Richland & Kabini rumours

[Nemesis 1]

Yeah, if the benchmark is a "score" where higher is better, then new/old is the logical way to go. If the benchmark is a "time to completion" style (like Sunspider), then old/new is the way to go (e.g. 100ms for new vs 200ms for old gives a new part that is twice as fast).

Thats OK . I don't care but for sandy we did the same thing and did it the other way . Numerator denuminator who new. expressed as a fraction than it would be 1400/1150 correct the numerator being 1400

 

[frozentundra123456]

Kitguru charted some extrapolations from performance claims by intel and AMD.

http://www.kitguru.net/components/g...vs-haswell-gt3-graphics-performance-analysed/

Richland would still have a healthy performance lead over haswell, provided the numbers are accurate. I assume AMD's numbers are since they show actual scores with their performance claims. Has intel given any performance numbers to back up it's 2x increase over ivy bridge? Couldn't find.

By the article's own admission, these are very speculative numbers based on each manufacturer's "best case" estimate. Lets not get too excited until we see real numbers by independent reviewers over a wide range of games.

Even using these projected best case numbers, one game is totally unplayable, and I would expect other demanding current and future games would be unplayable as well. Not to mention that by then the 8000 series cards will be out and hopefully will show a performance increase as well. So these igps will still trail low/mid range discrete cards by a huge amount. I just cant get excited about APUs in the desktop. Perhaps the lower power chips in tablets will be quite interesting though.

Edit: Based on the experience with my current Acer A100, I would never consider another Android tablet. I would only consider x86 running Windows, or (gasp) Apple.

 

[Nemesis 1]

Kitguru charted some extrapolations from performance claims by intel and AMD.

http://www.kitguru.net/components/g...vs-haswell-gt3-graphics-performance-analysed/

Richland would still have a healthy performance lead over haswell, provided the numbers are accurate. I assume AMD's numbers are since they show actual scores with their performance claims. Has intel given any performance numbers to back up it's 2x increase over ivy bridge? Couldn't find.

deleted I keep for getting A small part of this forum only cares about GPU performance Cpu doesn't matter . SO this was gpu only Why didn't kit include the GT3 numbers in haswell I mean the 55 watt mobile haswell with G3 isn't going to be 30% increase. thats for low wattage I would expect to see closer to 100% if not more He gives haswell a 15% gain that pure BS.

 

[Elixer]

Unless I missed it someplace, I wonder just when AMD will start doing PCIe 3.x ?
Surely the new crop of CPUs will offer this ?

 

[Gideon]

The score for the 2012 AMD A10-4600M on the PCMark 7 Overall benchmark was 1965 and the 2012 AMD A8-4555M was 1650, while the “Richland” 2013 AMD A10-5750M was 2175 and the 2013 AMD A8-5545M was 1850

http://www.amd.com/us/press-releases/Pages/amd_unveils_new_apus.aspx

The thing that really bothers me about these AMD tests is that they do not mention what storage was used, yet PCMark total score is heavily influenced by storage (look Llano HDD vs SSD)

Especially as they did mention a HDD in the battery life test (and only there). Excuse me for being so paranoid but AMD marketing was so damn notorious about Bulldozer that it wouldn't surprise me if they resorted to HDD -> SDD shenanigans when comparing Trinity -> Richland platform performance (I have no trouble believing Jaguar blows Bobcat away).

 

[dbcoopernz]

Given the low scores for all platforms, one might assume that all used HDD's?

 

[piesquared]

By the article's own admission, these are very speculative numbers based on each manufacturer's "best case" estimate. Lets not get too excited until we see real numbers by independent reviewers over a wide range of games.

Even using these projected best case numbers, one game is totally unplayable, and I would expect other demanding current and future games would be unplayable as well. Not to mention that by then the 8000 series cards will be out and hopefully will show a performance increase as well. So these igps will still trail low/mid range discrete cards by a huge amount. I just cant get excited about APUs in the desktop. Perhaps the lower power chips in tablets will be quite interesting though.

Edit: Based on the experience with my current Acer A100, I would never consider another Android tablet. I would only consider x86 running Windows, or (gasp) Apple.

Yep, I noted that they were claims from both manufacturers. But as the article suggests, even if it were best case against worst case, haswell would still be behind by a large margin. I just found it interesting because the general consensus seemed to be that haswell was going to dominate gaming. Richland appears be a quite potent chip and should be capable of fairly demanding game settings and resolution.

 

[Nemesis 1]

Ya me to I was just looking at Anand benches and that alot of meat to make up but SB and IB both work on the desk for what i want gaming is discrete only for me .

 

[Nemesis 1]

The richland is unchanged as far as cores go were did the GPU speed come from ?

 

[Nemesis 1]

Yep, I noted that they were claims from both manufacturers. But as the article suggests, even if it were best case against worst case, haswell would still be behind by a large margin. I just found it interesting because the general consensus seemed to be that haswell was going to dominate gaming. Richland appears be a quite potent chip and should be capable of fairly demanding game settings and resolution.

If you look Haswell is only showing 15% increase in performance on Igpu . Intel gave way better numbers than that . So no he didn't go by what intel said . Also didn't AT do a recent bench mark between these 2 and Intel made up alot of ground threw drivers The 55watt mobile with GT3 will be closer to 100% increase

 

[piesquared]

If you look Haswell is only showing 15% increase in performance on Igpu . Intel gave way better numbers than that . So no he didn't go by what intel said . Also didn't AT do a recent bench mark between these 2 and Intel made up alot of ground threw drivers The 55watt mobile with GT3 will be closer to 100% increase

The chart shows a 100% increase for GT3 over hd4000.

 

[Abwx]

The chart shows a 100% increase for GT3 over hd4000.

Yet , the same guy tried to explain us how to convert a performance
delta into a %age difference...

More seriously , i think that his aim is deliberate thread crapping.

 

[itsmydamnation]

from my readings GF have a better gatefirst process @28nm then they did @32. That could account for a bigger then expected performance difference. Most people have assumed not much of a bump from process.

 

[Abwx]

from my readings GF have a better gatefirst process @28nm then they did @32. That could account for a bigger then expected performance difference. Most people have assumed not much of a bump from process.

Process is not all , just give credit to whom it is due , AMD
announced better perfs/watt thanks to their automated design ,
seems that it did help them substancially given their limited
human ressources and they ultimately delivered the said promise....

http://technewspedia.com/the-new-mic...d-steamroller/

 

[itsmydamnation]

i said larger then expected, I didn't say it was only process. But if you consider that bulldozer released on 12oct 2011 , piledriver released in may 2012 and now richland is "shipping to oems" thats quite a perf bump for 15 months.

 

[MightyMalus]

To Nemesis: HSW iGPU's won't be that high performant. It will surpass Trinity's by very little.
Check the Haswell thread to see what I wrote, and help, if you can.


But, Into more important matters, anyone else noticed that AMD tested all the APU's with less than suggested RAM frequencies? I see room for improvements, worth while improvements.

 

[dbcoopernz]

Maybe they've improved their RAM controller?

 

[itsmydamnation]

Maybe they've improved their RAM controller?

theres nothing wrong with the memory controller in terms of throughput ( you cant push memory as hard as you can on intel). you will also notice how increasing NB freq has almost 0 performance increase. You can also see from GPU compute that the GPU can just about reach max theoretical throughput.

your looking at the wrong number in AIDA if you want to see the major screw up the memory system, unless you know of workloads that need more then 15gb/s of memory throughput for a single core.

 

[Abwx]

i said larger then expected, I didn't say it was only process. But if you consider that bulldozer released on 12oct 2011 , piledriver released in may 2012 and now richland is "shipping to oems" thats quite a perf bump for 15 months.

That s likely due to their eager adoption of automated design tools ,
a risky move but they had no choice given their shrinking workforce.

 

[MightyMalus]

I just wish they should have tested the APU's with their stock max RAM frequencies.

 

[mikk]

Kitguru charted some extrapolations from performance claims by intel and AMD.

http://www.kitguru.net/components/g...vs-haswell-gt3-graphics-performance-analysed/

Richland would still have a healthy performance lead over haswell, provided the numbers are accurate. I assume AMD's numbers are since they show actual scores with their performance claims. Has intel given any performance numbers to back up it's 2x increase over ivy bridge? Couldn't find.

This chart is nonsense. HSW GT3 is a mobile only chip and comparing a mobile chip against desktop chips is plain stupid. If you compare mobile vs mobile you will realize that (the higher TDP versions) HSW GT3 should have the upper hand this year. And these benches also are cherry picked by kitguru. 5800k is less than 2x to 3770k and not more than 2x.

 

[MightyMalus]

IVB HD4000 Peak performance is probably 294.4GFlops. (16*8*2*1.150MHz = 294.4GFlops)
HSW HD4600 Peak performance is probably 384GFlops. (20*8*2*1.200MHz = 384GFlops)

About a 30% increase. Obviously this is comparing the desktop GT2 iGPU's.

I expect the GT3 to be as performant as the GT2 on the desktop.
(40*8*2*600MHz = 384GFlops)

But surely out of the league of anything Richland, let alone Kaveri. Will have to see how it does against Kabini actually. If Kabini shows near performances of Trinity or slightly higher, <5%, in gaming and entertainment, HSW will have some competition for sure. And Atom will be of no use, on tablets or entertainment.

Doesn't really look bad for AMD. I think.

 

[mikk]

I expect the GT3 to be as performant as the GT2 on the desktop.
(40*8*2*600MHz = 384GFlops)

Nonsense. Even GT2 mobile is faster than GT2 desktop.

 

[inf64]

It actually looks much better than 2012 looked. I'm still surprised how well Kabini(Jaguar) performs. It's a tiny core but packs so much punch that it's crazy. It does it all while staying in the same or lower TDP bracket as Brazos,quite an achievement from perf./watt POV.

 

[krumme]

@inf64

Yes the outlook is better, but that doesnt say so much

Richland is a wellcome surprise, but i dont think it matter at all for the business, as its big die sold for low-mid prices. To what degree, does AMD still have to "support" GF? - the wsa is still dark and clouds all perpectives ? - we dont know anything. I fail to see the purpose of this product.

Kabine will sell like hotcakes especially on the emerging markets, exactly like Brazos, because Haswell is another price bracket, but its still low profit market.

The performance of Temash compared to the new Atoms will be very decisive for AMD profit in this year. Temash needs to have a performance edge, because Atom is going to be small, cheap and geared for very low power.