[Coolaler] Core i7 3930K vs Core i7 3960X

[StrangerGuy]

For those benchmarks your arguement holds true. 15MB to 12MB makes very little, if any, difference. But as I did mention, there are some applications (workstation) which love more cache. Once these CPUs are launched, and I seem some results from the workstation camp, and if the results are the same, I will eat my words and say you were right. I am holding my opinion until then however.

IIRC one of the pecularities of the ring bus of SB's L3 cache is that the lower the L3 capacity, the higher the bandwidth. I think that explains why the 6MB 2500K is virtually indistinguisable from the 8MB 2600K at the same clocks without HT.

 

[Ajay]

unless they were just being lazy and/or were just cutting corners to ensure stability, I can't say I'm too enthused by their overclocking results. 4.6GHz @ 1.44+v isn't too encouraging, not terrible, but not exactly much of an improvement from Gulftown.

Well, it is an ES sample, who knows what week it was produced.

 

[3DVagabond]

Wait is the 3960x ivy bridge ? If not, hold your horses everyone, let this pass. If you already have a Sandy, its a pointless upgrade you wont see a difference in real world usage. gl

Depends on what world you live in.

 

[LOL_Wut_Axel]

For those benchmarks your arguement holds true. 15MB to 12MB makes very little, if any, difference. But as I did mention, there are some applications (workstation) which love more cache. Once these CPUs are launched, and I seem some results from the workstation camp, and if the results are the same, I will eat my words and say you were right. I am holding my opinion until then however.

The difference could very well be bigger in those, but I did say from the beginning the extra money it costs could be worth it to some corporations with workstation workloads. It just depends on how much money it can save them in the long-term. But since this is a CPU enthusiast forum, that's who I was referring to. For desktop workloads it appears to have a smaller-than-1% difference if you have the same OCs. Since in enterprise you don't overclock then the difference could be anywhere from 2-5% faster.

 

[LOL_Wut_Axel]

All of which are GPU limited.

The leaked benchmarks in the OP do use 3-way GTX 580s. Unfortunately, the only thing that was run was 3DMark Vantage/11. The GTX 580s are clocked very high at 950MHz for a 3-way setup. Mine are at 850MHz and I still manage this:

Vantage - P59093:
http://3dmark.com/3dmv/3559058

11 - P16408:
http://3dmark.com/3dm11/2001148

I'm really hoping my future SB-E system performs better than these leaked benchmarks show.

Edit: I see what's wrong with their Vantage results. It's not using PhysX. Either it's disabled, or the driver isn't installed.

I wouldn't count on it, to be honest. Almost all games use two-four cores, and almost none care for Hyper-Threading. They also love high single-threaded performance, which is why you see even a Core i3-2100 beating a Phenom II X4 955 in gaming and no gain after you get to the i5-2500K. SB-E is the same architecture as SB, so nothing will change there. Ivy Bridge will bring with it a 5% IPC increase and ~5% higher clock speeds plus PCIe 3.0 support, so it'll probably be better for gaming than SB-E.

In any case, does it really matter? Sure, you get no gains in gaming compared to SB, but the six-cores will still be well over 30% faster than the quad-cores given the right workload.

 

[AdamK47]

How are you able to conclude that the IPC will be a 5% increase and the clock speeds will also be a 5% increase. Are there leaked benchmark/overclocking results for ES chips? I'm sure there are increases, but where is this 5% coming from?

 

[ed29a]

How are you able to conclude that the IPC will be a 5% increase and the clock speeds will also be a 5% increase. Are there leaked benchmark/overclocking results for ES chips? I'm sure there are increases, but where is this 5% coming from?

You must be new here. The 5% is coming straight from someone's behind. This is Anandtech forums man ... where people guess stuff left and right and play armchair chip designers!

Screw benchmarks on ES chips, we don't need facts here, just baseless speculation!

 

[LOL_Wut_Axel]

How are you able to conclude that the IPC will be a 5% increase and the clock speeds will also be a 5% increase. Are there leaked benchmark/overclocking results for ES chips? I'm sure there are increases, but where is this 5% coming from?

Ivy Bridge is considered a tick from the CPU perspective but a tock from the GPU perspective. On the CPU core side that means you can expect clock-for-clock performance improvements in the 4 - 6% range. Despite the limited improvement in core-level performance there's a lot of cleanup that went into the design. In order to maintain a strict design schedule it's not uncommon for a number of features not to make it into a design, only to be added later in the subsequent product. Ticks are great for this.

Source

Since it's at its core the same architecture we'll see more overclocking headroom, but not by much. Like going from Nehalem to Westmere, expect ~5% higher overclocking headroom. Comparing the difference was going from 4GHz to 4.2-4.3GHz on average. We'll see the same thing pan out here.

You must be new here. The 5% is coming straight from someone's behind. This is Anandtech forums man ... where people guess stuff left and right and play armchair chip designers!

Screw benchmarks on ES chips, we don't need facts here, just baseless speculation!

Nice try.

 

[ElFenix]

intel's naming is still stupid. can they stick with a convention for more than about 3 months?

core ix-3xxx should be haswell and derivatives, not SB derivatives.

 

[toyota]

lol, does the naming even matter? the people that need to know can easily figure it out. at the same time, the people that cant figure it out probably do not even need to.

 

[ed29a]

Source

Since it's at its core the same architecture we'll see more overclocking headroom, but not by much. Like going from Nehalem to Westmere, expect ~5% higher overclocking headroom. Comparing the difference was going from 4GHz to 4.2-4.3GHz on average. We'll see the same thing pan out here.

Nice try.

And the article you linked is full of ES/production CPU benchmarks ... right?

Didn't 'experts' predict Bulldozer is going to be great for multi threaded uses but only slightly behind 2600K in single threaded use? And we all know how those 'expert' predictions turned out! Not saying Ivy Bridge will be a flop, probably very far from it on the other end of the spectrum. But ... what if Intel is able to crank out more speed out of the die shrink? What if they can't? What if they'll have issues with their tri-gate process? Plenty of variables we don't know yet until someone benchmarks either the final CPU or the ES version.

Sorry, until then, it's speculation, not facts. Unless you consider company PR slides as 'facts'.

 

[LOL_Wut_Axel]

And the article you linked is full of ES/production CPU benchmarks ... right?

Didn't 'experts' predict Bulldozer is going to be great for multi threaded uses but only slightly behind 2600K in single threaded use? And we all know how those 'expert' predictions turned out! Not saying Ivy Bridge will be a flop, probably very far from it on the other end of the spectrum. But ... what if Intel is able to crank out more speed out of the die shrink? What if they can't? What if they'll have issues with their tri-gate process? Plenty of variables we don't know yet until someone benchmarks either the final CPU or the ES version.

Sorry, until then, it's speculation, not facts. Unless you consider company PR slides as 'facts'.

No one said that. You're making that up. Everyone knew Bulldozer wasn't gonna be a contender in single-threaded performance. The only ones that thought otherwise were AMD fanboys, and their opinions don't matter when looking at something objectively.

It's mostly a die shrink including some enhancements that didn't make it to Sandy Bridge. If Anandtech is saying it'll be a 4-6% IPC improvement it's because that's what will, in fact, happen. Or do you think they don't have engineers or other extremely knowledgeable people that analyze the architectural tweaks and deduct how much difference they'd make? They're not pulling this out of their ass. There's not "too many variables". Ivy Bridge is Sandy Bridge at 22nm with new 3D transistors and a few architectural enhancements. End of story.

 

[Arachnotronic]

I'm not sure why people think Ivy Bridge is anything more than a die shrink...

 

[ed29a]

No one said that.

The only ones that thought otherwise were AMD fanboys

'No one' excludes AMD fanboys?

This is a CPU that Tom's obtained and benchmarked.

Have a link for Tom's Ivy benchmark? I can't find it on their site. I am genuinely curious.

It's mostly a die shrink including some enhancements that didn't make it to Sandy Bridge. If Anandtech is saying it'll be a 4-6% IPC improvement it's because that's what will, in fact, happen. Or do you think they don't have engineers or other extremely knowledgeable people that analyze the architectural tweaks and deduct how much difference they'd make? They're not pulling this out of their ass.

The only proof they give is Intel PR slides. I don't consider those facts. Not saying Anandtech aren't right or wrong, but without benchmarks you don't know, I don't know, and people can only guess and speculate (my original point).

There's not "too many variables". Ivy Bridge is Sandy Bridge at 22nm with new 3D transistors and a few architectural enhancements. End of story.

Brand new way of making transistors and that alone is not source of uncertainty? You don't think that we should wait before seeing how that works out before throwing numbers left and right? Interesting logic there!

 

[LOL_Wut_Axel]

'No one' excludes AMD fanboys?

LOL, grasping at straws at its finest. No one, as in "no one with an actual rational mind." Fanboys of any type are included there.


Have a link for Tom's Ivy benchmark? I can't find it on their site. I am genuinely curious.

Meant to reply that in another comment. Deleting it.


The only proof they give is Intel PR slides. I don't consider those facts. Not saying Anandtech aren't right or wrong, but without benchmarks you don't know, I don't know, and people can only guess and speculate (my original point).

Well, we do know. You just make it out to be as if looking at the architecture we can't deduce what the performance improvements will be, which is something that CAN be done, and which is something in the article HAD been done. Again, do you think they have stupid monkeys there that don't know how architectural enhancements can make a CPU faster, and by how much? Didn't think so.

Only real thing that's left to "speculation" is if the enhancements make it faster on the lower-end scale (4&#37 or the higher-end (6%). To round it out I just choose 5%.

Also, this is exactly the same situation we had with Conroe and Penryn. Penryn is also based on the Core 2 microarchitecture but due to architectural improvements had 5% higher IPC than Conroe.


Brand new way of making transistors and that alone is not source of uncertainty? You don't think that we should wait before seeing how that works out before throwing numbers left and right? Interesting logic there!

Brand new way of making transistors affects nothing regarding what I wrote before about IPC. The new transistors make more efficient use of the die and will enable lower power consumption. They can also affect overclock-ability, but I'm taking the improvement we had from Nehalem to Westmere in that as a base for that. It won't deviate in any significant way from that when it comes to clock speeds.

What's in bold.