AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
And yet in x264 the FX is 25-26% faster, just shows how biased Cinebench really is.
Your conclusion is wrong because the GPU results dont have a negative effect in CPU Physics.
And yet in x264 the FX is 25-26% faster, just shows how biased Cinebench really is.
Your conclusion is wrong because the GPU results dont have a negative effect in CPU Physics.
DrMrLordX
Lifer
- Apr 27, 2000
- 22,696
- 12,651
- 136
They got better when they took over design of their own chipsets (though now they've farmed that off again for the stuff they can't/won't cram onto the die?). Remember when the best chipsets on the market for AMD products were usually Via junk? The 1.4 ghz Tbird was a fantastic processor, but the Via kt266/kt266a chipsets and the infamous IDE drive corruption thing was just lurvely. I'm lucky I didn't run into that on my system (go go ECS k7s5a!), but I know some others that had worse luck.
Sadly, there's more to the downstream than just making sure that motherboards hosting your CPUs aren't utter trash (and if you look at some FM2+ boards, you have to wonder if they've been slipping). If the boards are (relatively) solid but the OEM systems are not . . .
By the by, if anyone's really concerned that <insertpopularbenchmarkhere> is hurting AMD chips by showing them in an unfavorable light due to alleged Intel optimizations, it might be a good idea to stick to open source benchmarks and compare across a range of compilers. At least with open source, you know what the code is written to do (assuming you can read it). Then you can make your comparisons between gcc, ICC, MSVC, Open64, LLVM/Clang, or whatever else you can think of. That's also assuming your code is in C or C++.
Well you are comparing integer vs. FP. And Kaveri has relatively weak FP performance.
Beema (6410) posts x264 (11.8 and 54) numbers in line with BT, but FP numbers are a solid increase.
x264: pass one/pass 2
CB R15: single/multi
3dmark physics: ice storm/cloud gate
Kaveri 7500
x264: 68.5/13.9
CB: 55/153
3d mark: 19978/1721
N3540
x264: 54.3/11.1
CB: 43/158
3dmark: 21540/1880
A6-6410
x264: 54.5/11.8
CB: 49/165
3dmark: 23238/2006
Easy to see that Kaveri has strong integer performance and weak FP performance (2 vs. 4 FPU). CB may not be the best measure of overall performance but operating on the fact that it is biased isn't true either. Kaveri simply doesn't do FP as well as the market alternatives relatively speaking.
Yeah is SO biased that it gets a considerable ST lead but it goes to hell on MT...
wth has the gpu anything to do with CPU MT?
And btw, x264 benchmark takes about 2-3min MAX.
Last edited:
Would be great to provide the links of thoses reviews..
As for CB R15 score of Kaveri it varies of 10% depending of the used TDP, laptops that exhaust the CPU TDP get 62/167 :
http://www.notebookcheck.com/Test-HP-EliteBook-755-G2-J0X38AW-Notebook.125836.0.html
Others like the one you have opportunisticaly cherry picked dont exhaust the CPU TDP and provide scores like this 12"5 laptop at 55/151 with a CPU that is below 15W during rendering CPU tests :
http://www.notebookcheck.com/Test-HP-EliteBook-725-G2-Notebook-J0H65AW.126960.0.html
A 15"6 :
http://www.notebookcheck.com/Test-HP-Elitebook-745-G2-Notebook.125610.0.html
You will notice that all thoses three laptops have fans that spin at low speed even at full load if we are to look at the noise, and all use a A10 7350B....
Also, why no more Cinebench 11.5 and CB R15 instead, did you notice something that favour one brand going from the former to the latter..?.
So it appears that these are numbers from notebookcheck but they have no review of this extraordinary N3540 that get 4.8% better score than the N3530 in CB 11.5 and 10% better in CB R15, this imply 10% higher frequency while the CB 11.5 score tell us that it s running at 2.5.
Indeed at thoses score a Baytrail N3530 laptop will consume the same as the 12"5 HP laptop with the Kaveri 7350B :
http://www.notebookcheck.com/Test-Update-Asus-X751MA-TY148H-Notebook.134897.0.html
So all things being equal you have just demonstrated that Baytrail has not wonderfull perf/Watt once it s clocked at higher frequencies that the initialy intended target, the hard spinning fan at full load say it all, to compare with Kaveri s laptops silence..
Last edited:
I simply took the average numbers from the CPU page.
Make of it what you will. They are implementation numbers, not perfect numbers.
The max power numbers are only a measure of the maximum power the notebook can consume. They are not a measure of typical power or efficiency.
Case in point the BT 3530 runs at max clocks and consumes 17W more over max idle.
About 7W for the SOC
On the other hand Kaveri throttles during the tests.
23W over max idle and the chip throttles to 1.1 Ghz on the CPU and 211 mhz on the gpu.
The 15W elitebook also throttles to 1.1 Ghz on the CPU and 280-410 mhz. 27W over max idle power.
And before you start throwing stones at intel for not following TDP, the A10-7350B is rated at 19W yet both notebooks have a CPU package power consumption of 23W.
Make of it what you will. They are implementation numbers, not perfect numbers.
The max power numbers are only a measure of the maximum power the notebook can consume. They are not a measure of typical power or efficiency.
Case in point the BT 3530 runs at max clocks and consumes 17W more over max idle.
About 7W for the SOC
On the other hand Kaveri throttles during the tests.
23W over max idle and the chip throttles to 1.1 Ghz on the CPU and 211 mhz on the gpu.
The 15W elitebook also throttles to 1.1 Ghz on the CPU and 280-410 mhz. 27W over max idle power.
And before you start throwing stones at intel for not following TDP, the A10-7350B is rated at 19W yet both notebooks have a CPU package power consumption of 23W.
It s really impossible for you to not twist the numbers once it s Intel, from 6.3 to 27W i see a difference of 20.7W.
http://www.notebookcheck.com/Test-Update-Asus-X751MA-TY148H-Notebook.134897.0.html
Case in point is that you re deliberatly producing false numbers to get the 7.5W fake TDP being adequate with thoses actual 20.7W, wich they arent, as said a 15W beema laptop use typicaly 24-27W with 20-22W delta...
On the other hand Kaveri throttles during the tests.
23W over max idle and the chip throttles to 1.1 Ghz on the CPU and 211 mhz on the gpu.
The 15W elitebook also throttles to 1.1 Ghz on the CPU and 280-410 mhz. 27W over max idle power.
And before you start throwing stones at intel for not following TDP, the A10-7350B is rated at 19W yet both notebooks have a CPU package power consumption of 23W.
What you dont understand is that Intel always throttle the GPU, wich point to the CPU being the main power absorber, while AMD always throttle the CPU instead, power that is not used by the CPU is devoted to the GPU.
In the Kaveri laptops i linked you can see that fans have deliberatly kept silent, hence the less thermal headroom and 34-37db typical noises, to compare with the Baytrail "7.5W" that is at 43.8dB, obviously corporate dedicated laptops, like thoses HP, trade some perfs for less noise.
As for the infos reported by HW info obviously they are not accounting for all power, in no way there can be only 7.5W at the CPU level and as much as 20.3W at the main as delta...
Edit : The 12"5 Elite Book run at 26.4W at the main actualy.
Last edited:
Except Maxpower (Furmark + Prime) is done using max brightness, wifi on, high performance which is the 10.8W number (6.3W is min brightness, power saver, wifi off). Its 16W delta (27-10.8) before power losses from the power brick.
Different notebooks mean nothing fan/noise related can be compared.
The elitebook runs 24.4W under 3dmark and 32.5 under prime + Furmark (throttling).
Looks like someone did not get the memo that Intel does not throttle GPU anymore.
I was asking because the SDP driven Atom BT cant do that, its not about OEMs, its about, they cant, they can mantain full CPU turbos or use the igp and reduce the cpu clock to the base.
So i had no idea that notebook BT can sustain full turbo and IGP.
I was asking because the SDP driven Atom BT cant do that, its not about OEMs, its about, they cant, they can mantain full CPU turbos or use the igp and reduce the cpu clock to the base.
So i had no idea that notebook BT can sustain full turbo and IGP.
Last edited:
dave_the_nerd
Lifer
- Feb 25, 2011
- 16,991
- 1,620
- 126
So, is an FX-7500 better than an FX-6300? :awe:
(There is a special place in hell for the people who name some of these products.)
(There is a special place in hell for the people who name some of these products.)
What is amazing is that you re constantly using the excuse of screen and other parts that are apparently more power hungry on Intel based systems, it s not like AMD based such items are not tested at full brightness and so on, here a 17" Beema if it s the screens that bother you, difference is 1.7W at full load, so i guess that Beema is a 8.5W or so chip despite the better perfs...
http://www.notebookcheck.net/HP-Pavilion-17z-Notebook-Review.124669.0.html
This say that the 3530 is pushed way above 7.5W as its CB 11.5 score point to more than 2.4GHz on MT, indeed the pics you linked show a GPU that is 100% throttled and that the CPU can get the full, inflated, TDP headroom.
As for the elite book it run the stress test Prime + Furmark at :
26.4W after a short moment, i guess that you deliberatly didnt quote the full thing...
With a 20W delta at full loads that get our Kaveri at barely 12-13W on the long run, as i told you the fan is spinning slowly even in this 12.5" format, it s not by chance.
http://www.notebookcheck.com/Test-HP-EliteBook-725-G2-Notebook-J0H65AW.126960.0.html
Lol, Furmark say that the GPU is at 0hz..
Indeed let s compare thoses pics from Notebbokcheck and see where they are convenient with some manufacturer :
Let s start with AMD, i took the pic above :
And now with Intel N3530 :
Noticed.?.
The task bar has disappeared, we no more know since how much time P95 is running, they even masked its window, Furmark say that the GPU is, or was, running since one hour, at 0hz and with no %age of utilisation when the screen was "captured"....
At thoses conditions, and the pushed TDP, it can run at thoses settings...
I have a lot of trouble understand the exact meaning of your replies. Are you typing into google translate or something?
I can't follow this. Where is the GPU 100% throttled?
Anyway. BT has terrible driver support and stuff isn't reported properly on GPU-Z and other utilities.
As far as NBC, different reviewers use slightly different testing setups. Cue the different Furmark settings/versions or HWinfo vs. Hwmonitor.
I can't follow this. Where is the GPU 100% throttled?
Anyway. BT has terrible driver support and stuff isn't reported properly on GPU-Z and other utilities.
As far as NBC, different reviewers use slightly different testing setups. Cue the different Furmark settings/versions or HWinfo vs. Hwmonitor.
DrMrLordX
Lifer
- Apr 27, 2000
- 22,696
- 12,651
- 136
I don't know that I really agree with your statement. We're making that assumption based on what we think is going on in certain software and on module diagrams of Kaveri. If you actually look at simple software that does nothing but cram integer and/or floating-point operations down Kaveri's throat as quickly as possible and examine the results, it might surprise you.
In my own testing, Kaveri had very similar performance in raw integer and floating-point addition and multiplication operations. Granted, this was running Java, but the JVM is actually quite good at producing machine code from bytecode that isn't terribly complicated (and it can spit out AVX, which Kaveri supports and supports well). Floating-point division is leaps and bounds better on Kaveri than integer division.
At least, that was the case with a dataset that fit inside L1 and instructions that could easily be SIMD-optimized by the JVM.
I have limited data for what happens when Kaveri is running code that can not (or simply is not) optimized for SIMD instructions. What I did notice suggests that fp performance suffers more under such circumstances than integer.
Now, don't get me wrong, I'm not trying to say anything about Kaveri here in comparison to some other uarch. That's not the point. The point is, when comparing Kaveri to itself, Kaveri CAN sustain 32-bit fp performance in line with its 32-bit integer performance.
Last edited:
There can be a language barrier but to say it short a N3430 has 2.5W less TDP than a Beema 6310.
http://www.notebookcheck.com/Test-Update-Asus-X751MA-TY148H-Notebook.134897.0.html
http://www.notebookcheck.com/Test-Acer-Aspire-E17-E5-721-69FX-Notebook.129080.0.html
As pointed by the member above stress tests say only one thing on appearance, that is the allowed power comsumption for the CPU/GPU, now if you are to check how many frames were rendered in Furmark in one hour it s three time less with BT than with the power limited Kaveri, this put the stress tests numbers in perspective more than just checking their influence on comsumption alone...
Last edited:
coercitiv
Diamond Member
- Jan 24, 2014
- 7,225
- 16,982
- 136
Excel or equivalent software
https://docs.google.com/spreadsheets/d/1fLXY4XoLFj7x2nUEKHJd7qIZNhYgBpgD0FQFMDD4YJQ/edit?usp=sharing
there yah go
is it weird that there is a refresh rate option of 40Hz in catalyst/windows deisplay settings? is it possible that amd might be implementing mobile freesync?
Last edited:
coercitiv
Diamond Member
- Jan 24, 2014
- 7,225
- 16,982
- 136
Interesting: CPU is throttling, but it's not frequency related and temps are fine.
At 17:31 thread usage goes down fast and system power usage goes from 25-28W to 18-20W, and stays that way for the remainder of the test.
Does the unit get the same score while plugged in? What Power Plan are you using?
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter