TomsHardware: Core i7-3970X Extreme Review: Can It Stomp An Eight-Core Xeon?

[minitron]

The i7 completes the task in 91.5s
The FX completes the task in 116s

The i7 is not 21% faster than the FX [1 - (91.5/116)]
The i7 is 26% faster than the FX (116/91.5)

I understand there's 10 numbers and it's hard for some of you to master them all but it's something you'll probably have to learn eventually. Then again maybe not.

Take a book out of inf64's playbook: "I'm done with you.":'(

 

[Abwx]

The i7 completes the task in 91.5s
The FX completes the task in 116s


The i7 is 26% faster than the FX (116/91.5)

Now tell us by how much the 8350 is slower.......:whiste:

 

[AtenRa]

More time is not faster
Less time is Faster

116 Frames per Second are ~26,8% HIGHER or MORE or FASTER than 91.5 fps

116 seconds are ~26,8% SLOWER than 91,5 seconds (more time is SLOWER)

91,5 seconds are ~21,1% FASTER than 116 seconds (less time is FASTER)

Hope you got it now

 

[inf64]

Really guys it's pointless but I applaud your efforts . Maybe he is just trolling us all along and trolling himself in the process? I don't know,but what I know is he has no clue about some basic mathematical concepts.

 

[Idontcare]

The question I have is: with IB-E not coming out until Q3 2013, and with Haswell coming with much improved performance capabilities above and beyond even that of Ivy Bridge, is it actually plausible/probable that a consumer-grade quad-core Haswell is going to best the performance of even the 150W TDP extreme 6-core SB-E chips on release?

 

[inf64]

10-15% over 3770K can hardly best 6C SB-E @ 3.5Ghz. Unless you recompile your workload yourself and your code benefits from AVX2 or FMA.

 

[Makaveli]

The question I have is: with IB-E not coming out until Q3 2013, and with Haswell coming with much improved performance capabilities above and beyond even that of Ivy Bridge, is it actually plausible/probable that a consumer-grade quad-core Haswell is going to best the performance of even the 150W TDP extreme 6-core SB-E chips on release?

10-15% over 3770K can hardly best 6C SB-E @ 3.5Ghz. Unless you recompile your workload yourself and your code benefits from AVX2 or FMA.

Agreed I think you will need a much larger IPC gap and some compiler tricks to bridge the gap in multithreaded work loads.

However in the Single threaded to 4 threaded workloads I can see Haswell doing very well and will win some benchmarks.

 

[Idontcare]

10-15% over 3770K can hardly best 6C SB-E @ 3.5Ghz. Unless you recompile your workload yourself and your code benefits from AVX2 or FMA.

Are you sure the top K model for a quad Haswell will be 3.5GHz?

Also, considering the microarchitectural improvements that are going in for bandwidth purposes, I'd expect MT workloads to see a disproportionate uplift with haswell versus IB. (which is why I am wondering if it might actually exceed that of SB-E, you've got a process node and two microarch iterations advantage at that point)

 

[ShintaiDK]

The speed and power gonna be very interesting, even with the VRM added ondie.

Or as David Kanter says:

Haswell is the first family of SoCs that have been tailored to take advantage of Intel’s 22nm FinFET process technology. While Ivy Bridge is also 22nm, Intel’s circuit design team sacrificed power and performance in favor of a swift migration to a process with a radically new transistor architecture.

Haswell-DT vs SB-E/IB-E in singlethreaded gonna be an utter nightmare for LGA2011 owners. And SB-E atleast might easily lose more than it wins in MT.

 

[Makaveli]

The speed and power gonna be very interesting, even with the VRM added ondie.

Or as David Kanter says:



Haswell-DT vs SB-E/IB-E in singlethreaded gonna be an utter nightmare for LGA2011 owners. And SB-E atleast might easily lose more than it wins in MT.

Lets say the IPC gap is 15% from Ivy to Haswell and it may be lower single digits.

At 15% its the same gap as nehalem and SB and the performance difference isn't that great depending on the application at the same clock speed.

so how would that be an utter nightmare?

 

[Sheep221]

Are you sure the top K model for a quad Haswell will be 3.5GHz?

Also, considering the microarchitectural improvements that are going in for bandwidth purposes, I'd expect MT workloads to see a disproportionate uplift with haswell versus IB. (which is why I am wondering if it might actually exceed that of SB-E, you've got a process node and two microarch iterations advantage at that point)

It depends on various things, first the conductive lines in such a small scale dies are not really anymore intended for operation on such a high frequencies and for a second every year the time space between software hardware gets wider and makes high frequencies less effective in obtaining more performance. In fact, due to more and more advanced hardware there maybe complete OC dismiss in the future.

 

[inf64]

Are you sure the top K model for a quad Haswell will be 3.5GHz?

Also, considering the microarchitectural improvements that are going in for bandwidth purposes, I'd expect MT workloads to see a disproportionate uplift with haswell versus IB. (which is why I am wondering if it might actually exceed that of SB-E, you've got a process node and two microarch iterations advantage at that point)

I give it 3.6Ghz with 4Ghz Turbo . And remember ,Haswell is all about power efficiency not super high clocks. The 10-15% speedup intel states is from clock+ipc. I suppose Haswell may be able to run at sustained turbo state longer than IVB/SB so there might be some benefit from there too.

 

[myocardia]

Lets say the IPC gap is 15% from Ivy to Haswell and it may be lower single digits.

At 15% its the same gap as nehalem and SB and the performance difference isn't that great depending on the application at the same clock speed.

so how would that be an utter nightmare?

I love how some people like to pretend as if Haswell isn't gonna have new instructions like TX and AVX2. With the right software, and even if Haswell only has a ~10% gain in IPC, Haswell could easily see up to a 100% gain per core*, at the same clock speed. Of course, that will only be in one or two of the best written, newly updated softwares. Still, it would suck to see your 6 core 3930k get slaughtered by a non-overclocked Haswell quad in the one software that you use the most!

* I personally only expect improvements of up to ~75-80% from TX/AVX2, but remember that you have to also add in the same IPC gain that Haswell will have with all other, older software, which may be as high as 15%.

 

[Lepton87]

Just look how sensitive WoW is to the cache size, despite clock frequency deficit and/or architectural improvements it still wins this benchmark.

Based on those numbers Xeon could easily be 130W part as it draws less power than 130W TDP 3960X. I wish Intel would release a fully unlocked SB-E procesor with 8 cores and no multiplier lock, as it turns out 1000$ is not enough for them to price their best processor. Extreme Edition should be Extreme if they can't or won't sell thir best processors for 1000$
they should price them at 2000$ and really keep them extreme. right now there's just a hair separating 3960k from their "Extreme" offering. I see nothing extreme in their current extreme line-up besides price.

Basically going with the "extreme" offering over 3960k makes sense only if that 400$ is pocket change for you.

 

[Makaveli]

I love how some people like to pretend as if Haswell isn't gonna have new instructions like TX and AVX2. With the right software, and even if Haswell only has a ~10% gain in IPC, Haswell could easily see up to a 100% gain per core*, at the same clock speed. Of course, that will only be in one or two of the best written, newly updated softwares. Still, it would suck to see your 6 core 3930k get slaughtered by a non-overclocked Haswell quad in the one software that you use the most!

* I personally only expect improvements of up to ~75-80% from TX/AVX2, but remember that you have to also add in the same IPC gain that Haswell will have with all other, older software, which may be as high as 15%.

Because history speaks for itself.

SB has had AVX for how long now and how many applications use it?

 

[myocardia]

Because history speaks for itself.

SB has had AVX for how long now and how many applications use it?

Haha, very true. I honestly don't know enough about this stuff to know for sure whether it is the type of software that would benefit or not, but if it benefits in any way at all, I definitely expect at the least to see most if not all CAD software that's commercially available to have AVX2 support, if not both AVX2 and TX. I also wouldn't be surprised to see PhotoShop having some support for either or both, but probably not immediately.

 

[Makaveli]

Haha, very true. I honestly don't know enough about this stuff to know for sure whether it is the type of software that would benefit or not, but if it benefits in any way at all, I definitely expect at the least to see most if not all CAD software that's commercially available to have AVX2 support, if not both AVX2 and TX. I also wouldn't be surprised to see PhotoShop having some support for either or both, but probably not immediately.

I do believe we will see more of a push to have software recompiled for AVX2 over AVX because it will provide more of a benefit.

The question is when?

And will it be within the first year of release or are you going to be waiting for broadwell before you start seeing it used in alot of apps.

 

[ShintaiDK]

Because history speaks for itself.

SB has had AVX for how long now and how many applications use it?

More than you might think. The issue is that AVX1 is FP and limited by that SB/IB/DB/PD needs 2 cycles to execute it. Haswell will do singlecycle.

Also the main benefit is AVX2 with vectoring of all scalar SSE functions plus gather. Visual Studio, Intel compilers etc already support and generate AVX2 code today.

 

[myocardia]

I do believe we will see more of a push to have software recompiled for AVX2 over AVX because it will provide more of a benefit.

The question is when?

And will it be within the first year of release or are you going to be waiting for broadwell before you start seeing it used in alot of apps.

Well, none of us can predict the future, but I'd put my money on it already having at least minimal support the day that Haswell launches/asap after it launches. That will be in the high dollar software like AutoCAD, and it's competition, of course. I highly doubt that people like you and I will see any benefit from it for at least a year or two, if not quite a long time after that. Although I am guessing that Adobe's next version of Photoshop will include support for it. Then again, I don't buy Photoshop, and you most likely don't either.

 

[moonbogg]

Well, none of us can predict the future, but I'd put my money on it already having at least minimal support the day that Haswell launches/asap after it launches. That will be in the high dollar software like AutoCAD, and it's competition, of course. I highly doubt that people like you and I will see any benefit from it for at least a year or two, if not quite a long time after that. Although I am guessing that Adobe's next version of Photoshop will include support for it. Then again, I don't buy Photoshop, and you most likely don't either.

I have seen this too many times to not comment on it. The majority of "CAD" work is single threaded. There, thats it. The only time more cores does anything is if you do a photo realistic rendering or perhaps a physics simulation of a part or assembly, which is done, but not as often as the "CAD" work. The actual "CAD" part of a "CAD" program doesn't give a crap how many cores you got and it won't help. So please, if you happen to be a "CAD" user and want a better "CAD" experience, upgrade your video card to a Quadro or firepro GL.

 

[Sheep221]

I have seen this too many times to not comment on it. The majority of "CAD" work is single threaded. There, thats it. The only time more cores does anything is if you do a photo realistic rendering or perhaps a physics simulation of a part or assembly, which is done, but not as often as the "CAD" work. The actual "CAD" part of a "CAD" program doesn't give a crap how many cores you got and it won't help. So please, if you happen to be a "CAD" user and want a better "CAD" experience, upgrade your video card to a Quadro or firepro GL.

CAD programs are still more likely to render with CPU instead of GPU, while there are CAD programs which are multi-threaded as well. The DDS-Solidworks is one of the more known programs to use multiple threads.

 

[moonbogg]

CAD programs are still more likely to render with CPU instead of GPU, while there are CAD software which are multi-threaded as well. The DDS-Solidworks is one of the more known programs to use multiple threads.

I use solidworks every day. Guess what functions of the program are multithreaded? Things like the SAVE command, etc. Seriously. The important part, that is, the actual design work, is single threaded.

 

[phenomkid7]

Funny that 8350 has been constantly sold out at newegg. So much for not going so well for AMD. I'm willing to say that AMD will exceed their sales expectation with Vishera line,far outselling FX8/6/4 series this and next year.

Yes because Newegg == the entire market for chips.

 

[tweakboy]

I'm sure there are enthusiasts who can handle a high clocked 8 core Sandy Bridge.

Like me. I get 45c when its not a hot hot sunny day on full load @ 1.384v

with that voltage I can take a SB E to 4.7Ghz then go 1.40v 4.8Ghz stable ?

 

[Haserath]

CPU A=120 seconds
CPU B= 12 seconds

A/B= 1000% faster(magnitude of 10)
And
A-B=C

C/A= 90% less time

So...

116/91.5= 26.8% faster

And

116-91.5

24.5/116= 21.1% less time