Sandy Bridge E starts at $294 (FUD)

[OCGuy]

Which means I can hopefully pick one up @ Microcenter for less!

 

[exar333]

Which means I can hopefully pick one up @ Microcenter for less!

!!!!!!MC rocks!

 

[AdamK47]

I'm glad there's a Micro Center near where I live. My wife isn't so glad though.

 

[Arachnotronic]

Hoping that the bet I made on AMD stock will pay off before SNB-E comes out.

 

[aphelion02]

Hoping that the bet I made on AMD stock will pay off before SNB-E comes out.

^lol

 

[pantsaregood]

Then you realize at 2133MHz memory speed, up to 6% performance improvement is a performance increase you otherwise would not have had.

Again, that <6% performance increase isn't something that's going to show up outside of synthetic benchmarks. Sandy Bridge isn't bandwidth starved. CPU clock speeds can, ideally, scale linearly. The same isn't the case for memory bandwidth - if it isn't saturated, more isn't better.

 

[mrjoltcola]

True, but there is more to memory speed than bandwidth, right? There is latency, which becomes more of an issue as CPU clock speeds scale. Otherwise we wouldn't need caches. It is a three-legged race.

 

[grkM3]

Again, that <6% performance increase isn't something that's going to show up outside of synthetic benchmarks. Sandy Bridge isn't bandwidth starved. CPU clock speeds can, ideally, scale linearly. The same isn't the case for memory bandwidth - if it isn't saturated, more isn't better.

Saved me about 10min ripping a bluray disk.

 

[lol123]

True, but there is more to memory speed than bandwidth, right? There is latency, which becomes more of an issue as CPU clock speeds scale. Otherwise we wouldn't need caches. It is a three-legged race.

Yeah, and it's easy to forget that increasing memory frequency will also decrease latency.

 

[Edrick]

Again, that <6% performance increase isn't something that's going to show up outside of synthetic benchmarks. Sandy Bridge isn't bandwidth starved. CPU clock speeds can, ideally, scale linearly. The same isn't the case for memory bandwidth - if it isn't saturated, more isn't better.

So a free (more or less) 6% performance increase is nothing?

And all the AMD fans keep posting how SB is not bandwidth starved, and how SB-E does not need quad channel. The fact that nobody knows how a 6 core-12 thread SB-E is going to perform. And basing any conclusions on SB-E solely on the results of quad core LGA1155 SB CPUs is just plain dumb. And with 8/10 core SB-E or IB-E CPUs a realistic possibility within the next year, I can almost guarantee that they will choke on dual channel memory.

 

[grkM3]

This sounds amazing.

Then you realize at 5.0 GHz, "300 MHz worth of clock speed" results in a theoretical performance increase of 6%. Given that programs never scale perfectly, that's under a 6% performance advantage.

Again, Sandy Bridge is not bandwidth starved.

That is, however, some beast RAM.

I dont get what your saying? it scored a lot higher in cinebench with the tighter timings.

If I ran the tight timings at 5300 mhz it would of mopped the floor with the loose timings run at 5300 mhz.

the faster you clock the cpu the more the extra bandwith will help,Im sure for stock clocked sandys there fine but once you load up 8 threads at 5.3ghz you start to use more of that bandwidth.

Its the very reason intel is going quad channel since there adding up to 10 cores soon.

10 cores running 20 threads at 5.3 ghz would bottle neck even my ram on dual channel.

I cant wait to try these runing quad channel,prolly break 60,000 MB/SEC lol.That is over kill but its nice to have bragging rights

I wish I stocked up on this ram,It was 129 for 2x2gb and now they sell for 300-350 if you can find them

 

[Arachnotronic]

So a free (more or less) 6% performance increase is nothing?

And all the AMD fans keep posting how SB is not bandwidth starved, and how SB-E does not need quad channel. The fact that nobody knows how a 6 core-12 thread SB-E is going to perform. And basing any conclusions on SB-E solely on the results of quad core LGA1155 SB CPUs is just plain dumb. And with 8/10 core SB-E or IB-E CPUs a realistic possibility within the next year, I can almost guarantee that they will choke on dual channel memory.

You miss the THG i7 3960X preview?

 

[Mhorydyn]

You're getting two more cores and because of that it's 40% faster than the 2600K in heavily multi-threaded workloads and the same speed in single-threaded.

Noob-ish question, but is that last bit about single threaded performance accurate? Will the 3930K at 3.2Ghz be comparable to the 3820 at 3.6Ghz at every task? Going for the hex core model won't be a decrease in performance in any area?

 

[Edrick]

You miss the THG i7 3960X preview?

What about it?

 

[mrjoltcola]

Noob-ish question, but is that last bit about single threaded performance accurate? Will the 3930K at 3.2Ghz be comparable to the 3820 at 3.6Ghz at every task? Going for the hex core model won't be a decrease in performance in any area?

Hi Mhorydyn,

Assuming it is still the same Sandy Bridge architecture inside, no it won't be comparable for every task. Single threaded performance for the same architecture doesn't depend on number of cores, it depends on how fast one core runs and the per-clock performance of that core (not counting other things like cache).

So no, 3.2Ghz won't be comparable to 3.6Ghz from 3930K -> 3820 for CPU bound, single-threaded applications.

This is why, when the first Core-2 Duo Extreme's came out, the dual core 2.93Ghz Extreme Edition, or an overclocked Duo was in demand somewhat even after quad core 2.66Ghz appeared, because for gaming and many other things, the dual-core was faster.

 

[Mhorydyn]

Hi Mhorydyn,

Assuming it is still the same Sandy Bridge architecture inside, no it won't be comparable for every task. Single threaded performance for the same architecture doesn't depend on number of cores, it depends on how fast one core runs and the per-clock performance of that core (not counting other things like cache).

So no, 3.2Ghz won't be comparable to 3.6Ghz from 3930K -> 3820 for CPU bound, single-threaded applications.

This is why, when the first Core-2 Duo Extreme's came out, the dual core 2.93Ghz Extreme Edition, or an overclocked Duo was in demand somewhat even after quad core 2.66Ghz appeared, because for gaming and many other things, the dual-core was faster.

I know that that's normally the case, but with the max turbo boost frequency of the two CPUs being so similar (3.8 vs 3.9), I figured that may play a role in balancing out the performance between the two despite the difference in base clock speed. Sadly, I haven't yet properly read up on turbo boost, and am unsure of the effect it'd have in comparisons like this.

 

[mrjoltcola]

I know that that's normally the case, but with the max turbo boost frequency of the two CPUs being so similar (3.8 vs 3.9), I figured that may play a role in balancing out the performance between the two despite the difference in base clock speed. Sadly, I haven't yet properly read up on turbo boost, and am unsure of the effect it'd have in comparisons like this.

Oh, well, I also haven't read that much on the chips either, so I might be wrong. I was assuming similar turbo ratios from their base speeds.

 

[Grooveriding]

3930K and a Gigabyte X79-UD5 are in my future. Not happy about spending $500 on a CPU, what I always liked about CPUs was spending about $250-$350 and getting the $1000 performance with overclocking, but oh well.

I want the unlocked multi and Intel knows it, bastards.

 

[pantsaregood]

So a free (more or less) 6% performance increase is nothing?

And all the AMD fans keep posting how SB is not bandwidth starved, and how SB-E does not need quad channel. The fact that nobody knows how a 6 core-12 thread SB-E is going to perform. And basing any conclusions on SB-E solely on the results of quad core LGA1155 SB CPUs is just plain dumb. And with 8/10 core SB-E or IB-E CPUs a realistic possibility within the next year, I can almost guarantee that they will choke on dual channel memory.

I'm an AMD fan now? I'm running a 2500K and consider it somewhere between extremely unlikely and impossible for Bulldozer to match Sandy Bridge.

Think about this for just a moment.

50% more cores means that, theoretically, 50% more bandwidth could be used. The SB memory controller is designed for DDR3 1333. Assuming core scaling is linear (it isn't perfect), six-core SB-E should be happy with dual channel DDR3 2000.

You could even claim it would be happy with DDR3 1600 in dual channel, given SB doesn't see any significant performance degradation with DDR3 1066.

You are also mistaken. We know EXACTLY how SB-E will perform. SB-E is just Sandy Bridge. We know how Sandy Bridge performs. There's nothing magically different about SB-E. It has two more memory channels, extra cache, and (in some cases) extra cores. The CPU itself, however, is an unchanged SB.

 

[Edrick]

50&#37; more cores means that, theoretically, 50% more bandwidth could be used. The SB memory controller is designed for DDR3 1333. Assuming core scaling is linear (it isn't perfect), six-core SB-E should be happy with dual channel DDR3 2000.

And by that same logic, the 8 core versions (they exist already) will require, theoretically, 100% more bandwidth. And start thinking ahead to OCed IB-E cpus, which should run faster than SB-E, will also require the additional bandwidth.

I am not saying it will make a huge difference with the initial release of SB-E, but I am positive it will make a difference during the lifecycle of the lga2011 socket.

But even performance increases of 3-7% may be worth it to many people. Remember there are many applications which love memory bandwidth. Granted they are more "workstation specific", but not everyone uses their PC just to game and surf the internet.

I'm an AMD fan now?

Sorry to imply that you were. I have just been argueing with known AMD fans about this same topic for months now.

 

[mrjoltcola]

There are a lot of softwares that are license-constrained per CPU socket. I have single socket VMware servers in my datacenter that do expensive document conversion, pdf generation, and image scaling on the fly within web applications, and 6&#37; is 6% more capacity before I have to migrate VMs.


I have Oracle Standard Edition One database that is licensed per processor socket, but doesn't limit me based on cores, up to a point (6 maybe), so per-socket performance allows me to stretch my budget on more than just the price of the $300 CPU. Oracle SE1 costs $5500 per socket. When I move a database from an old Netburst Xeon to a E3-1275 I quadruple my capacity without adding more licenses, for a fixed hardware expense. That's why some of us don't hesitate to add $100 for 6% more hardware (though I wouldn't overclock my RAM to get the 6%, but just commenting on the significance of 6%).

 

[LOL_Wut_Axel]

Noob-ish question, but is that last bit about single threaded performance accurate? Will the 3930K at 3.2Ghz be comparable to the 3820 at 3.6Ghz at every task? Going for the hex core model won't be a decrease in performance in any area?

I'm comparing the 2600K and 3930K...

They'll be very similar in single-threaded, but in multi-threaded the Hexa-Cores pull away by ~40%. After all, they do have 50% more cores.

 

[Mhorydyn]

I'm comparing the 2600K and 3930K...

They'll be very similar in single-threaded, but in multi-threaded the Hexa-Cores pull away by ~40%. After all, they do have 50% more cores.

Since the 2600K (3.4Ghz/3.8Ghz) and the 3820 (3.6Ghz/3.9Ghz) are, on paper, somewhat similar, I'm assuming that the 3930K and the 3820 should also perform similarly on single threaded tasks. I'm perfectly willing to spend extra for the 3930K for the few times I'd want the extra multi-threaded performance as long as the single-threaded performance isn't markedly worse than the 3820 (which seems to be the case).

 

[PreferLinux]

3930K and a Gigabyte X79-UD5 are in my future. Not happy about spending $500 on a CPU, what I always liked about CPUs was spending about $250-$350 and getting the $1000 performance with overclocking, but oh well.

I want the unlocked multi and Intel knows it, bastards.

What for? You can overclock the one with locked multipliers just fine, according to the page done by Bit-tech, which AnandTech then re-wrote.