer boards with ECC). That and provide and easy way to disable CPU downclocking under iGPU load.
 

Essence_of_War

Platinum Member
Feb 21, 2013
2,650
4
81
The AM3+ and AM1 consumer processors have ECC capability enabled. For those processors is up to the mobo manufacturer to provide ECC support on the board (which some do).

I am aware. I was talking specifically about berlin which was supposed to be on the FM platform.
 

Shehriazad

Senior member
Nov 3, 2014
555
2
46
I haven't heard anything "official" about the refresh in months now....siiiigh. AMD be AMDing. xD