Will Bobcat be the home run AMD is looking for?

[cbn]

Simple poll. (be aware that votes are not public).

When I say "home run", I mean a chip that is more desirable than the Intel equivalent (ie, atom) even if it is made on larger process. (eg, 90nm Athlon 64 vs. 65nm Netburst)

EDIT: Here is a new Bobcat update article from Techreport. (courtesy of wlee15 in post #69 of this thread).

Here is Part 1 of the Anandtech Brazos preview article.

EDIT 2:
(Benchmarks are up. Before these were posted, the bobcat poll was showing 75 votes "Yes" and 59 votes "No".)
http://www.anandtech.com/show/4023/the-brazos-performance-preview-amd-e350-benchmarked
http://techreport.com/articles.x/19981
http://hothardware.com/Reviews/AMD-Zacate-E350-Processor-Performance-Preview/
http://www.legitreviews.com/article/1470/1/
http://www.pcper.com/article.php?aid=1039


EDIT 3:
http://www.tomshardware.com/reviews/amd-fusion-brazos-performance,2790.html

 

[StrangerGuy]

http://www.chip-architect.com/news/AMD_Ontario_Bobcat_vs_Intel_Pineview_Atom.jpg

So we have a 15.2 mm^2 40nm CPU that is 90% as fast a same clocked C2D @ 45nm 107 mm^2? And that is not even including the 80 SP GPU...*jaws drop*. This is far and away the most impressive product from AMD since K7.

AMD can slap 6 Bobcat cores into one die and still come out smaller than a Wolfdale.

 

[ModestGamer]

http://www.chip-architect.com/news/AMD_Ontario_Bobcat_vs_Intel_Pineview_Atom.jpg

So we have a 15.2 mm^2 40nm CPU that is 90% as fast a same clocked C2D @ 45nm 107 mm^2? And that is not even including the 80 SP GPU...*jaws drop*. This is far and away the most impressive product from AMD since K7.

AMD can slap 6 Bobcat cores into one die and still come out smaller than a Wolfdale.

kind of makes you wonder how good bulldozer will be eh ?

 

[KingstonU]

I can't help but think that if there was such a linear increase in performance to how many bobcat cores are put together AMD would have figured it out and we would have heard some whispers of it.

In other words, can someone with more education on it speculate on if it's as easy as 8 bobcat cores = 4x performance as 2 bobcat cores?

Fingers are crossed though for the sake of the industry and us consumers.

 

[nyker96]

from most pre-reports, I get the sense that BD cannot beat intel offering in per core performance, amd design team just doesn't have the resource to pull it off. So they are doing the opposite by upping threaded performances. I don't know if this is a home run or not depend on software, but it would be an interesting alternative approach to intel cpu design. It could sell very well, if its overall performance/power/price/OC potential is on target. If that's the case, I'd consider it a successful product even if it's not the fastest product on the market.

 

[ModestGamer]

from most pre-reports, I get the sense that BD cannot beat intel offering in per core performance, amd design team just doesn't have the resource to pull it off. So they are doing the opposite by upping threaded performances. I don't know if this is a home run or not depend on software, but it would be an interesting alternative approach to intel cpu design. It could sell very well, if its overall performance/power/price/OC potential is on target. If that's the case, I'd consider it a successful product even if it's not the fastest product on the market.

I don't get that sense at all.we will have to wiat for the testing to commence in real world applications to see how it really measures up, being AMD is so tight lipped its tough to determine anything.

better to wait then judge. Its not always about fab budgets and engineering resources, all it takes is a few good ideas to really improve something alot.

 

[sandorski]

Being the Under Dog, AMD has every reason to keep any kind of Pwnt Chip under wraps. They can't really afford to go toe-toe with Intel on a constant basis. So the better strategy is to develop something that catches Intel off guard, then ride the wave for a few generations until Intel counters.

 

[ModestGamer]

Being the Under Dog, AMD has every reason to keep any kind of Pwnt Chip under wraps. They can't really afford to go toe-toe with Intel on a constant basis. So the better strategy is to develop something that catches Intel off guard, then ride the wave for a few generations until Intel counters.

I think this is the mor elikely scenario, in fact Intels giant marketing efforts lately most likely indicate how badass the comming AMD products are. Intel is now feeling pressure from ARM and a few other market items. Imagine if arm takes root in netbooks and notebooks. They do have those new 2ghz dual core arm cpu's.

I can only imagine intel leadership is more then a bit nervous and it explains all the bluster.

 

[cbn]

Intel is now feeling pressure from ARM and a few other market items. Imagine if arm takes root in netbooks and notebooks. They do have those new 2ghz dual core arm cpu's.

What you are saying is very interesting.

Furthermore, I have been wondering how Cloud computing could integrate into buyer habits?

For example, What offers lower total cost of ownership, versatility and/or battery life:

1. Lightweight computer (eg, AMD Bobcat, Cortex A15 or Intel) + Cloud services (for video editing, etc)

vs.

2. Dedicated Quad+ core laptop that does all the heavy lifting directly within the machine.

Advantages and Disadvantages to each?

 

[JFAMD]

from most pre-reports, I get the sense that BD cannot beat intel offering in per core performance, amd design team just doesn't have the resource to pull it off. So they are doing the opposite by upping threaded performances. I don't know if this is a home run or not depend on software, but it would be an interesting alternative approach to intel cpu design. It could sell very well, if its overall performance/power/price/OC potential is on target. If that's the case, I'd consider it a successful product even if it's not the fastest product on the market.

First off, I would wait until all products are out until anyone makes any performance claims.

But, more importantly, per core performance is becoming less and less critical. 5 years ago all of the sofftware was written for a single executed thread. Then it was expecting two threads. That moved to 4 and 6. Next year it will be 8. As software developers create their applications they are thinking less and less about single core and now utilizing multiple cores to achieve better performance.

So as you look at per core performance, that is far less important ass time goes on. the computers that people buy in 2011 will last most through 2014, so if you are developing applications today, you are thinking about core counts over the next few years.

Think about the music industry. They have gone out of their way to keep people on physical CD's when the public has spoken and said electronic downloads are the format that they want. Every year there is less demand for physical CDs and more demand for downloads. As the music business do you try to add things to the CDs to make them more desireable or do you shift your weight towards making downloads better?

Single core performance is a dying trend. If you don't believe me, look at intel. Are they working on a 10GHz single core for the future or are they trying to shop a 48-core design?

 

[jaydee]

Great post JFAMD. You know a lot, but you can also bring the technical level down to us non-computer engineers. I was under the same impression (as nyker96), that AMD is having trouble with per core/clock performance. Fortuanately for them however, the industry is moving towards more threads and how well that threading is implemented.

Given how long Bulldozer has been in development with this in mind, I'm interested to see how this plays out because the full benefit of cores/modules will probably (surely?) not be seen right away. Perhaps that's one reason why Bulldozer has been delayed like it has? 2009-2010 software has not caught up to see the benefits that BD offers, and a 2009 launch would have made AMD look really bad? Or put more mildly, took the rush off of making it happen in a certain quarter so they can further improve performance while waiting for the software to catch up? Just a guess.

With this in mind, I see Bobcat being the more important launch of the two in the short-term. I think OEM Mini-ITX is about to grow big-time in the next two years (with power bricks instead of traditional PSU's). CPU + GPU price/performance/power are about to hit the sweet spot along with the SSD's hitting their stride. Nettops will become extremely practical on all accounts.

 

[cbn]

But, more importantly, per core performance is becoming less and less critical. 5 years ago all of the sofftware was written for a single executed thread. Then it was expecting two threads. That moved to 4 and 6. Next year it will be 8. As software developers create their applications they are thinking less and less about single core and now utilizing multiple cores to achieve better performance.

So as you look at per core performance, that is far less important ass time goes on. the computers that people buy in 2011 will last most through 2014, so if you are developing applications today, you are thinking about core counts over the next few years.

A few questions:

1. I thought multiplying cores (for applications outside of a data center) resulted in diminishing returns? At some point wouldn't it start to make sense to look into boosting the performance of each core again?

2. What about Open CL for the few desktop applications that happen to be highly parrallel?

3. Or does nothing I am saying really matter because Cloud computing is the future and all that is really needed for the majority is bobcat for client and "many core" CPUs for the servers doing the heavy lifting?

 

[ModestGamer]

A few questions:

1. I thought multiplying cores (for applications outside of a data center) resulted in diminishing returns? At some point wouldn't it start to make sense to look into boosting the performance of each core again?

well that depends. If you can run a process in each core you get some good performance gains. multithreading doesn;t have diminishg returns so to speak, it just comes with a thread managaing overhead. Erlang looks to remove alot of the overhead however. the way it handles threads removes alot of the overhead in doing so.

2. What about Open CL for the few desktop applications that happen to be highly parrallel?

3. Or does nothing I am saying really matter because Cloud computing is the future and all that is really needed for the majority is bobcat for client and "many core" CPUs for the servers doing the heavy lifting?

I personally don't think cloud computing is the future but multicore fits in well in the cloud model for sure.

 

[OBLAMA2009]

NO, and nobody thinks it will, not even amd. it will underperform intel chips but will be attractive based on price and core count, as usual

 

[ModestGamer]

NO, and nobody thinks it will, not even amd. it will underperform intel chips but will be attractive based on price and core count, as usual

ahh I am not seeing AMD indicating this anywhere. so where is your info comming from ? intel marketing ?

 

[adlep]

Please AMD make it Socket AM3 compatible.

 

[cbn]

If you can run a process in each core you get some good performance gains. multithreading doesn;t have diminishg returns so to speak

http://en.wikipedia.org/wiki/File:AmdahlsLaw.svg

I was referring to this graph.

If speedup reaches diminishing returns wouldn't a point be reached where electricity was better spent making less cores go faster?

 

[ModestGamer]

http://en.wikipedia.org/wiki/File:AmdahlsLaw.svg

I was referring to this graph.

If speedup reaches diminishing returns wouldn't a point be reached where electricity was better spent making less cores go faster?

well it depends on how you use the cores that matters. threading may have dimishing returns in a application by application or task basis, overall though more cores is going to equal more power. the issue is creating software that extracts this performance efficiently.

its hard to say but I don't subscribe to the notion of Amdahls law. it is to a large degree based on todays programming models which aren't really the most efficient and alot of software is still heavily serialized not parrelled.

 

[quest55720]

I don't know how the subject got changed but I think bobcat could be a home run if AMD can market it right. It basically will destroy the atom in the netbook market. Be competitive enough in the low-mid range notebook to give Intel fits. I personally can't wait for a bobcat netbook to replace my current under powered netbook. I would love to see a windows 7 based tablet based on the bobcat.

 

[JFAMD]

There is definitely diminishing returns on the core count. 2 cores are significantly faster than 1 core. And 4 cores are faster than 2, but not by the same measure. Where does the curve break? Depends on the software.

Most models will tell you that diminishing returns really happen at ~10-12 cores on the desktop side and probably ~24-32 cores on the server side.

What I mean by diminishing returns is where the bulk of the customers stop seeing ROI on adding more cores.

This is of course dependent on a lot of variables including the pace of software development and the application patterns of the users.

The key is that software keeps continuing to become more threaded and people continue to run more concurrent applications because they have the resources.

Remember when you used to actually open a web browser? Well, most people keep these open all day long. I know at work I typically have outlook, excel, powerpoint and a browser open at all times. Plus antivirus and 20 other processes that the corporate IT has running on my behalf.

We live in a multithreaded world and will continue to live there. the biggest challenge may well be the core to memory ratio or bandwidth more than the core counts.

As for openCL, this really only helps you in a parallelized workload. Cores help you in much more random workloads.

 

[firewolfsm]

A few questions:

1. I thought multiplying cores (for applications outside of a data center) resulted in diminishing returns? At some point wouldn't it start to make sense to look into boosting the performance of each core again?

Honestly, I don't see the trend to increase core count ever reversing as long as we have cores in our processors. We're always trying to design processors more like our own brains, which are massively parallel.

Essentially, as these diminishing returns hit, we can run the same process three (or more) times, but with lower accuracy, such that, say, 66% accuracy is achieved. This would boost the speed of individual calculations exponentially. Error is eliminated through redundancy, and the correct calculation is pushed forward (to a negligible degree of inaccuracy). Then we can spend another ten or twenty years improving the population of our calculations and the accuracy of each calculation per watt.

We will have a similar method in place by the time we hit hundreds or thousands of cores, at this rate, in about 20 years.

You heard it from me first.

 

[cbn]

I don't know how the subject got changed but I think bobcat could be a home run if AMD can market it right. It basically will destroy the atom in the netbook market. Be competitive enough in the low-mid range notebook to give Intel fits. I personally can't wait for a bobcat netbook to replace my current under powered netbook. I would love to see a windows 7 based tablet based on the bobcat.

Yep, I am really interested to see if Bobcat (more accurately Ontario/Zacate) can reverse trends in the netbook market.

http://forums.anandtech.com/showthread.php?t=2118063

Lots of people claim atom is just not sufficient, maybe Ontario (with 2 to 4 GB of soldered in memory) will finally get the netbook segment over the hump into the range of "good enough"?

 

[Dark_Archonis]

First off, I would wait until all products are out until anyone makes any performance claims.

But, more importantly, per core performance is becoming less and less critical. 5 years ago all of the sofftware was written for a single executed thread. Then it was expecting two threads. That moved to 4 and 6. Next year it will be 8. As software developers create their applications they are thinking less and less about single core and now utilizing multiple cores to achieve better performance.

So as you look at per core performance, that is far less important ass time goes on. the computers that people buy in 2011 will last most through 2014, so if you are developing applications today, you are thinking about core counts over the next few years.

Think about the music industry. They have gone out of their way to keep people on physical CD's when the public has spoken and said electronic downloads are the format that they want. Every year there is less demand for physical CDs and more demand for downloads. As the music business do you try to add things to the CDs to make them more desireable or do you shift your weight towards making downloads better?

Single core performance is a dying trend. If you don't believe me, look at intel. Are they working on a 10GHz single core for the future or are they trying to shop a 48-core design?

Sounds like a lot of marketing talk and explanation for AMD's Bulldozer strategy, to be quite honest.

Per-core performance matters less and less? Really? Well it doesn't matter how many cores you have if each individual core only offers average performance.

Why should there be a compromise? So as software becomes more and more multi-threaded, performance per core no longer needs to be improved?

The big problem with this approach, AMD's approach, is that the approach the main competition (Intel) is going for is very strong per-core performance combined with multiple cores. Intel is not compromising with their approach. Their approach will offer the best of both worlds.

Almost all software out there still benefits from better per-core performance. Even highly multi-threaded software benefits from better per-core performance.

Furthermore, a lot of software today that is multi-threaded is poorly coded and not very optimized.

It will still take several years at the very least for efficient multi-threaded software to become common.

Additionally, there is a limit to how many cores you can keep adding before you start seeing very little return.

The other problem with this approach is that it relies on a big assumption that software will become much better in terms of efficient multi-threaded coding.

Intel's approach is to offer a performance increase for current and past software as well as future efficiently multi-threaded software.

Single core performance is a dying trend. If you don't believe me, look at intel. Are they working on a 10GHz single core for the future or are they trying to shop a 48-core design?

I don't believe you quite frankly, and my response to that first off is that it depends who you ask.

Secondly, you're right, let's look at Intel. Specifically, look at the changes that the Sandy Bridge architecture will bring forth. Top Sandy Bridge CPUs will have turbo modes close to 4Ghz, quite a bit higher than the turbo modes of top Intel i7 CPUs today. Furthermore, core counts will generally increase *and* per-core performance will increase quite noticeably.

Going back to your question, no Intel is not working on a 10Ghz single core. However, is Intel working on a 2Ghz 16-core or 32-core design? Not really, no. Intel is combining the best of both worlds. They are increasing per-core performance AND increasing clock speeds AND increasing the core counts.

Ultimately Intel's approach gives a person supreme flexibility. Past and current software ends up running faster, while you also have the capability of running future software faster as well.

 

[Medikit]

I'm fairly confident that everyone will want a Bobcat in their laptop for the next 2 years.

 

[veri745]

Sounds like ...snip... faster as well.

Well thank you both for derailing the topic entirely. This has nothing to do with bobcat / ontario.

On topic:

I think ontario has the potential to put netbooks in a highly desirable place in the price/performance spectrum of the laptop market. low-end notebooks have always suffered from horrendous graphics and "not good enough" performance for many users, and this could change with ontario.