AMD Kaveri OC On Planet Neptune

inf64

Diamond Member
Mar 11, 2011
3,221
2,354
136
It's nice to see it can operate at such temps but I wouldn't hold my breath for some BD/PD-like frequency records. PD is done on a superior process node.
 

Ajay

Lifer
Jan 8, 2001
11,464
5,195
136
Yes please. This is exactly how BD was "leaked". Lots of clock speed numbers on exotic cooling that nobody will actually use along with no bench numbers.
This is LN2. BD was overclocked on LHe (-269 degrees Celsius), IIRC. Now, liquid Helium is definitely exotic!
 

schmuckley

Platinum Member
Aug 18, 2011
2,335
1
0
LHe is about 7x the cost of LN2..if you can even get it..helium is rare..
LN2 can be extracted from air :D
 

Ajay

Lifer
Jan 8, 2001
11,464
5,195
136
LHe is about 7x the cost of LN2..if you can even get it..helium is rare..
LN2 can be extracted from air :D
Well, AMD was sponsoring that effort and they have the resources to get industrial grade supplies, like LHe. I can get LN2 for around $2/liter from UNH Durham. A bit of a drive, but worth it and it would be cool if I had the dosh to build a LN2 system (pun intended).
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
998
126
Now give the ram some liquid nitrogen so the graphics can really perform. :)
 
Last edited:

Gikaseixas

Platinum Member
Jul 1, 2004
2,836
218
106
Disappointed from the moment i knew it would not have a GPU as beefier as the ones in the current generation consoles. At this pace Intel will catch up in no time.
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
There are pictures on Facebook that show AMD Kaveri being overclocked at temps of -186C (-295F) which is like high noon on Neptune! Since Intel and AMD have CPUS that overcook to more than 7 and 8 ghz, it could be that Kaveri may reach a new record.
Maybe this is AMD strategy to beat Intel to the punch. Buy an AMD APU and get 12 months worth of LN2 supply.

Gentlemen, this is very strange:

- No performance leaks
- "we see growth in desktop"
- "Why partner with AMD?"
- Kaveri overclock on LN2

Things are not looking good for Kaveri launch.
 

avtek21

Member
Oct 26, 2013
54
0
0

The AMD leaks this time included "bench leadership" slides and playing BF4 without Mantle plus these OC pictures, so that is more than drip of info for their APU.
 

inf64

Diamond Member
Mar 11, 2011
3,221
2,354
136
That slide has already been posted and discussed.

Here is short version:
OpenCL-> barely relevant benchmark for desktop users
40% "uplift" over 4670K in 3Dmark firestrike is not the same as it being that much faster in actual games. 3D mark is basically "marketing benchmark".
BF4 bundle is a nice deal to attract some customers. Bad luck the game is infested with bugs and won't run that well in actual MP on any APU, Kaveri included.
8% over 4670K in PCMark8 can be disregarded since 6800K already outperforms 4670K in same benchmark by ~4%. So top Kaveri will be faster than 6800K by whooping ~4%? That's not good at all. 6800K can overclock like crazy too, let's see Kaveri on that front.
New SR cores is more of a marketing bullet point than anything else
 
Last edited:

NTMBK

Diamond Member
Nov 14, 2011
9,785
3,814
136
Maybe this is AMD strategy to beat Intel to the punch. Buy an AMD APU and get 12 months worth of LN2 supply.

Gentlemen, this is very strange:

- No performance leaks
- "we see growth in desktop"
- "Why partner with AMD?"
- Kaveri overclock on LN2

Things are not looking good for Kaveri launch.
Or AMD have learnt from Bulldozer and are keeping their mouth shut before launch. There's a reason they fired most of their PR team. ;)

We didn't see any reliable Hawaii performance leaks before launch, either.
 

inf64

Diamond Member
Mar 11, 2011
3,221
2,354
136
Or AMD have learnt from Bulldozer and are keeping their mouth shut before launch. There's a reason they fired most of their PR team. ;)

We didn't see any reliable Hawaii performance leaks before launch, either.
I'm afraid this won't be a reprise of Hawaii launch. :/
 

NaroonGTX

Member
Nov 6, 2013
106
0
76
What are people really expecting from Kaveri? Do people honestly expect it to somehow be on par with Haswell? It'll be decently faster than Richland in both CPU/GPU and be the first APU with true HSA support (which is honestly the biggest deal about the processor.) Only people with ridiculous expectations will be disappointed.

As for no performance leaks, that's good. As mentioned, there weren't any reliable leaks for Hawaii, either. AMD is keeping everyone quiet until the 'big unveil' next month, which is good. Otherwise we would've probably had another JF-AMD running around telling people completely fabricated things, getting peopled hyped in the wrong way.
 

inf64

Diamond Member
Mar 11, 2011
3,221
2,354
136
What are people really expecting from Kaveri? Do people honestly expect it to somehow be on par with Haswell? It'll be decently faster than Richland in both CPU/GPU and be the first APU with true HSA support (which is honestly the biggest deal about the processor.) Only people with ridiculous expectations will be disappointed.

As for no performance leaks, that's good. As mentioned, there weren't any reliable leaks for Hawaii, either. AMD is keeping everyone quiet until the 'big unveil' next month, which is good. Otherwise we would've probably had another JF-AMD running around telling people completely fabricated things, getting peopled hyped in the wrong way.
Out of curiosity, what do you consider "decently faster than Richland in both CPU/GPU", in pure % numbers?
Remember, Richland that it will be compared to will be 6800K and not some 3.7Ghz base clock part "just to make it fair".
 

NaroonGTX

Member
Nov 6, 2013
106
0
76
I don't wanna throw out any random percentages, but the jump of PD from BD was pretty marginal because it was some quick and easy fixes. SR was a much bigger change in the uarch than BD -> PD, so I'm pretty sure it'll be a big leap (for AMD). They need it, since it's 2013 and right now the best AMD has is Piledriver, which per core is still only offering Phenom II levels of performance, a uarch from ~2008.
 

Ajay

Lifer
Jan 8, 2001
11,464
5,195
136
I don't wanna throw out any random percentages, but the jump of PD from BD was pretty marginal because it was some quick and easy fixes. SR was a much bigger change in the uarch than BD -> PD, so I'm pretty sure it'll be a big leap (for AMD). They need it, since it's 2013 and right now the best AMD has is Piledriver, which per core is still only offering Phenom II levels of performance, a uarch from ~2008.
It certainly would have been a bigger jump if Kaveri was being produced on a very high performance process (28nm SOI/FD-SOI w/hidden gate). But, it's looking like 28nm Bulk is going to give the stock clocks a considerable haircut - hence the big mystery in terms of actual performance gains. On the face of it, it looks like AMD made some solid improvement to Kaveri B, but their sub par foundry partner may eat up much of AMD's gains.
 

avtek21

Member
Oct 26, 2013
54
0
0



Here is short version:
OpenCL-> barely relevant benchmark for desktop users
It is very relevant if bench provides measure of performance for type of apps that are used by hundreds of millions of users in the world.

x264
x264 is the world leading video encoder library when measured with quality, speed and downloads in mind. It is estimated there has been over a billion downloads of x264 as part of other applications such as VLC and Handbrake to name a two of hundreds.

MCW ported x264's look ahead pre-encode thread to be nearly entirely OpenCL accelerated. *All slice-cost projections are performed on the GPU, giving a speedup for presets where the look ahead is a bottleneck, and providing an offload for CPUs with few cores.

VLC
VLC is the world’s most downloaded free and open source cross-platform multimedia player and framework that plays most types of multimedia files. In our optimization work, we first integrated an AMD APU exclusive hardware feature - steady video - into VLC, to stabilize shaky video played with VLC. Then we used OpenCL to optimize VLC’s scaling filter, which is used to enlarge or shrink the video on the fly during playback. This OpenCL optimization has achieved up to 10x speedup on MD Llano and 18x speedup on AMD Trinity compared to the Intel SandyBridge CPU. De-noise filter in VLC was used to reduce the noise in video. Together these features/optimizations enable much better video playback experience on AMD Fusion platforms.



http://www.multicorewareinc.com/services.html
 

inf64

Diamond Member
Mar 11, 2011
3,221
2,354
136
It certainly would have been a bigger jump if Kaveri was being produced on a very high performance process (28nm SOI/FD-SOI w/hidden gate). But, it's looking like 28nm Bulk is going to give the stock clocks a considerable haircut - hence the big mystery in terms of actual performance gains. On the face of it, it looks like AMD made some solid improvement to Kaveri B, but their sub par foundry partner may eat up much of AMD's gains.
Hit the nail on the head with your post ;).
 

Homeles

Platinum Member
Dec 9, 2011
2,580
0
0
But, it's looking like 28nm Bulk is going to give the stock clocks a considerable haircut - hence the big mystery in terms of actual performance gains. On the face of it, it looks like AMD made some solid improvement to Kaveri B, but their sub par foundry partner may eat up much of AMD's gains.
The Fab Club seems to have intentionally obfuscated the performance advantages of 28nm bulk over 32nm bulk, making the guessing game difficult. It's hard to get an idea of just how bad the hit will be, but we can likely assume that the Fab Club's gate-first 28nm has equal or lower performance than TSMC's gate-last 28nm. Given that TSMC's 28nm is decidely slower than GloFo's 32nm PDSOI, we can safely assume that there will be a noticeable regression in clock speed. Correspondingly, Kaveri's frequencies suffer.

The regression from SOI to bulk for AMD is somewhat comparable to the regression from planar to FinFETs for Intel. Intel's FinFETs only regress at high voltage, while making considerable gains at low voltage, while the loss of SOI should result in a penalty across all operating voltages.

Basically, in a sense, at 28nm (AMD) and 22nm (Intel), enthusiasts have encountered and will continue to encounter frequency regressions while overclocking. AMD's stock clocks will also take a hit.

There's light at the end of the tunnel, though. GloFo moves to gate last with 20nm, which will result in better performance in addition to traditional scaling. On the Intel side of things, the change from planar to FinFETs will only be felt once, and since it's a "true full node improvement," we should once again see maximum operating frequencies, i.e. max overclocks, push forward. Broadwell should overclock at least as well as Sandy Bridge, excluding unforeseen problems. Whatever AMD ends up putting on 20nm should clock very high as well, especially if they move back to SOI.
 

ASK THE COMMUNITY