How can a mobile CPU appear to be so fast?

Toggle sidebar Toggle sidebar
G

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
A friend of mine just bought a laptop with an Intel Core i7 4700HQ @ 2.4 GHz and I saw on different benchmark websites that my CPU (an Intel Xeon E5 1620 v2 @ 3.7 GHz) is only about 20% faster than his, how can that be ?!?

The core count is the same (4), the number of threads is the same (8), the lithography is the same (22nm) and I would presume that the architecture is just about the same.

But the clock frequency is at least 50% faster and there are most likely quite a few architectural trade-offs must have been made to fit a smaller mobile form-factor and yield lower wattage (such as a narrower and slower memory bus, fewer PCIe lanes/QPI links etc.).

So why is the performance difference not bigger? Or could we safely assume that those synthetic benchmarks are not representative for the real world workloads?
 
Solution
jaguarskx


Assuming Turbo Boost is being used the i7-4700HQ runs at 3.4GHz and the E5-1620 runs at 3.8GHz. Since a Haswell generation CPU is on average 6% more powerful than Ivy Bridge generation, the i7-4700HQ is roughly equivalent to a 3.6GHz Ivy Bridge CPU. From a raw CPU performance perspective, the difference between 3.8GHz and 3.6GHz is about 5.5%...
Sort by date Sort by votes
Someone Somewhere

Someone Somewhere

Titan
Sep 23, 2012
22,291
8
84,965
Yours is Ivy Bridge, his is Haswell. The architectural improvements are ~15%, IIRC.

Clocking down and less I/O is about the only change mobile chips have. Desktop and laptop have same memory and are usually the same die.

Synthetics are hopeless, though. And performance doesn't always change directly with clock rate, either.
 
Upvote 0 Downvote
tadej petric

tadej petric

Honorable
Feb 9, 2013
826
0
11,010
Mostly beacuse yours is older (IB<HW) and beacuse clock rates are not linear (you wouldn't get 100% preformance gain from 100% clock gain)
Yours is from xeon family (workstation) that has its ups and downs compared to core family as well.

But 20% can be quite a lot, depending on the benchmark.
 
Upvote 0 Downvote
G

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
I actually looked into the 'cpubenchmark.net' website and they apparently use the PassMark benchmark for comparison. I'm not sure how PassMark works but if it is single-threaded then the Turbo Boost may kick in and increase the clock to 3.4 and 3.9 GHz for the 4700HQ and E5-1620v2 respectively.

The maximum memory bandwidth of the 4700HQ is 25.6 GB/s with only 2 memory channels whereas is is 59.7 GB/s for the E5-1620v2 with 4 memory channels.

So memory intensive workloads ought to have a much bigger performance difference between these two CPUs than what these benchmarks show.
 
Upvote 0 Downvote
jaguarskx

jaguarskx

Titan
Apr 19, 2006
27,979
3
84,965


Assuming the same # of cores and same clockspeed, the average improvement from Ivy Bridge to Haswell is only 6%.
 
Upvote 0 Downvote
jaguarskx

jaguarskx

Titan
Apr 19, 2006
27,979
3
84,965


Assuming Turbo Boost is being used the i7-4700HQ runs at 3.4GHz and the E5-1620 runs at 3.8GHz. Since a Haswell generation CPU is on average 6% more powerful than Ivy Bridge generation, the i7-4700HQ is roughly equivalent to a 3.6GHz Ivy Bridge CPU. From a raw CPU performance perspective, the difference between 3.8GHz and 3.6GHz is about 5.5%.

http://ark.intel.com/products/75116/Intel-Core-i7-4700HQ-Processor-6M-Cache-up-to-3_40-GHz

http://ark.intel.com/products/64621/Intel-Xeon-Processor-E5-1620-10M-Cache-3_60-GHz-0_0-GTs-Intel-QPI

Because the Xeon's memory bandwidth is 2x that of the Haswell, the difference in the benchmark results are likely due to the memory bandwidth.
 
Upvote 0 Downvote
Solution
G

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
Ok, now I'm getting confused here; if a Haswell CPU is on average 6% faster than an equivalent Ivy Bridge CPU (in terms of # cores/threads, cache and clock frequency), then how on earth can a 2.4GHz Haswell be equivalent to a 3.6 GHz Ivy Bridge CPU?!? It doesn't make any sense to me.

And btw, Intel Xeon E5-1620v2 has a clock speed of 3.7 GHz, not 3.6, and it turbo boosts to 3.9 Ghz, not 3.8.

And if we should nitpick here, the effective bandwidth of the Xeon is 4 times as much as the 4700HQ, but I cannot see how that difference shows up in the PassMark benchmark.

But yes, if we assume that the benchmark is single-threaded and both CPUs operate in full turbo boost mode, then the difference, on average shouldn't be more than 8% (3.9/3.4/1.06) if no memory access is taking place.

But it appears that PassMark is a multi-threaded benchmark tool, so it doesn't seem likely that TurboBoost is enabled during this test.
 
Upvote 0 Downvote
jaguarskx

jaguarskx

Titan
Apr 19, 2006
27,979
3
84,965
Without Turbo Boost the Core i7-4700HQ runs at only 2.4GHz, but with Turbo Boost that max core speed jumps to 3.4GHz.


The max bandwidth for the E5-1620 v2 is 59.7 GB/s vs. 25.6 GB/s for Haswell. Given the 20% performance difference between the processors from the benchmarks and an estimated 5.5% performance difference in raw CPU performance, the logical conclusion is the benchmarks are very sensitive to memory bandwidth.

http://ark.intel.com/products/75779/Intel-Xeon-Processor-E5-1620-v2-10M-Cache-3_70-GHz?wapkw=e5-1620v2
 
Upvote 0 Downvote
G

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
I tend to disagree here, if the benchmark never trigs the turbo boost, then given the 6% advantage of the slower CPU and that no memory access is taking place, the performance difference should be 45%. I.e. the Xeon should be 45% faster than the mobile i7. If memory bandwidth would have mattered here, the performance difference would have been even bigger, i.e. the Xeon CPU would have been even faster than the mobile i7 CPU.

But that is not the case here since the benchmark only shows a 20% difference in performance.
 
Upvote 0 Downvote
jaguarskx

jaguarskx

Titan
Apr 19, 2006
27,979
3
84,965
If the benchmarks do not cause the i7-4700HQ exceed 2.4GHz, then perhaps there is something faulty with the synthetic benchmarks. It generally tend to ignore synthetic benchmarks and focus on real application benchmarks.

For example, since I encode videos from time to time, I look at video encoding performance with programs like Handbrake to see the actual performance difference between two CPUs.
 
Upvote 0 Downvote
G

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
Well, we're talking about PassMark here since it is the only benchmark I could find that shows a measured score for both CPUs. When investigating this benchmark further it turns out that this benchmark "will run as many simultaneous tests as the processor is able to handle in parallel", as written in the page I linked to. In such a scenario, it is very unlikely that the CPU will run in TurboBoost mode.
 
Upvote 0 Downvote
whoratesit

whoratesit

Honorable
Jun 8, 2012
201
0
10,760
On the 4700HQ the 2.4 GHz base clock is simply to save batteries, four cores will happily turbo up to 3.2 GHz. On your E5 1620 V2 the four active cores will run at the base clock 3.7 GHz and when fewer cores are active they may boost to a maximum of 3.9 GHz.

For single core apps you have a clock advantage of : 3.9 vs 3.4 = 14.7%
For multi core apps you have a clock advantage of : 3.7 vs 3.2 = 15.6%

Your advantage is lowered by the fact that Haswell can complete more ops per clock.

According to user benchmarks your actual single core CPU advantage is : 90.2 vs 83.2 = 8.4%
According to user benchmarks your actual multi-core CPU advantage is : 538 vs 464 = 15.9%

Bottom line, that laptop CPU is awesome!

sources:
http://cpu.userbenchmark.com/SpeedTest/71/IntelR-CoreTM-i7-4700HQ-CPU---240GHz
http://cpu.userbenchmark.com/SpeedTest/8354/IntelR-XeonR-CPU-E5-1620-v2---370GHz
http://www.notebookcheck.net/Intel-Core-i7-4700HQ-Notebook-Processor.93264.0.html
 
Upvote 1 Downvote
G

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
Yes indeed, that CPU is awesome until it breaks. That soldered unreplaceable BGA package (blergh!!!) of CPU means that all you can do once it breaks is to throw the laptop in the garbage bin.

While those spittle-flecked environmentalist idiots are talking their throats off about starving polar bears and global warming, the one thing that they can do and don't is to enforce manufacturers to make products that last longer. The rate of obsolescence has slowed down considerably compared to only 10 years ago, that's why manufacturers make products impossible to repair combined with diminishingly short warranty periods so as to force people to ceaselessly buy new products and throw the old stuff into the ever-growing mountain of toxic waste.

Let's hope that they will release some high clocked Haswell based Xeon for the LGA2011 platform, that would have been nice.

Edit: I looked once again into the charts and saw that the Intel Xeon E3-1630 3.2GHz is pretty much on par with the 4700HQ in spite of the two generations older Sandy Bridge platform that this Xeon is based upon and its slightly slower memory bus (2 channels @21GB/s). So if the 4700HQ runs at 3.2GHz then there is no advantage of it, if we were to take the PassMark benchmarks seriously.
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom