4core/dual channel vs 6core/quad channel?

[benjagan91]

Hi guys,

Just trying to get the general consensus here. Assuming the same clock speeds on the quad/hex cpus and memory modules (though the quad channel will have double the memory bandwidth) - What's the benefit to the end user?

I do heavy multitasking and like a responsive system (So an SSD is obviously a necessity for starters).

Thanks guys

[Renza]

quad channel won't have much real world benefit for just multitasking. depends what you call multitasking though. If you do a lot of computational stuff, then quad channel may make a difference.

you should base your decision more on if you need 6 cores or not, and whether the increased cost is worth it for you

[DarkYendor]

Memory bandwidth is rarely going to be the limiting factor in your performance, especially now that the memory controller is in the CPU.

For example, the 2500k has a maximum memory bandwidth of 21GB/s. You'd be hard-pressed to find a situation that can utilise that. Maybe churning certain data in databases or matlab, but not in anything you'd often find running on a desktop rig.

However, if you could saturate the 21GB/s, you could see a benefit by going up to the 3930k, with its 51.2GB/s bandwidth.

[bobbth]

If you require a very responsive system then it may be worth looking at memory latency over memory bandwidth, from memory Sandy Bridge has quite lower memory latency than Sandy Bridge E so you may actually find the 4cores/dual channel a better choice.

[gmeup]

Originally Posted by bobbth

If you require a very responsive system then it may be worth looking at memory latency over memory bandwidth, from memory Sandy Bridge has quite lower memory latency than Sandy Bridge E so you may actually find the 4cores/dual channel a better choice.

wouldnt memory latency be more to do with the memory sticks rather than the platform. quad channel over dual channel should still be faster pound for pound using the same sticks of ram with same specs. Dont know if you need quad channel or even 6 core cpus unless you are using software which can take advantage of all that hardware.

[benjagan91]

Originally Posted by bobbth

If you require a very responsive system then it may be worth looking at memory latency over memory bandwidth, from memory Sandy Bridge has quite lower memory latency than Sandy Bridge E so you may actually find the 4cores/dual channel a better choice.

I believe memory latency and memory clocks both determine memory performance. To simplify: 5ns memory with 500mb/s bandwidth equals 10ns memory with 1000mb/s bandwidth. Please correct me if I'm wrong.

I'll be running dual monitors for web development, have the usual 50 browser tabs, security suite, torrents, multimedia playback, RDP/Teamviewer, photoshop, skype, dropbox/backup/raid and a VM or two running.

At the moment I'm looking at the two following CPUs:

• Xeon 6 core SB-EP , 2.0GHz , 3/3/4/4/5/5 turbo, 15MB L3, 95W TDP, 7.2GT/s, Quad Channel DDR3
  
• Xeon 4 core IB, 2.5GHz , 6/7/9/10 turbo, 8MB L3, 45W TDP, DMI 2.0, Dual Channel DDR3

Thermals and energy efficiency are also important to me. I understand that using the 4 core Ivy Bridge chip allows me to run my dual monitors through the integrated CPU graphics rather than installing a HD5450 for my needs?

[HobartTas]

Greetings

Originally Posted by benjagan91

Hi guys,

Just trying to get the general consensus here. Assuming the same clock speeds on the quad/hex cpus and memory modules (though the quad channel will have double the memory bandwidth) - What's the benefit to the end user?

I do heavy multitasking and like a responsive system (So an SSD is obviously a necessity for starters).

Thanks guys

I concur, there is not going to be a lot in it either way as you probably only really notice it if your gaming.

Originally Posted by benjagan91

I believe memory latency and memory clocks both determine memory performance. To simplify: 5ns memory with 500mb/s bandwidth equals 10ns memory with 1000mb/s bandwidth. Please correct me if I'm wrong.

I'm not a memory expert but how RAM actually gets used is a bit complicated, firstly if its in the CPU's cache then memory gets bypassed altogether, secondly Intel CPU's in normal usage never saturate the memory bus, and thirdly I believe if they access memory in burst mode the subsequent accesses after the initial memory address is a bit quicker (but I could be totally wrong on this aspect).

If you really want to know how much of an impact this is in real life then do this run Memtest and make special note of the memory bandwidth, also run 3dmark, prime etc, then

(1) go into your BIOS and downclock your existing ram speed by say 20% and do the benchmarks again.

(2) put the memory back to normal speed and bump up the timings e.g. if they are 10-10-10-27 make them 11-11-11-30 or 12-12-12-34, run benchmarks again

(3) drop the speed AND increase the timings and run benchmarks again

(4) if you have enought memory in your current system, take out one channel's RAM and leave the other one alone, run benchmarks but just be aware that if your really low on memory it will severely impact the benchmarks but Memtest would still be OK however so have a look at memory throughput in that case.

Why reduce? because it will be a rough proxy (in reverse) for what would happen if you got faster memory with tight timings (not strictly accurate but good enough), you'll probably find that performance will only drop a couple of a percent and I reckon maybe 10% at the most for test (3).

Anyway, just looking at what you are going to use it for I guess any platform would do the job you want (assuming there's a generous amount of memory in it) as I can't see that your really going to stress any particular part of the machine, here are a couple points to consider

1. Socket 2011 generally will consume more power than S1155 (30-40? watts at idle).

2. S1155 motherboards are limited to 32GB ram (chipset limitation), Socket 2011 consumer motherboards are limited to 64GB (chipset limitation), server 2011 motherboards are limited to 196GB.

3. If going the S2011 route will you be needing more than 64Gb memory? If no then it would probably be easier to get a consumer S2011 board and either a 3820 or 3930K as you can always overclock them later if you want to.

4. Do you want or need ECC memory, in that case you will need a server motherboard and you MUST also have a Xeon to utilise it, as a 3820,3930K won't support it, also if you must have the Xeon and you put it into a consumer S2011 you can't use ECC either.

5. ECC memory is a lot more expensive than normal non ECC memory, even more so for the larger sized sticks.

5. If VM's are important then

(a) the 3820,3930K has VT-D and VT-X and is also overclockable whereas

http://ark.intel.com/products/63697/...up-to-3_80-GHz)

(b) Xeon's are not overclockable, also

(c) the Non-"K" S1155 cpu's have VT-D but the "K" versions are overclockable but have it disabled, compare these two

http://ark.intel.com/products/65719/...up-to-3_90-GHz)

http://ark.intel.com/products/65523/...up-to-3_90-GHz)

6. Windows 7-64 home premium is licensed only for 16GB max memory, if you want more you'll need a higher version

http://msdn.microsoft.com/en-us/libr...(v=vs.85).aspx

7. Socket 2011 doesn't have the integrated graphics as it's not on the S2011 cpu's itself like it is on the S1155 one, also how do you run integrated on two monitors? can you like use the HDMI and DVI simultaneously on the S1155 motherboard? You will probably have to get a crappy low end card for graphics.

8.

Originally Posted by benjagan91

Thermals and energy efficiency are also important to me. I understand that using the 4 core Ivy Bridge chip allows me to run my dual monitors through the integrated CPU graphics rather than installing a HD5450 for my needs?

Why exactly are thermals and power such a huge consideration, is it just for the cost of the power consumed? or are you trying to cram all this stuff into a small HTPC case with poor cooling? There's no point getting something that's underpowered or limits you too much as you might be replacing it sooner rather than later.

So thinking about what your going to be doing with it I'm guessing that what you predominantly want is a system that has plently of spare resources available so that if you start something afresh it can get going without having to share the workload with something else already in use (or currently overloaded) so I would recommend that you get;

(a) Socket 2011 and a hexacore cpu as all S2011 cpu's are 2 threads per core (unlike S1155 where only some are) so If I had to recommend something it would be the 3930K for 6 cores and 12 threads.

(b) something like the gigabyte X79-UD5 principaly because it has 8 memory slots and.

http://www.gigabyte.com.au/products/...px?pid=4049#ov

(c) PCCG has 32GB of G.Skill at $219.00 so get 64GB of this for a total of $448.00 (8 x 8GB sticks).

(d) Windows 7 ultimate.

Of couse if you want to go the Xeon route there's no real problems in that regard, it will cost you a bit more for a server motherboard and a lot more for the ECC memory especially if you get a lot of it.

So what you could do for example is say assign a VM to use say 16GB and use the affinities in task manager to only let it use cpu's 11 and 12 and a second VM to use another 16GB and to only use cpu's 9 and 10 and your still left with the equivalent of a quad core (8 threads) and the remaining 32GB for the rest of your stuff, so I just can't see somthing like this system ever getting overloaded for your intended usage.

This setup would cost a bit more but should have greater longevity, especially once you place more demands on it later on such as running more VM's. etc, etc.

Food for thought


Cheers

[trevor68]

Interesting article on Xbit Labs about memory speeds/latency. Apparently latency is not so important these days, but you can see up to 20% gains depending on speed.

http://www.xbitlabs.com/articles/mem...idge-ddr3.html

[Soarer GT]

I actually purchased a i7 3820 and 64GB kit from Scorptec last month. Just as a caveat if you use an ASUS P9X79 board, it will need a bios flash to use the latest CPU, so have a usb key handy (the USB bios update button is freaking AWESOME).

My thoughts on 64gb? It's only useful for databases and vmware (which are still limited by disk IO generally). As a comparison, my home machine I run a Q6600@3.2ghz (intel 520 SSD for OS) with 8GB ram (DDR2-800 dual channel) and it's as fast for "daily" tasks (Word, Excel, Browsing, 1 or 2 VM's) as the i7 3820 with 64GB ram.

Database crunching is where the main difference is. If you're not doing "intense" workloads, there's probably no difference.