CPU performance with dedicated Sound/LAN card?

[dumbfire]

Hi All!
A few clarifications on why this thread is categorized under CPUs:
1. This is NOT about sound quality.
2. This is NOT about network performance.
3. This IS about CPU performance.

I am interested in learning if there is a CPU performance difference in using a dedicated sound/network card vs using the motherboard's onboard sound/network solutions with modern 2011 processors. I am also interested if there is a different level of difference between dual and quad (and hexa) core processors for the above test. My theory is that quad core processors will have less of a performance hit than dual core processors from using onboard sound/networking. Thus, leading to an additional advantage for purchasing a quad core (or hexa) processor over a dual core.

Obvious benchmarks for the above would be Frames per second in games (as they use sound and can use networking). However, other tests could be applicable such as playing music and running a winzip benchmark or doing multiple FTP transfers to a local server while running a winzip benchmark.

Anyone have any benchmarks proving or disproving my theory?

Much appreciated.
Thanks!

 

[Zap]

There can be a measurable difference, but with modern multi-core CPUs there likely won't be a seat-of-the-pants difference. AKA: You won't notice it. You will notice a far, far greater performance difference by sticking $150 into a better graphics card or an SSD than into a sound card + NIC. Better yet, just put $150 more into the CPU itself (or better cooling so you can OC higher). For gaining CPU performance, money put into discrete sound cards and NICs is money NOT well spent.

from this AnandTech review

Note that this was at maximum throughput. CPU utilization goes down with lower throughput.

Also note that a number of PCI and PCIe NICs got higher CPU utilization than one of the motherboard NICs. This goes to show that drivers and PHY design affects it more than which interface it sits on.

Speaking of which, these days the motherboard NICs are exactly the same as a PCI/PCIe NIC, just that it sits on the motherboard and not in a slot. For instance you can buy Realtek NICs to go in a slot using the same chipset and drivers that motherboards use. Heck, some boards like the Asus P8P67 Pro use an Intel NIC just like you can buy separately. The motherboard NICs are even connected using PCIe.

 

[dumbfire]

Hi Zap,
Thanks for the reply. I think your comments invalidated an erroneous assumption I had which led to this thread...

"Speaking of which, these days the motherboard NICs are exactly the same as a PCI/PCIe NIC, just that it sits on the motherboard and not in a slot. For instance you can buy Realtek NICs to go in a slot using the same chipset and drivers that motherboards use. Heck, some boards like the Asus P8P67 Pro use an Intel NIC just like you can buy separately. The motherboard NICs are even connected using PCIe."

Interesting. I was unaware of the above. This information makes my question of dedicated vs onboard audio/network cards irrelevant.

I guess the next question would be more simple.

Suppose you have the top of the line dual core processor running XYZ gaming benchmark and it gets a score of 100.
Suppose you have the top of the line quad core processor running XYZ gaming benchmark and it gets a score of 100.

Now suppose you run the same benchmark with an added network load. My assumption would be that the quad core processor would have less of a performance hit than the dual core processor. Do you think this is correct or so marginal that it does not warrant investigation?

According to the article you linked there was roughly a 20% CPU utilization with the network under full load on an older dual core processor (E6300). Would you expect there to be a 20% hit to the XYZ benchmark making it 80 in my example above?
What about the quad core processor: also a a score of 80 in my example above?

I guess the point of my questions are to identify the additional practical advantages of going quad core vs dual core when running multiple benchmarks simultaneously. We all know quad core processors excel at certain thread optimized processes but I wonder if they also excel at executing multiple single thread optimized processes.

 

[Schmide]

It's not going to matter. Here's a little perspective to help you out.

Sound data is measured in kilobytes - even with effects on multi channel 100x oversampling you're never going to reach a rate more than a few megabytes or so a second.

Internet throughput is measured in megabits even high end connections and local lans work out to marginal megabytes of transfers. A gigabit connection theoretically can transfer 125 megabytes a second but typically maxes out at 35 megabytes a second.

The PCI-e lanes that handle all this transfer. They are measured in gigabytes. From the v1 at 250 megabyte a second per lane to the v3 1 gigabyte per second per lane. A typical 16 lane link can handle 4-16 gigabytes per second v1-v3 respectively.

On to the memory/frontside bus/cpu cache. Hear you're going from 5-15 gigabytes a second for memory to near 100 gigabytes a second for the L1 cache.

So putting it in perspective a maximum transfer on a gigabit lan comes in at less than 1% of memory bandwidth. Now there is some housekeeping that makes this many factors of magnitude more, but nothing that can put any dent into 1 let alone 4 CPUs.

 

[tweakboy]

Back in the day it was bad. But now these new mobo have great sound, even some video cards have great sound too, using hdmi ,,its better quality compred to 5 years ago. It doesnt hinder CPU at all.

 

[smakme7757]

Yea as stated above. With any of the more recent CPUs (last decade or so) you will not notice a performance loss at all and if you do it won't impact your games or productivity.

 

[Zap]

Now suppose you run the same benchmark with an added network load. My assumption would be that the quad core processor would have less of a performance hit than the dual core processor. Do you think this is correct or so marginal that it does not warrant investigation?

Marginal. I'm sure it is measurable in some way, but the difference would be so little that you would be unable to notice the difference without some software telling you there was a difference.

 

[Cerb]

Unless you get a NIC with a TCP offload engine, the performance difference between NICs is going to be barely measurable (even then, it will be rare that a single-user desktop will able to get much benefit from a TOE). There was quite a difference early on w/ GbE, but everyone got their chips and drivers sorted out, so now it's all a wash. Also, lower network utilization results in lower CPU utilization, for any given NIC, so the several KB/s used for gaming is not even worth a second glance.

Here are two reviews I was able to find, that are modern:
http://www.hardocp.com/article/2010/10/14/gigabyte_880gmaud2h_motherboard_review/3
http://www.hardocp.com/article/2011/01/27/gigabyte_p67aud4_motherboard_review/3

Here are two that aren't:
http://www.hardocp.com/article/2006/09/07/epox_ep5p945pro/4
http://www.hardocp.com/article/2006/02/15/asus_a8rmvp/5

Notice the differences? The new ones are good enough, to the point that all such tests can do is show that there is not a problem needing to be fixed. It takes what, 75%+ NIC utilization, to need to use 10%+ CPU, and that includes some rather significant overhead for windows file transfers. I've seen purely synthetic tests show better throughput and CPU utilization (I just can't find them via Google, ATM).