AMD vs Intel Performance With OpenCL Realtime Face Animation

[rootheday3]

regarding AtenRA's post...

First, the comparison there is versus HD4000 (Ivybridge), not HD4600 or other Haswell generation Intel gpus.

Second, even if Musemage & Photozoom are faster on an A1-6800K, cherrypicking benchmarks proves nothing - especially for OpenCL. Look at this article and see how much things shift from test to test in terms of AMD A8-6800K 8670D vs HD 4600 (Core i7 GPU). OpenCL workloads are highly sensitive to differences in vector width, register file size, cost of atomics and shared memory accesses. An algorithm that is tuned for one gpu will typically not do well on a different one. As such, to evaluate OpenCL performance generically, you have to look across a lot of benchmarks or you have to know that the single benchmark you've chosen exactly reflects your intended usage model/application.

Third, the OC'd results are shown only for the AMD gpu. I've seen a lot of forum threads with Intel GPUs OC'd ~50% to 1.55-1.65GHz+ and getting excellent scaling.

 

[Enigmoid]

20% is a lot of performance difference.

I very much doubt that Intel Core i5 4250U (HD5000) 15W TDP will be faster than AMD A8-5545M (384 cores) 19W TDP. Yes i know they are not the same TDP but you also will have to take cost in to consideration, Battery capacity etc.

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested
Also, if A10-4600M Trinity at 35W is within a few frames distance against a desktop HD4600 at 87W on the Core i7 4770K in the majority of the games, then Intel HD5100 at 28W will be slower than Ritchland LV A10-5745M at 35W which is faster than A10-4600M.

Just because Intel has the Iris Pro (HD5200) doesnt mean it is faster than Ritchland APUs in iGPU department at comparable TDP, Power consumption and cost.

I will really like to see a review of Laptops with AMD A8-5545M (19W) vs Intel HD5000 (15W) or A10-5745M vs HD5100 (28W) and HD4600 (37W).

20% difference in mobile products is common where CPUs may boost differently depending on thermals, and GPU have different clock speeds and vram type. Yes its kinda confusing but I feel intel's branding solution is better than AMD's where you have about 5 billion names, different between desktop and mobile (at least they got rid of the llano hybrid crossfire naming scheme).

The haswell CPU will be much faster. About twice as fast CPU wise. Here are three reviews of the a8-4555m (trinity).

http://www.notebookcheck.net/Review-HP-Pavilion-Sleekbook-15z-b000-Notebook.88783.0.html
http://www.notebookcheck.net/Review-Lenovo-IdeaPad-S405-Notebook.86795.0.html
http://www.notebookcheck.net/Review-Toshiba-Satellite-S955D-S5150-Notebook.94413.0.html

CPU sucks 1.3-1.4 points in cinebench multi.

The satellite is unfortunately the only one with dual channel RAM. It honestly looks slightly better than the HD 4000 but as soon as the CPU gets a heavy load too performance generally falls. And between a 15 watt haswell and a 19 watt richland + PCH the haswell is going to use significantly less energy.

We haven't seen the performance of the HD 5100 yet.

Do you actually believe that a Mobile 37W Intel HD4600 will be faster than A10-4600M when the Desktop 84W HD4600 is only a few fps faster ???


Edit: just a few examples. And just to remind you all, Desktop Haswell HD4600 runs on 2400MHz Memory in the Review. In Laptops it will be at 1600MHz maximum and lots of the time single slot.
[remover images]

Not faster no but within margin of error on notebooks (within 10%). 37 and 47 watt HD 4600 and going to have basically the same performance to within 5% (same clocks, lots of tdp headroom, only difference is cache).

http://www.notebookcheck.net/Review-Intel-HD-Graphics-4600.93188.0.html

Assume the typical HD 4600 is somewhere between the HD 4000 and the HD 4600 used there (which has high clocks and lots of cache). They are quite similar.

Lets look at the last couple games from notebookcheck (who actually reviews integrated graphics regularly).

Metro LL-Basically equal, not sure on hd 4000 discrepancy.

Grid 2 - AMD advantage

Dota 2- basically equal
http://www.notebookcheck.com/Dota-2-Benchmarks.99462.0.html

Company of heroes 2-Equal

SplinterCell Blacklist- Missing some settings and looks CPU bottlenecked

The Bureau

Saints Row-

(note the 750m is using ddr3 which is why its behind the 660m). Sadly there is little difference between the HD 4000 and the HD 4600.

 

[Abwx]

So we have to believe your own benchmarks? From a person that even signed up for AMDs PR program with the task of filling forums up with this kind of nonsense. Your credibility is zero.

http://forum.oktabit.gr/topic/amd-a10-6800k-overclocked-part-1-desktop-applications

?

Ad Hominem attack as an argument..?

And your own credibility is better with such agressions ?..

He has shown numbers while you re relying on deffamation
to downplay his findings , hardly what one would call an
unbiaised critic...

 

[Abwx]

Sadly there is little difference between the HD 4000 and the HD 4600.

The Trinity 4600M is strictly limited to 35W TDP while
the IB in your graphs is a 45W that can go up to 55W
on short times , the HW is a 47W that burst to 60W levels ,
that is , you re comparing a 35W TDP to litteraly desktop parts
and then claim that Intel is not that far from AMD GPU wise.

Indeed , if you take parts that have basicaly 30% higher TDPs
that can be pushed to 60% higher TDP than Trinity , but then
give the latter the same TDP envelloppe and we ll see if your
conclusions will be stretched the same way...

 

[AtenRa]

20% difference in mobile products is common where CPUs may boost differently depending on thermals, and GPU have different clock speeds and vram type. Yes its kinda confusing but I feel intel's branding solution is better than AMD's where you have about 5 billion names, different between desktop and mobile (at least they got rid of the llano hybrid crossfire naming scheme).

The haswell CPU will be much faster. About twice as fast CPU wise. Here are three reviews of the a8-4555m (trinity).

http://www.notebookcheck.net/Review-HP-Pavilion-Sleekbook-15z-b000-Notebook.88783.0.html
http://www.notebookcheck.net/Review-Lenovo-IdeaPad-S405-Notebook.86795.0.html
http://www.notebookcheck.net/Review-Toshiba-Satellite-S955D-S5150-Notebook.94413.0.html

CPU sucks 1.3-1.4 points in cinebench multi.

The satellite is unfortunately the only one with dual channel RAM. It honestly looks slightly better than the HD 4000 but as soon as the CPU gets a heavy load too performance generally falls. And between a 15 watt haswell and a 19 watt richland + PCH the haswell is going to use significantly less energy.

We haven't seen the performance of the HD 5100 yet.

I was talking about the A8-5545m Ritchland, what you post is about Trinity A8-4555m.

Not faster no but within margin of error on notebooks (within 10%). 37 and 47 watt HD 4600 and going to have basically the same performance to within 5% (same clocks, lots of tdp headroom, only difference is cache).

http://www.notebookcheck.net/Review-Intel-HD-Graphics-4600.93188.0.html

Assume the typical HD 4600 is somewhere between the HD 4000 and the HD 4600 used there (which has high clocks and lots of cache). They are quite similar.

You have to be kidding me. You are using a 55W Core i7 4930MX for the Mobile HD4600 to show that it is very close to a 35W A10-4600M ??
And not only that, even that Intel 55W HD4600 losses at 1366x768 in many games to the last year 35W Trinity A10-4600M. Just image how bud a 37W Intel HD4600 will perform against 35W Ritchland A10-5750M that also supports 1866MHz memory.

Lets look at the last couple games from notebookcheck (who actually reviews integrated graphics regularly).

Metro LL-Basically equal, not sure on hd 4000 discrepancy.

Lets take one game at a time,

Intel Core i5 3360M is a 35W, 3MB L3, 1.2GHz GPU
Intel Core i7 3610QM is 45W, 6MB L3, 1.1GHz GPU

Trinity A10-4600M is equal to 45W HD4000 and 66% faster than 35W HD4000 at 1366x768 High and almost 2x faster at Medium.

Grid 2 - AMD advantage

35W A10-4600M is 70% faster than 45W HD4000 at 1366x768 High.

Dota 2- basically equal
http://www.notebookcheck.com/Dota-2-Benchmarks.99462.0.html

35W A10-4600M equal to 55W HD4000

Company of heroes 2-Equal

35W Trinity A10-4600M equal with 45W HD4000 and 47W HD4600 Haswell

The Bureau

35W Trinity 18fps, 45W IvyBridge HD4000 16fps and 47W Haswell HD4600 17fps.

I will give this to Trinity just because it is a 35W APU.

Saints Row-

Again, 35W Trinity A10-4600M is faster than 45W HD4000 and 47W HD4600

(note the 750m is using ddr3 which is why its behind the 660m). Sadly there is little difference between the HD 4000 and the HD 4600.

And if you will see the performance of a 37W Haswell HD4600 it will be even slower.
Everyone says that Intel has the edge in Mobile but even with a node advantage and higher TDPs HD4600 is slower than 35W one year old Trinity.
And we havent seen any performance results for 35W Ritchland A10-5750M with 1866MHz memory. It will completely destroy any Intel HD graphics at the same TDP levels.

And now the good part, HD5000 and 5100 will be at an even lower TDP of 28W and Bellow. I really like to see the results of 15W HD5000 because i strongly believe that all those 40EUs will hardly work at the maximum frequency at that low TDP levels.

So, where people see that Intel has the advantage in Mobiles iGPU ???

 

[Sweepr]

Cant wait to see Iris HD5100 (28W TDP) results. It will be fun if it actually beats AMD's 35W APUs in graphics, cause the CPU performance advantage is a given. Mobile Trinity was a mere 20% faster than mobile HD4000 (inside a quad-core) according to AnandTech, time to close the gap and beat them.

 

[AtenRa]

Cant wait to see Iris HD5100 (28W TDP) results. It will be fun if it actually beats AMD's 35W APUs in graphics, cause the CPU performance advantage is a given. Mobile Trinity was a mere 20% faster than mobile HD4000 (inside a quad-core) according to AnandTech, time to close the gap and beat them.

It may beat Trinity but i doubt it will even be close to 35W Ritchland with 1866MHz memory.

 

[boxleitnerb]

The A10-5750M is barely faster than the A10-4600M in GPU-bottlenecked scenarios:
http://www.anandtech.com/show/7106/...art-1-the-apu-and-radeon-hd-8650g-performance

Dream on.

 

[AtenRa]

Dream on user self, they used 1600MHz memory in that review. It is only logical that there will be not a lot of performance difference between the two in situations when they are memory bandwidth starved.

But i clearly said it will not be close to Ritchland when using 1866MHz memory.

 

[piesquared]

Is that another AMD marketing slide ??

Or maybe this one ??

You have to understand that when you code specifically for a giver architecture you get much higher performance. The A8-4500M in Mixamo Face Plus is faster than Intel HD4600 either you like it or not.

You will see the same with AES and AVX, CPUs that dont have those instructions will perform poorly against modern hardware no matter if they are 8-cores 400mm2.

Nice test. This is just the tip of the Iceberg and they are already dominating. hUMA enable products are where jaws start to drop. The fact that software is already shifting to heterogeneous computing with dozens on board for HSA and the efficiency it brings means developers will be substantially rewarded.

 

[Abwx]

Mobile Trinity was a mere 20% faster than mobile HD4000 (inside a quad-core) according to AnandTech

Anand used a 45W IB , that is , 30% higher TDP than Trinity ,
increase the latter by 30% TDP and the difference will be 50%,
so much for your flawed 20%...

 

[mikk]

Is that another AMD marketing slide ??

Or maybe this one ??

You have to understand that when you code specifically for a giver architecture you get much higher performance. The A8-4500M in Mixamo Face Plus is faster than Intel HD4600 either you like it or not.

You will see the same with AES and AVX, CPUs that dont have those instructions will perform poorly against modern hardware no matter if they are 8-cores 400mm2.

The Musemage benchmark is a OpenGL one and yes Intel sucks here, their GPU utilization is very bad in this benchmark what makes me think it's either a driver or application issue. PhotoZoom on the other side proves my point. i3-3225 with a relatively low clocking 1050 Mhz HD4000 only 5 seconds slower than AMD fastest 100W APU. Compare it with the marketing slide from post 1 and you should be able to see the difference. In general, Intel is doing better in GPGPU computing applications due to the bad GPGPU handling from AMDs VLIW4 architecture. You can see it in lots of OpenCL based benchmarks like Luxmark, Folding@Home, CLBenchmark, but also browser GPU benchmarks like Fish Bowl. On paper HD4600 is slower and you would expect AMD would crush Intel in all of these GPGPU benchmarks, in real life HD4600 can often compete with AMDs on paper faster GPUs or even outpace them. A8-4500M is different, it's so much slower in gaming and GPGPU.

 

[Enigmoid]

Dream on user self, they used 1600MHz memory in that review. It is only logical that there will be not a lot of performance difference between the two in situations when they are memory bandwidth starved.

But i clearly said it will not be close to Ritchland when using 1866MHz memory.

Very few systems use 1866 mhz memory. Especially $500-600 notebooks.

The Trinity 4600M is strictly limited to 35W TDP while
the IB in your graphs is a 45W that can go up to 55W
on short times , the HW is a 47W that burst to 60W levels ,
that is , you re comparing a 35W TDP to litteraly desktop parts
and then claim that Intel is not that far from AMD GPU wise.

Indeed , if you take parts that have basicaly 30% higher TDPs
that can be pushed to 60% higher TDP than Trinity , but then
give the latter the same TDP envelloppe and we ll see if your
conclusions will be stretched the same way...

Do you critics even look up the specs of the CPUs in question?

The i7-3610qm has a max turbo of 1100 mhz and 6 MB L3. The i5 chips run the igp at anywhere from 1100 mhz to 1250 mhz with 3 MB L3. With the exception of the i3-3110m, i3's run the igp at 1100 mhz too with 3MB cache. I would consider the 35 watt ivy i5 mobile chip at 1150 to run pretty much the same as the 1100 mhz i7-3610qm.

I was talking about the A8-5545m Ritchland, what you post is about Trinity A8-4555m.

Yes but there are almost no benchmarks on that chip. Add maybe (generously) 20% to that.

You have to be kidding me. You are using a 55W Core i7 4930MX for the Mobile HD4600 to show that it is very close to a 35W A10-4600M ??
And not only that, even that Intel 55W HD4600 losses at 1366x768 in many games to the last year 35W Trinity A10-4600M. Just image how bud a 37W Intel HD4600 will perform against 35W Ritchland A10-5750M that also supports 1866MHz memory.

Lets take one game at a time,

Intel Core i5 3360M is a 35W, 3MB L3, 1.2GHz GPU
Intel Core i7 3610QM is 45W, 6MB L3, 1.1GHz GPU

Trinity A10-4600M is equal to 45W HD4000 and 66% faster than 35W HD4000 at 1366x768 High and almost 2x faster at Medium.

35W A10-4600M is 70% faster than 45W HD4000 at 1366x768 High.

35W A10-4600M equal to 55W HD4000

35W Trinity A10-4600M equal with 45W HD4000 and 47W HD4600 Haswell


35W Trinity 18fps, 45W IvyBridge HD4000 16fps and 47W Haswell HD4600 17fps.

I will give this to Trinity just because it is a 35W APU.

Again, 35W Trinity A10-4600M is faster than 45W HD4000 and 47W HD4600

1. The only setting which metro is even remotely close to playable is 1024 x 768 low in which they are basically equal. Something is funny on that bench because the i5 should not be that slow.

2. Grid 2 favours AMD at higher settings in that test. Yes.

3. Dota 2, where are you getting 55 watts? Notebook routinely uses the i7-3610m in their gaming tests.

4. Again continued, 35 watt and 45 watt ivy bridge don't show that much difference.

Fast clocked i7 against a ULV i5. The quad is a whopping 36% faster. And thats a higher clocker quad with 8 MB cache vs a ULV. For 35/35 vs 45/47 watt cpus the difference will be much smaller. 35 watts is plenty to keep two cores at 3.2 ghz and the igp at 1100-1200 mhz in games.

I also said that the typical HD 4600 is going to be between the ivy bridge and the haswell cpu on that test.

And if you will see the performance of a 37W Haswell HD4600 it will be even slower.
Everyone says that Intel has the edge in Mobile but even with a node advantage and higher TDPs HD4600 is slower than 35W one year old Trinity.
And we havent seen any performance results for 35W Ritchland A10-5750M with 1866MHz memory. It will completely destroy any Intel HD graphics at the same TDP levels.

And now the good part, HD5000 and 5100 will be at an even lower TDP of 28W and Bellow. I really like to see the results of 15W HD5000 because i strongly believe that all those 40EUs will hardly work at the maximum frequency at that low TDP levels.

So, where people see that Intel has the advantage in Mobiles iGPU ???

If you look at the MBA review you can see that the advantage of the HD 5000 at 15 w is limited but around 10-15% on average that 15 watt HD 4400, higher in situations where the CPU is stressed less.

 

[AtenRa]

The Musemage benchmark is a OpenGL one and yes Intel sucks here, their GPU utilization is very bad in this benchmark what makes me think it's either a driver or application issue. PhotoZoom on the other side proves my point. i3-3225 with a relatively low clocking 1050 Mhz HD4000 only 5 seconds slower than AMD fastest 100W APU. Compare it with the marketing slide from post 1 and you should be able to see the difference. In general, Intel is doing better in GPGPU computing applications due to the bad GPGPU handling from AMDs VLIW4 architecture. You can see it in lots of OpenCL based benchmarks like Luxmark, Folding@Home, CLBenchmark, but also browser GPU benchmarks like Fish Bowl. On paper HD4600 is slower and you would expect AMD would crush Intel in all of these GPGPU benchmarks, in real life HD4600 can often compete with AMDs on paper faster GPUs or even outpace them. A8-4500M is different, it's so much slower in gaming and GPGPU.

That doesnt look like 5 secs to me, it is 23 secs difference. 41% faster on A10-6800K than Core i3

 

[MisterMac]

That doesnt look like 5 secs to me, it is 23 secs difference. 41% faster on A10-6800K than Core i3

If your only trumph card is an obscure program that resizes images - (and costs a ton, when free alternatives achieve somewhere around 95% percentile feature\quality) that probably doesn't even have an active install base of 500k in the entire world....

Well, you tell me what this information does for the average user-joe.
You think that's a helpful information-base - to make a decision on?

You think the target market where richland M APUs appear consider they're photo imaging and resizing time spent?

Really?


If you have to grasp for this - (and don't start on MuseMage it's clearly professional targetted app) as a way to prove superiority - in a niche area that your chips will be used for probably by less than 1% - don't you think it's better consider other alternatives ?

...And do you really wanna sell this to the public as your trumph card? Really?

 

[Markfw]

I have had it with this PR post and the replies and flames.

Locking this flame bait.