Intel Core i7-3960X and i7-3930K CPUs Transitioning to C2 stepping in January

[PaulieG]

Yeah, I'm not saying that every application is going to be faster in a large way. But usage scenarios that do take advantage of the extra bandwidth that the platform provides, do excel. There are beenfits to be had, but I connot deny that the average user isn't going to benefit from it. Those that run 3 or 4 VGAs, and multiple monitors...yes, they will.

I mean really, 1155 is Intel's entry-level platform, IGP included. It is a killer entry level paltform though, without a doubt, and will provide 99% of users will more power than they need. I am not one of those users.

The extra power consumption is a tough pill to swallow, but like I care. For the games I play, it's better.

SiSoft Sandra is the only bench right now that shows the extra Bandwidth. It's just about the algorithms used for the tests that are at fault. Many need updates, but I do use Sandra in my memory reviews, so you can see the difference there, at least.

Really, if you want affordable highly threaded, FX-8150 is the CPU of my own choice. Cost/Perforamnce is damn good.

If it must be X79, 3930K with and OC warranty, with 4.8 GHz is damn snappy. Xeons, with no OC, and no OC warranty, are not something I'd personally consider.


For mutli-GPU gamers on X79, the 4-core with HT, if priced right, would be the CPU to get, because as mentioned the extra cores don't offer much, overall.

So, Dave, what is the benefit of x79 vs P67/Z68 besides the extra bandwith? Beyond gaming with multi gpus, what usage scenarios would justify choosing x79 over the significantly cheaper platform? I've tried to come up with a shred of justification to do the upgrade, just like I do with every new platform. However, finding this justification has been more difficult than usual. Really not being rhetorical here...

 

[radrok]

Really, if you want affordable highly threaded, FX-8150 is the CPU of my own choice. Cost/Perforamnce is damn good.

If it must be X79, 3930K with and OC warranty, with 4.8 GHz is damn snappy. Xeons, with no OC, and no OC warranty, are not something I'd personally consider.


For mutli-GPU gamers on X79, the 4-core with HT, if priced right, would be the CPU to get, because as mentioned the extra cores don't offer much, overall.

The FX would be a downgrade for me, my 980x does 4.4-4.5 GHz easily so that's why I was hesitant on X79, I guess I'm going to wait a little more.

 

[cadaveca]

So, Dave, what is the benefit of x79 vs P67/Z68 besides the extra bandwith? Beyond gaming with multi gpus, what usage scenarios would justify choosing x79 over the significantly cheaper platform? I've tried to come up with a shred of justification to do the upgrade, just like I do with every new platform. However, finding this justification has been more difficult than usual. Really not being rhetorical here...

That's an easy question. There is no use outside of that specific scenario, for the users that tend to be here on TPU. That's why I have the 3960X...it's not useful for W1zz in GPU reviews using a single card, and his already posted CPU compare shows that in that regard, the 1366 platform is still more than viable for most users.


This is cutting edge stuff. unless you live on the edge of high-end performance, and have those sort of performance needs, SKT 1155 is more than adequate. The only reason I'm using the 3960X it is becuase it was free. As soon as I can get a quad for this socket, I'm gonna. I'm not even overclocked right now, as the higher power consumption just isn't worth it, and CPU performance @ stock is enough for me, getting ~70FPS or more(usually 140 FPS or so in BF3 with 6950 Crossfire at stock, with stock shaders. With 1155, that dips below 60 quite often.

I mean, I kinda stopped really using an OC'ed system for gaming largely becuase of the realization that for what I want, memory bandwidth is more important. Plus, at stock, the difference between minimum framerates and maximums is much smaller, leading to a smoother gameplay expereince for me.

The FX would be a downgrade for me, my 980x does 4.4-4.5 GHz easily so that's why I was hesitant on X79, I guess I'm going to wait a little more.

If you have that, I see no need for you to upgrade at all.

 

[PaulieG]

That's an easy question. There is no use outside of that specific scenario, for the users that tend to be here on TPU. That's why I have the 3960X...it's not useful for W1zz in GPU reviews using a single card, and his already posted CPU compare shows that in that regard, the 1366 platform is still more than viable for most users.


This is cutting edge stuff. unless you live on the edge of high-end performance, and have those sort of performance needs, SKT 1155 is more than adequate. The only reason I'm using the 3960X it is becuase it was free. As soon as I can get a quad for this socket, I'm gonna. I'm not even overclocked right now, as the higher power consumption just isn't worth it, and CPU performance @ stock is enough for me, getting ~70FPS or more(usually 140 FPS or so in BF3 with 6950 Crossfire at stock, with stock shaders. With 1155, that dips below 60 quite often.

I mean, I kinda stopped really using an OC'ed system for gaming largely becuase of the realization that for what I want, memory bandwidth is more important. Plus, at stock, the difference between minimum framerates and maximums is much smaller, leading to a smoother gameplay expereince for me.



If you have that, I see no need for you to upgrade at all.

Thing is, I'm all about "bleeding edge". I've been an early adopter for years, because the curiosity of new platforms just drives me crazy, both with AMD and Intel. Hell, I even bought a couple of Bulldozer chips to play with. I'm not feeling that way with this platform though. I guess it's a matter of limited performance increase and features over SB that just does not justify spending double the price of a 2600K, even from the perspective of a dedicated enthusiast. Still, it's killing me that there's a new platform, and I won't give into temptation.

 

[cadaveca]

Thing is, I'm all about "bleeding edge". I've been an early adopter for years, because the curiosity of new platforms just drives me crazy, both with AMD and Intel. Hell, I even bought a couple of Bulldozer chips to play with. I'm not feeling that way with this platform though. I guess it's a matter of limited performance increase and features over SB that just does not justify spending double the price of a 2600K.

That's it exactly. I mean, I don't feel there's any need to point out the obvious, that for rendering and such, computational-heavy workloads, X79 excels, for sure.

Unfortunately, 1155 has little bus adjsutment, and on 2011, bus adjustments have no noticble impact on cache or memory performance, so even the extra available clocking isn't really all that useful, either. Core performance on 2011 doesn't affect memory bandwidth much, either, like it does on 1155, so even overclocking only has so much benefit on 2011, when it comes to gaming. I mean, of course it's faster when OC'd, but power consumption jumps so quickly, it's hard to justify

For me, I want three GPUs, to go with my three monitors for Eyefinity. Single monitor, 1155 is more than adequate.

 

[radrok]

I think this chip REALLY needs a 22nm shrink if it's hot as Dave says, imagine what would have happened if they released a complete SB-E die, 8/16 with unlocked multi and the frequency of the 3960X.
Would TDP have been 150W or higher? I believe the SB-EP 8c part is 150W so it may have been higher.

 

[cadaveca]

Probably more like 170-180W, based on my crappy sample.

And that is STOCK.

Xeons have more cores, but core frequency is much lower 2.1 - 2.6 GHz, I think? NOt 100% sure on that, TBH, like i said, no OC, means no use for me. NOt that I personally OC, but I need chips that will push boards properly so you guys can see what's what.

I don't really care who's faster, more expensive..that's up to you and your wallet. Every CPU has it's pros and cons.

 

[PaulieG]

Probably more like 170-180W, based on my crappy sample.

Damn.

 

[cadaveca]

130w divided by 6 cores, multiplied by 8 cores = 178w. probably a bit less in real-world, due to the shared NB and such, but still...not that much less.


Why do ya think those that already have chips are interested in revisions? It's very clear that power consumption holds the SKT 2011 platform back a fair bit in 24/7 clocking. 4.3 GHz on air on all cores may be pushing it for a lot of chips. Most boards have 4.3 GHz for "Auto-clock"

 

[Aquinus]

22nm die shrink of SB-E will be IVB-E but with tri-gate transistors. Just keep in mind that smaller circuitry makes for higher resistance, so similar voltages will result in more heat. (I can run my Phenom II 940 at 1.575v on air safely (granted I have a massive cooler), I don't think you would want to do that with a 32nm chip.) But we don't know. There might be 3 sources and 3 drains, but there is still 1 22nm wire.

Also Xeons are designed for highly parallel workloads that require an extremely reliable platform. Memory is CRC'ed using ECC so any errors that crop up in memory (which already is very uncommon in server platforms,) can usually get corrected, but for this, memory performance takes a hit. But since workloads are highly parallelized and you have 4 memory controllers per CPU (or two depending if it is a lesser or low-voltage model), you have a massive amount of raw power. It's a matter of being able to utilize the resources the server has, so ECC usually isn't an issue. It's what these chips were designed to do, but even in server environments, you will find that for a web server, a single request will most likely be single threaded if its a script in PHP or something like that, but it's the 50 people trying to run a script at the same time which is where the Xeon shines.

Edit: I'm also curious how Intel plans on getting around the quantum tunneling issue at sizes <= 16nm circuitry, but that's just a tangent.

Edit 2:

130w divided by 6 cores, multiplied by 8 cores = 178w. probably a bit less in real-world, due to the shared NB and such, but still...not that much less.

Keep in mind that the CPU has many shared components, like the L3 cache, memory controller, and DMI so it doesn't quite scale like that. Also if this is really is based on a Xeon die, then a CPU with all 8 cores activated would be downclocked to match the 130-watt TDP. You won't see higher TDPs in servers.