AT Mobile Kaveri CPU Performance Preview

[Enigmoid]

@Enigmoid - i think HD4400 is the mobile part and 4600 desktop one

@USER8000 - comparing ur HD5000 numbers with that of Kaveri, I see a max 5fps difference in specific games. not bad I would think. frame times would give a better idea

HD 4600 also exists in mobile and is what notebookcheck is using.

So you are comparing a low TDP part to full TDP Intel parts??

Emm,why don't you put the HD5000 results in there too?

Its for comparison's sake. Given how well 35W kaveri demolishes HD 4600 I expected parity.

 

[bullzz]

@AtenRa - are you comparing the chips or laptops with them. if ur talking abt laptops, I see the HP one costs lower and is comparable in specs and in most cases has better performance

@Enigmoid - ah ok. I blame Intels naming on that slip

 

[AtenRa]

@AtenRa - are you comparing the chips or laptops with them. if ur talking abt laptops, I see the HP one costs lower and is comparable in specs and in most cases has better performance

@Enigmoid - ah ok. I blame Intels naming on that slip

The Laptop with the HD5000 cost lower than the one with A10-7300 ??

About the HD4600, it is only available with 37W TDP and above SKUs.

 

[AtenRa]

I can't believe that's one of their fastest ULV 19W quad-core parts (384 cores iGPU). If those numbers actually represent mobile Kaveri's performance then AMD better make some very creative PCMark 8 marketing slides.
+1 year old 15W Haswell parts destroy it CPU-wise and hold up pretty well GPU-wise, lets see what Broadwell does.

Here's a quick CB11.5 comparison:

- Atom Z3770 (2013 Tablet chip): 0.4 / 1.47 pt
- Core i5 4200U (mid 2013 15W ''U Series'' chip): 1.15 / 2.45 pts

So according to you, PC MARK 8 that actually uses everyday applications that the majority of users are doing every day with their laptops (Web browsing, gaming, Video/photo editing etc) is AMD marketing.
But Cinebench that is heavily Intel Optimized and only professionals and some minority of users is using, is the benchmark to compare against Intel APUs

 

[Erenhardt]

- Core i5 4200U (mid 2013 15W ''U Series'' chip): 1.15 / 2.45 pts

That is some serious thermal issue there. 2.13x scaling?

 

[sm625]

Compare to this:

http://www.amazon.com/Lenovo-IdeaPad...75df25c769bbd9

That's $530. It has a mobile haswell, not an ultramobile. But you can manually throttle it to make the battery last longer so I dont see the ultramobile being that big of a selling point anyway. Also, it weights 0.4 pounds more. But its actually a heck of a lot faster since its not a ULV chip. I think the gaming performance might even be higher than the A10-7300. Due to the inherent power efficiency of intel (except while gaming), I believe this model is a much more direct competitor for the "19W" kaveri.

 

[Enigmoid]

That is some serious thermal issue there. 2.13x scaling?

Its only a dual core and its exceeding 2x scaling due to HT.

So according to you, PC MARK 8 that actually uses everyday applications that the majority of users are doing every day with their laptops (Web browsing, gaming, Video/photo editing etc) is AMD marketing.
But Cinebench that is heavily Intel Optimized and only professionals and some minority of users is using, is the benchmark to compare against Intel APUs

PCmark is difficult to directly compare due to its huge storage dependency. Pcmark8 also is openCL accelerated to an enormous extent (considering most consumers won't care about openCL). Cinebench is pretty much completely CPU bound and makes for easy comparisons. Its not the best but it works pretty well.

The Laptop with the HD5000 cost lower than the one with A10-7300 ??

About the HD4600, it is only available with 37W TDP and above SKUs.

But a lot of times a 35W CPU can easily fit in the chassis. Case in point the kaveri notebook I linked (15.6", 25 mm thick).

 

[USER8000]

How much % faster do think the A10-7300 machine would be over this?

Probably much better when doing concurrent tasks for sure!! The laptop I was using had a Intel IGP!! D:

HD 4600 also exists in mobile and is what notebookcheck is using.



Its for comparison's sake. Given how well 35W kaveri demolishes HD 4600 I expected parity.

Thats the thing - the HD5000 should be massively faster than an HD4600,but due to TDP limitations cannot always Turbo as much as it should. In some of the benchmarks the HD4400 can exceed it and the HD4600 is usually faster,and the HD5000 has double the shaders the HD4400 and HD4600 have and similar clockspeeds:

HD Graphics 4600
20 @ 200 - 1350MHz
HD Graphics 5000 40 @ 200 - 1100MHz
HD Graphics 4400
20 @ 200 - 1100MHz


Intel is using a better process node,so AMD is hitting probably worse issues when the TDP is lowered.

Edit to post.

All the parts the A10 7300 is competing with use the HD4400,ie,the Core i5 4200U and i5 4300U,so it still does better at the parts it is targeting(at least looking at local European prices).

The HD5000 is found in much more expensive notebooks and the HD4600 is found in higher TDP parts.

 

[frozentundra123456]

Kaveri ULV review

http://www.notebookcheck.net/Acer-Aspire-E5-551-T8X3-Kaveri-A10-7300-Notebook-Review.122063.0.html

GPU is nowhere near as good as expected, CPU is abyssmal. Barely faster than Kabini with CB 11.5 0.49 single, 1.74 multi.

Throttles like mad under load. Energy consumption is nothing spectacular.

GPU performance is seriously bad. Either its extremely TDP limited or the drivers need to be updated.

http://www.notebookcheck.net/Computer-Games-on-Laptop-Graphic-Cards.13849.0.html

Better than HD 4400 but not as good as HD 4600 (mobile). Was expecting the chip to at least match HD 4600.

The Ghosts and Dota 2 results must be reversed. The table is showing higher FPS for medium than low. In any case, all three are very limited for gaming. It actually does not seem to me that the Kaveri gaming results are that bad, but I would not buy a low power chip like this for gaming in any case.

 

[AtenRa]

Pcmark8 also is openCL accelerated to an enormous extent (considering most consumers won't care about openCL). Cinebench is pretty much completely CPU bound and makes for easy comparisons. Its not the best but it works pretty well.

So because it is OpenCL accelerated we dismiss it and we keep a heavily Intel optimized CPU only benchmark. We dont take in to consideration that more and more applications that the majority of users are using every day with their Laptops and PCs but we only keep a Intel Optimized benchmark that only professionals are using(small minority of users).

The PCMark 8 Home benchmark test contains the following workloads.
- Web Browsing
- Writing
- Casual Gaming
- Photo Editing
- Video Chat

The PCMark 8 Creative benchmark test contains the following workloads:
- Web Browsing
- Photo Editing
- Batch Photo Editing
- Video Editing
- Media To Go
- Mainstream Gaming
- Video Group Chat

vs

Cinebench


I dont know about you, but that seams extremely bias to me.

ps: I dont have a problem with Cinebench and you can see that im using it in my reviews, but only taking Cinebench as a metric of CPU performance is highly misleading and bias.

 

[Sweepr]

There's nothing to hide here AtenRa, soon there will be a lot more benchies to compare. Those CB11.5 scores are like Bay Trail-M (7.5W chip) CPU performance + slightly better than Core iX 15W HD4400 (but still worse than HD4600) iGPU performance in a 19W package. If that does represent A10 7300's performance (and it's not an isolated throttling issue of this particular notebook) then most of the time a Core i5 4200U/4300U notebook will have much faster CPU performance - hint: not just CB. I can easily notice the difference between a Bay Trail-M and a Core i5 notebook in my everyday computer tasks, not sure about a smaller than HD4400 -> HD4600 graphics performance gap.

 

[AtenRa]

There's nothing to hide here AtenRa, soon there will be a lot more benchies to compare. Those CB11.5 scores are like Bay Trail-M (7.5W chip) CPU performance + slightly better than Core iX 15W HD4400 (but still worse than HD4600) iGPU performance in a 19W package. If that does represent A10 7300's performance (and it's not an isolated throttling issue of this particular notebook) then most of the time a Core i5 4200U/4300U notebook will have much faster CPU performance - hint: not just CB. I can easily notice the difference between a Bay Trail-M and a Core i5 notebook in my everyday computer tasks, not sure about a smaller than HD440 -> HD4600 graphics performance gap.

Cinebench CPU performance is not indicative of average CPU performance. Why dont you use Pov-Ray or x264 or any other application and always and specifically address CPU performance by using a single benchmark like Cinebench??

If i post this graph that shows 45W TDP A10-7600 Kaveri being 6% faster than 54W TDP Core i3 4330 and say that Intel better come up with a Cinebench Marketing slide, everyone will call me fanboy, bias etc etc.

I believe you see the picture now, dont you ??

http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/11

 

[USER8000]

Who the heck runs CB let alone intensive video editing or any of those on a cheap laptop anyway?? You are talking £400 to £450 laptops. I know plenty of geeky people(and plenty of non-geeks) and the only ones who would be running anything remotely intensive would be doing so on much more expensive laptops anyway not budget ones.

People are just going to be browsing the web,doing some light image(not serious image editing) editing,office software and maybe a bit of gaming on these kind of laptops.

None of these(outside the gaming part) are CPU intensive. Plus the screens are not that great anyway,which would not help with video and image tasks.

Like I said I used a Pentium M based laptop to do quite a bit of work,and Core2 based MacBook Airs too.

The A10s or the ULV Core i3 CPUs,are comfortably faster than any of these,especially with 4 threads(which makes them far better for multi-tasking) - if anything for most tasks the speed of the OS drive and the amount of RAM are going to be more important considerations.

 

[Sweepr]

Cinebench CPU performance is not indicative of average CPU performance. Why dont you use Pov-Ray or x264 or any other application and always and specifically address CPU performance by using a single benchmark like Cinebench??

Maybe because they're not available in NotebookCheck's article.

I believe you see the picture now, dont you ??

I do. Everything but OpenCL accelerated apps likely won't make it into AMD's marketing slides. That's fine though, they're just showing their design's strenghts.

 

[Enigmoid]

Thats the thing - the HD5000 should be massively faster than an HD4600,but due to TDP limitations cannot always Turbo as much as it should. In some of the benchmarks the HD4400 can exceed it and the HD4600 is usually faster,and the HD5000 has double the shaders the HD4400 and HD4600 have and similar clockspeeds:

HD Graphics 4600
20 @ 200 - 1350MHz
HD Graphics 5000 40 @ 200 - 1100MHz
HD Graphics 4400
20 @ 200 - 1100MHz

Intel is using a better process node,so AMD is hitting probably worse issues when the TDP is lowered.

Edit to post.

All the parts the A10 7300 is competing with use the HD4400,ie,the Core i5 4200U and i5 4300U,so it still does better at the parts it is targeting(at least looking at local European prices).

The HD5000 is found in much more expensive notebooks and the HD4600 is found in higher TDP parts.

Yeah it should be competing with lower tdp parts but really notebooks also compete with regards to chassis. The chassis this CPU is in (15.6" 25 mm) could easily hold a 35W i5. To the consumer looking for a 15.6" laptop all other 15.6" laptops are fair game. Consumers don't usually go: "I want ULV in a 15.6" notebook", they want the best bang for their buck in a given size/weight/performance class; all notebooks similar in size are competition, just as when looking at 15.6" notebooks (screen size preference) you will come across models with ULV, SV and Quad intel processors. All those CPUs, despite the TDP difference are in direct competition with each other.

By placing a ULV CPU in a certain chassis you invite competition against other notebooks in the same chassis. Hell there is competition against the 35W kaveri in a similar 15.6" chassis despite any TDP differences.

So because it is OpenCL accelerated we dismiss it and we keep a heavily Intel optimized CPU only benchmark. We dont take in to consideration that more and more applications that the majority of users are using every day with their Laptops and PCs but we only keep a Intel Optimized benchmark that only professionals are using(small minority of users).

The PCMark 8 Home benchmark test contains the following workloads.
- Web Browsing
- Writing
- Casual Gaming
- Photo Editing
- Video Chat

The PCMark 8 Creative benchmark test contains the following workloads:
- Web Browsing
- Photo Editing
- Batch Photo Editing
- Video Editing
- Media To Go
- Mainstream Gaming
- Video Group Chat

vs

Cinebench
I dont know about you, but that seams extremely bias to me.
ps: I dont have a problem with Cinebench and you can see that im using it in my reviews, but only taking Cinebench as a metric of CPU performance is highly misleading and bias.

I agree. But you notice something. PCmark is not a pure CPU benchmark. There is a large GPU component (gaming). Thus it is not a good benchmark for pure CPU performance. There is also a large storage dependency. It is a system benchmark, not a CPU benchmark.

As far as CB goes

Using Hardware.fr (http://www.hardware.fr/articles/913-7/cpu-performances-applicatives.html)


The i5 4670k is 60% faster using this amalgamation of tests. Using values from AT bench the i5 gets 6.2 in CB 11.5 while kaveri gets 3.6, an increase of 72%. It appears that CB favours intel, however that could just as likely be caused by the FP heavy calculations and the fact that the a10-7850k has only two FPUs. Other FPU heavy calculations show a strong bias toward Haswell.

 

[AtenRa]

Maybe because they're not available in NotebookCheck's article. :thumbsup:

wrong, there are more CPU benchmarks like x264 or Geekbench etc

http://forums.anandtech.com/showpost.php?p=36549515&postcount=67

It is funny that nobody talks about those CPU benchmarks but everyone takes Cinebench as CPU performance gospel.

 

[USER8000]

Yeah it should be competing with lower tdp parts but really notebooks also compete with regards to chassis. The chassis this CPU is in (15.6" 25 mm) could easily hold a 35W i5. To the consumer looking for a 15.6" laptop all other 15.6" laptops are fair game. Consumers don't usually go: "I want ULV in a 15.6" notebook", they want the best bang for their buck in a given size/weight/performance class; all notebooks similar in size are competition, just as when looking at 15.6" notebooks (screen size preference) you will come across models with ULV, SV and Quad intel processors. All those CPUs, despite the TDP difference are in direct competition with each other.

By placing a ULV CPU in a certain chassis you invite competition against other notebooks in the same chassis. Hell there is competition against the 35W kaveri in a similar 15.6" chassis despite any TDP differences.

I don't know how it is in the US,but here in the UK upto around the £550 mark,the vast majority of Intel based laptops have either Bay Trail or Core i3 or Core i5 ULV CPUs,and that includes one of the biggest retailers here with nearly 900 stores around Northern Europe. Most are not thin and light laptops either and have 15.4" or 15.6" screens. Most of the AMD APU laptops(the Trinity and Richland A10 and A8 based ones) are under £550 but have the higher TDP SKUs like the A10 5750M.

Edit to post.

I think the current Richland and Trinity A10 and A8 based laptops might be the bigger competition over here.

 

[AtenRa]

I agree. But you notice something. PCmark is not a pure CPU benchmark. There is a large GPU component (gaming). Thus it is not a good benchmark for pure CPU performance. There is also a large storage dependency. It is a system benchmark, not a CPU benchmark.

As far as CB goes

Using Hardware.fr (http://www.hardware.fr/articles/913-7/cpu-performances-applicatives.html)


The i5 4670k is 60% faster using this amalgamation of tests. Using values from AT bench the i5 gets 6.2 in CB 11.5 while kaveri gets 3.6, an increase of 72%. It appears that CB favours intel, however that could just as likely be caused by the FP heavy calculations and the fact that the a10-7850k has only two FPUs. Other FPU heavy calculations show a strong bias toward Haswell.

First off all, PC MARK 8 is not storage depended like PC MARK 7 is.
Secondly, i never said that PC MARK 8 is a CPU benchmark. I said we shouldn't dismiss it because of the OpenCL acceleration.
Thirdly, if Cinebench is so Intel Optimized why do we still use it as a base for CPU performance and not use a more neutral benchmark like x264 or Pov-Ray or anything else. ??

As i have said before, i dont have anything against Cinebench, i use it as well. It is very usefull to compare Intel CPUs against each other or AMD CPUs against each other. But it is completely misleading to compare Intel vs AMD CPUs using Cinebench alone. As it would be completely misleading to compare AMD vs Intel only on OpenCL.

 

[Enigmoid]

I don't know how it is in the US,but here in the UK upto around the £550 mark,the vast majority of Intel based laptops have either Bay Trail or Core i3 or Core i5 ULV CPUs,and that includes one of the biggest retailers here with nearly 900 stores around Northern Europe. Most are not thin and light laptops either and have 15.4" or 15.6" screens. Most of the AMD APU laptops(the Trinity and Richland A10 and A8 based ones) are under £550 but have the higher TDP SKUs like the A10 5750M.

Edit to post.

I think the current Richland and Trinity A10 and A8 based laptops might be the bigger competition over here.

Yes. I'm saying that these chips compete primarily on price, not TDP. If you are getting 35W notebooks for the same price and similar features then those 35W chips are direct competition, regardless of TDP. As far as the US, newegg has the cheapest quad at $600 (http://www.newegg.com/Product/Produc...9SIA1J21UP0115). The is little targeting a specfic tdp, rather a price and form factor is targeted.

First off all, PC MARK 8 is not storage depended like PC MARK 7 is.
Secondly, i never said that PC MARK 8 is a CPU benchmark. I said we shouldn't dismiss it because of the OpenCL acceleration.
Thirdly, if Cinebench is so Intel Optimized why do we still use it as a base for CPU performance and not use a more neutral benchmark like x264 or Pov-Ray or anything else. ??

As i have said before, i dont have anything against Cinebench, i use it as well. It is very usefull to compare Intel CPUs against each other or AMD CPUs against each other. But it is completely misleading to compare Intel vs AMD CPUs using Cinebench alone. As it would be completely misleading to compare AMD vs Intel only on OpenCL.

Yes but CB being a purely CPU test has more merit (theoretically) than a mixed CPU + GPU test. I'm not saying CB is perfect but you have to eliminate variables.

(I didn't realize that they factored the storage test separately, I stand corrected).

 

[AtenRa]

Yes. I'm saying that these chips compete primarily on price, not TDP. If you are getting 35W notebooks for the same price and similar features then those 35W chips are direct competition, regardless of TDP. As far as the US, newegg has the cheapest quad at $600 (http://www.newegg.com/Product/Produc...9SIA1J21UP0115). The is little targeting a specfic tdp, rather a price and form factor is targeted.

Just because you have two products at the same price doesnt mean they are targeting the same buyer.

Yes but CB being a purely CPU test has more merit (theoretically) than a mixed CPU + GPU test. I'm not saying CB is perfect but you have to eliminate variables.

(I didn't realize that they factored the storage test separately, I stand corrected).

But i have never said to replace CB with PC MARK 8, i said we need more than one benchmark to evaluate CPU performance. I have already mentioned two neutral Benchmarks like Pov-Ray and x264 and im sure there are more. But some people here only use CB which is Intel Optimized to evaluate CPU performance and at the same time they want to dismiss PC Mark 8 because of its OpenCL acceleration.

Edit: Not to mention that both AMD and Intel among others are PC MARK 8 program members, unlike CineBench that is only optimized for Intel CPUs.

http://www.futuremark.com/pressreleases/announcing-pcmark-8

PCMark 8 has been developed in partnership with the members of our Benchmark Development Program. PCMark BDP members include Acer, AMD, Condusiv Technologies, Dell, HGST, HP, Intel, Microsoft, NVIDIA, Samsung, SanDisk, Seagate and Western Digital.

 

[lefty2]

I agree with AtenRa. The benchmark choosen in that article shows the A10-7300 in a very poor light. The Cinebench R11.5: CPU Multi 64Bit score is only the same level as a Pentium N3520, compare that to Octane V2: A10-7300 scores 3 times higher than Pentium N3520!
You can compare the two processors in detail here:
http://www.notebookcheck.net/Intel-Pentium-N3520-Notebook-Processor.108993.0.html
http://www.notebookcheck.net/AMD-Kaveri-A10-7300-Notebook-Processor.117333.0.html

 

[Enigmoid]

I agree with AtenRa. The benchmark choosen in that article shows the A10-7300 in a very poor light. The Cinebench R11.5: CPU Multi 64Bit score is only the same level as a Pentium N3520, compare that to Octane V2: A10-7300 scores 3 times higher than Pentium N3520!
You can compare the two processors in detail here:
http://www.notebookcheck.net/Intel-Pentium-N3520-Notebook-Processor.108993.0.html
http://www.notebookcheck.net/AMD-Kaveri-A10-7300-Notebook-Processor.117333.0.html

You really really have to be careful with notebookcheck scores and eliminate obvious mistakes. Kraken shows the i7 4600U as the lead performer, well over the 2600k. This is so obviously wrong (different browsers).

Also Octane is quite short and doesnt stress the CPU much. CB is a much heaver load for the TDP limited 7300.

Avoid browser benchmarks like the plague.

Better would be x264 or winrar.

 

[lefty2]

You really really have to be careful with notebookcheck scores and eliminate obvious mistakes. Kraken shows the i7 4600U as the lead performer, well over the 2600k. This is so obviously wrong (different browsers).

Also Octane is quite short and doesnt stress the CPU much. CB is a much heaver load for the TDP limited 7300.

Avoid browser benchmarks like the plague.

Better would be x264 or winrar.

Ok, maybe Octane was a bad example. Skimming through the benchmarks there, I see most of them are showing the 7300 beating the N3520 by around 30%

 

[jdubs03]

From what I have seen notebookcheck does a real good job in determining CPU performance aggregation, but when devices are compared against one another it is harder to get a clear picture.

For instance in the Acer Aspire review with the A10-7300 it has PCMark 7 comparing a i5-4200U only scoring 2533, but on Anandtech all the PCMark 7 scores are over 4000. And in the Surface Pro 3 review, they don't even include PCMark 8, but they did with their latest review of the HP Elitebook Revolve (in which the Acer A10 score higher in home, and slightly lower in creative, yet the PCMark 7 was only 4131, but SP3 is 5076 with the same processor - i5-4300U). And from Anandtech the core i5 SP3's score is lower in PCMark 8 then the A10 in both home and work.

Very contradictory. And my question is why is there such a large difference between PCMark 7 and 8 results?, where clearly PCMark 7 favors the i5-4300U by almost triple (in the case of SP3) over the A10-7300.

 

[Sweepr]

Very contradictory. And my question is why is there such a large difference between PCMark 7 and 8 results?, where clearly PCMark 7 favors the i5-4300U by almost triple (in the case of SP3) over the A10-7300.

PCMark 8 uses OpenCL acceleration, hence why it favours faster IGPs and why its a constant in AMD's most recent comparison marketing slides.