Temash peformance tablet SoC

[Exophase]

You are overestimating thoses ARM15 , likely that they have
at most half Kabini s throughput at same frequency/core.

I don't think you're well informed regarding Cortex-A15's performance at all. Look at this benchmark for instance:

http://7-cpu.com/

The relevant comparison is 2 thread vs 2 thread, where the Cortex-A15 has higher perf/MHz than the Bobcat. When I said 1.2GHz vs 1GHz I was actually being pretty generous.

 

[Abwx]

I don't think you're well informed regarding Cortex-A15's performance at all. Look at this benchmark for instance:

http://7-cpu.com/

The relevant comparison is 2 thread vs 2 thread, where the Cortex-A15 has higher perf/MHz than the Bobcat. When I said 1.2GHz vs 1GHz I was actually being pretty generous.

Almost on par with a Sandy Bridge at same frequency...

Let s see : on decompression mode :

A15 2 Threads 1.7ghz 3560 -------> 2094/GHZ

i5 2400 2 Threads 3.1ghz 6830 -------> 2203/GHZ

I dont think that theses numbers have any relevance
to compare the respective architectures performances
given that they are hugely optimistic for the A15.

95% of a SB IPC , are you serious..??..

 

[ShintaiDK]

Compression numbers are terrible in terms of CPU benchmarking.

 

[Idontcare]

Compression numbers are terrible in terms of CPU benchmarking.

Why is that? Because it is limited by ram bandwidth or something? Just curious, never thought much about it before.

 

[Ferzerp]

Why is that? Because it is limited by ram bandwidth or something? Just curious, never thought much about it before.

Perhaps he means because sometimes certain architectures are 1 trick ponies and showing their single strength can create a misleading view? Look at BD, if all we looked at were 7-zip benchmarks, it would have appeared a very compelling processor.


Of course, the state of bottom of the barrel CPU benchmarking is atrocious anyway. People actually believe that running javascript benchmarks on different rendering engines provides meaningful data in regards to the performance of a given pair of CPUs.

 

[Exophase]

Perhaps he means because sometimes certain architectures are 1 trick ponies and showing their single strength can create a misleading view? Look at BD, if all we looked at were 7-zip benchmarks, it would have appeared a very compelling processor.

Probably because it scales well with threads, and BD/PD highly favor well threaded workloads. The case I gave is like for like, where both processors have the same core and thread count and both tests use 2 threads (I ignored the 1 and 4 thread runs).

Note that it includes both compression and decompression. If the argument is that it's a poor benchmark because it has no real world relevance then that's not a strong case either, given that decompression is used all the time...

The other reason I like this test is because it's native code, compiled with a compiler that's relatively mature for both CPUs, and it doesn't have a ton of hand coded SIMD. I'd much sooner take it over the usual fare of Javascript benchmarks.

Still, I'd like to see even one benchmark or technical rationale that would suggest Cortex-A15 has anywhere close to only half the perf/MHz that Jaguar does.

 

[sm625]

Unless AMD can manage something truly innovative, like 4GB of PCM integrated into their SoC, they really have no chance. Hell, even with 4 or even 8GB of PCM integrated right into the cpu, the part probably still wouldnt really take off. That is how broken and compromised this market is. Intel would crush the oems for using AMD chips, and microsoft would do the same simply by not making use of the nonvolatile RAM in a halfway intelligent manner. PCM (and other nonvolatile RAM techs) are knocking on the door, but microsoft hasnt even begun to recode their OS to use it. So if you arent an intel, apple or samsung, you could be totally hung out to dry by investing in an expensive new technology only to be thwarted by a bigger companies deliberate lack of support for it. That problem has always plagued AMD.

 

[ShintaiDK]

Why is that? Because it is limited by ram bandwidth or something? Just curious, never thought much about it before.

Memory bandiwidth got a large impact. Latency a minor impact, but still. In WinRAR alone you can easily get a 10% performance delta just by small differences in memory.

The decompression number was also the only used. Note that the i5 2400 gets an even higher number compressing (7370 vs 6830). Meaning there is a bottleneck somewhere.

The A15 in compression gets 2270 at 1.7Ghz. Or 1335/Ghz.
The i5 2400 (without turbo) gets 7370 at 3.1Ghz. Or 2377/Ghz.

Then the rosy picture is suddenly quite different.

 

[grimpr]

Just use 7zip on AMD chips, faster, better and free.

 

[Abwx]

Memory bandiwidth got a large impact. Latency a minor impact, but still. In WinRAR alone you can easily get a 10% performance delta just by small differences in memory.

The decompression number was also the only used. Note that the i5 2400 gets an even higher number compressing (7370 vs 6830). Meaning there is a bottleneck somewhere.

The A15 in compression gets 2270 at 1.7Ghz. Or 1335/Ghz.
The i5 2400 (without turbo) gets 7370 at 3.1Ghz. Or 2377/Ghz.

Then the rosy picture is suddenly quite different.

It is specified that compression doesnt use fully the cores
and is more influenced by bandwith and I/O than decompression
where the CPU is at 100%...

Did you think that i used the decompression numbers by chance.?..

 

[Abwx]

Still, I'd like to see even one benchmark or technical rationale that would suggest Cortex-A15 has anywhere close to only half the perf/MHz that Jaguar does.

If it was remotely on par it would populate the notebooks
this year but you wont see this happen and it s not by
chance either....

 

[ShintaiDK]

It is specified that compression doesnt use fully the cores
and is more influenced by bandwith and I/O than decompression
where the CPU is at 100%...

Did you think that i used the decompression numbers by chance.?..

Oh yes.

Compression doesnt use the cores fully? But decompression does? Thats certainly a new one.

7-Zip (r) [64] 9.22 beta : Igor Pavlov : Public domain : 2011-04-18
RAM size: 16353 MB, # CPU hardware threads: 4
RAM usage: 850 MB, # Benchmark threads: 4
Dict Compressing | Decompressing
Speed Usage R/U Rating | Speed Usage R/U Rating
KB/s % MIPS MIPS | KB/s % MIPS MIPS
22: 13000 295 4287 12646 | 150558 352 3856 13584
23: 12168 287 4313 12397 | 145962 347 3848 13357
24: 12550 322 4190 13494 | 144597 353 3804 13415
25: 12342 328 4302 14092 | 142696 355 3821 13559

The entire benchmark also looks wierd. 13000KB/sec gets a lower rating than 12342KB/sec. 150558KB/sec basicly equals 142696KB/sec.

And the hardware, in terms of the i5 2400 vs the A15.
2 GB DDR3L-1600 32-bit, 2-port, 12.8 GB/s. Samsung Chromebook (Samsung XE303C12).
Nothing is listed under the i5 3100. But I would assume that DDR3-1333 is being used. So in terms of bandwidth scaling compared to the core speed, the A15 is better equipped.

 

[Nemesis 1]

If it was remotely on par it would populate the notebooks
this year but you wont see this happen and it s not by
chance either....

So all of the sudden power usage doesn't matter. AMD is going to get big share with this product . They have a small window that will be closed befor there products make it to market.

The arm guys may not like it but the game is over . Medfield ended it .This is now back to AMD/Intel struggle with intel holding up arm in the phone space while AMD gets product ready to enter that space in the future,

 

[piesquared]

So all of the sudden power usage doesn't matter. AMD is going to get big share with this product . They have a small window that will be closed befor there products make it to market.

The arm guys may not like it but the game is over . Medfield ended it .This is now back to AMD/Intel struggle with intel holding up arm in the phone space while AMD gets product ready to enter that space in the future,

Intel is dead in smartphones. Samsung and Qualcomm will rule the roost there for years to come.

 

[Nemesis 1]

Intel is alive in phones and taking share . Your not being honest here and everyone knows it . The haswell GT3 demo you were not being honest and now this . Thats 2 I will keep count of your honesty from here on out , publicly

 

[piesquared]

Intel is alive in phones and taking share . Your not being honest here and everyone knows it .

Intel is dead in phones. Not only do they have to overcome the ARM ecosystem, they have to overcome Snapdragon 800 and Exnyos 5 Octa.

I think you got the honesty part mixed up... It was intel that faked the HD4000 circus and used slight of hand for their 7W ivy bridge claim. I think you can understand why not many take their latest haswell 'demo' very seriously.

 

[Nemesis 1]

The 900 looks pretty good, What part you missing about medfield being intels first phone that your not precieving. How is it that your here talking about AMD future products singing it praises yet no one else has anything in the works that compare . I am intel fanboy . But you are twisted

 

[piesquared]

The 900 looks pretty good, What part you missing about medfield being intels first phone that your not preceiving. How is it that your here talking about AMD future products singing it praises yet no one else has anything in the works that compare . I am intel fanboy . But you are twisted

Personal attacks don't bother me in the least. I grew up with 5 brothers and raised on a ranch, take your best shot!

I'm praising AMD's products because they are awesome products. They co-own the gaming ecosystem, intel is an afterthought for big game developers, other than to have a fall back path for the lowest possible hardware specifications. People like to blame consoles for some of the poorer game content. In reality, game developers need to allow intel hardware to run since they weaseled their way into the market by saddling the industries worst GPU's onto their CPU's, and taking the market share lead. And before that they milked the industry with GMA latched onto motherboards.

Anyway, Temash is shaping up to be an awesome product that can power a performance tablet and last all day. Not really concerned about whether or not I can throw it in the closet and have it last 3 weeks. That's of no use to me.

 

[Nemesis 1]

The personnal attack was a good touch fully expected from someone like you . You attack yourself by being dishonest. Why do i say dishonest . Because no one can be that stupid . See I just complamented you . Temash looks good on paper . We just wait to see what other players have to offer , Haswell SoC/SOIX will likely walk all over temash in performance . Both Graphics and compute . Silvermont will likely = or better it in compute I don't know . But being Intel fanboy . I choose to say an unleleased products is faster than another unreleased product. Who has the upper hand in graphics below 2 watts We shall have to wait till temash gets to those levels . Silvermont scales up to 12watts and down to below 3 watts it runs in phones and is the result of years of work . 5 years to be exact . Amd is throwing things together to compete

 

[Exophase]

It is specified that compression doesnt use fully the cores
and is more influenced by bandwith and I/O than decompression
where the CPU is at 100%...

Did you think that i used the decompression numbers by chance.?..

I used the decompression numbers and they too were very favorable for the Cortex-A15 vs the Bobcat.

If it was remotely on par it would populate the notebooks
this year but you wont see this happen and it s not by
chance either....

People want to run Windows software on notebooks so an ARM SoC is a non-starter no matter how powerful it is. Of course, it's not really fair to hold a prediction of what will happen against them, when so far there has only been one Cortex-A15 SoC even available, only for a couple months, and from a company that historically sells their SoC for little outside of their own products. And has been in a netbook-class device that has sold pretty decently.

Please, show something, anything to justify your outrageous claim that Cortex-A15 should be considered at only half the perf/MHz that Jaguar does.

 

[IntelUser2000]

There are better benchmarks for one thing.

Compare the 875K to the 2600K. The latter is ~20% faster in average but in here, its only 5-6%.

I do agree with Exphase on the A15 not being inferior in IPC to Bobcat. Half the IPC of Bobcat would mean it performs less than Atom, which is not true. I assume at best, Bobcat would be ~10% faster.

 

[Abwx]

Please, show something, anything to justify your outrageous claim that Cortex-A15 should be considered at only half the perf/MHz that Jaguar does.

I somewhat exagerated but not to the point of standing corrected ,
it should not be far from that , anyway.

Those ARM CPUs are usable if you re not throwing at them too much
real computing but as soon as you start using FPUs and arithmetic
ops the picture will turn ugly.

 

[Exophase]

Please give some actual evidence to support your position :/ You sound like you're just making things up. Cortex-A15 can perform 128-bit FMADD in one cycle and can dual issue single-cycle 128-bit integer operations. I've confirmed it myself. I can't fathom what other "arithmetic ops" you may be referring to.

You shouldn't generalize all CPUs based on their architecture and vague feelings.

 

[Abwx]

I didnt find anything other than Core Mark as an eventual common
bench but there s no A15 , it is said that it is 40% faster than the
Cortex A9 wich is average with the said bench that is more about
data manipulation than actual computing.

ARM themselves only make claims without giving significant numbers
other than comparisons with their previous uarch.

Despite the improvements you re quoting they are surely
poor devices once PC like tasks are the purpose otherwise
we would have seen them popping in ultra books , yet , nothing ,
they are even the limiting factor in tablets set apart on power usage.

 

[Exophase]

Alright, so I guess it's settled, they must suck because you say so. I already gave some pretty good reasons for why the Cortex-A15 laptop market isn't taking off that don't have anything to do with its capabilities.