Sorry, fell for it again . One of us should have started a new thread on Intel's financials in expectation that it would start and AMD vs. Intel debate.
What's with all this Ryzen garbage in the Intel thread?
This whole AVX512 support favoring Intel story is a bit of overhyping, IMO. Think of applications that are CPU FPU-bound - linear algebra(LINPACK), signal processing(FFTs), data analysis(lots of math functions) - these are the things that would obviously benefit from wider vector SIMD registers. But there are a lot of other workloads that are mostly I/O bound, especially with large data sets. Eulerian CFD, FEM analysis, basically most of the things that involve solving differential equations are I/O bound.
There is a point after which one has to really think, are all these vector computations done on the CPU actually worth the trade-offs in heat and power consumption, due to fitting bigger and bigger registers on the CPU for a specialized task like vectorized data processing? At that point, the logical thing to do would be to move to things like Xeon Phi or GPUs.
None of the things I mentioned can be termed as 'regular software'. Encoding benefits from AVX because it uses a lot of techniques that I mentioned are FPU-intensive. Vectorizing your code is the only way to extract the best out of those special registers that you have on your CPU, compilers can only do so much. Writing vectorized code, and identifying the parts of code that can be accelerated through AVX is a non-trivial task, and is beyond the realm of capabilities of your average user.We need to check instruccions and registers, but the whole point of AVX-512 on Xeons is to integrate the capabilities of the Phi Coprocessor.
Not sure how gonna impact on regular software... maybe encoders?
TSX didn't work on Haswell. And neither did it on Broadwell. Not easy to get it right it seems.
This whole AVX512 support favoring Intel story is a bit of overhyping, IMO. Think of applications that are CPU FPU-bound - linear algebra(LINPACK), signal processing(FFTs), data analysis(lots of math functions) - these are the things that would obviously benefit from wider vector SIMD registers. But there are a lot of other workloads that are mostly I/O bound, especially with large data sets. Eulerian CFD, FEM analysis, basically most of the things that involve solving differential equations are I/O bound.
There is a point after which one has to really think, are all these vector computations done on the CPU actually worth the trade-offs in heat and power consumption, due to fitting bigger and bigger registers on the CPU for a specialized task like vectorized data processing? At that point, the logical thing to do would be to move to things like Xeon Phi or GPUs.
That the TDP numbers are fairy tale numbers. 95W TDP that uses 120-130W etc.
Just double confirmed this bit of information with a source:
Skylake-SP LCC = 10C/20T
Skylake-SP MCC = 18C/36T
Skylake-SP HCC = 28C/56T
This means the Skylake-X 12C SKU for LGA 2066 is based on the MCC server die. The new HEDT platform (Basin Falls) is able to accommodate 14C, 16C and 18C beasts, if Intel eventually chooses to release them for desktop users.
The chart is system power, though. Not sure how accurately we can deduce CPU power from that.Damn, Intel lied about the TDP of the 7700k - it's supposed to be a 91W part
Well it started with this quote "If the pretty strong rumors are true, Intel will be up against 16 core/ 32 thread Ryzen cpu's not AMD'S mainstream platform. AFAIK, Skylake-X will be offering a top end 12 core to compete with AMD's hedt."What's with all this Ryzen garbage in the Intel thread?
$400 for a 6c/12t CPU will loose to a Ryzen 8c/16t@$330Again, that is not true, unless AMD wants to put the 16C cheaper than a 1800X, the 6C Skykake X starts at about the same price range as the 6800K, 400.
Skylake-X 6C/12T will compare very farourably to any Ryzen part as far as performance is concerned. This SKU has the potential to offer better performance per clock than a regular Skylake/Kaby Lake/Coffee Lake, with the added advantage of higher overclocking potential than the competition (4.5 GHz+ should be doable). And as far as mainstream platforms are concerned, a possible Coffee Lake-S 6C/6T part replacing Core i5-7600K + cheap compatible LGA 1151 should become the new gamer's favorite.
As far as this thread, it is pertinent to post about what these new CPU's will be up against, but the Intel fanboys will not listen, they just keep spouting the same old crap. If we could have a reasonable discussion about the situation, it would be more enlightening.
So an Intel 14c/28t CPU can barely beat a 8c/16t Ryzen, but you are trying to convince everyone that a NEW 6c/12t will compare. Based on what ? No benchmarks ? Just speculation, so don't say "WILL", thats trolling. "I would guess" or more appropriate.Skylake-X 6C/12T will compare very farourably to any Ryzen part as far as performance is concerned. This SKU has the potential to offer better performance per clock than a regular Skylake-S, with the added advantage of higher overclocks than the competition (4.5 GHz+ should be doable). And as far as mainstream platforms are concerned, a possible Coffee Lake-S 6C/6T part replacing Core i5-7600K + cheap compatible LGA 1151 should become the new gamer's favorite.
So an Intel 14c/28t CPU can barely beat a 8c/16t Ryzen, but you are trying to convince everyone that a NEW 6c/12t will compare. Based on what ? No benchmarks ? Just speculation, so don't say "WILL", thats trolling. "I would guess" or more appropriate.
and AVX-512 for people who might be able to make use of that (video encoding?)
Ryzen 8C barely beats Kabylake 4C, with decent clock speeds Intels 6C on Skylake uArch will be the better choice overall.
Ryzen 8C barely beats Kabylake 4C, with decent clock speeds Intels 6C on Skylake uArch will be the better choice overall.
$400 for a 6c/12t CPU will loose to a Ryzen 8c/16t@$330
So, what is not true ?
And my "guesses" as based on facts about todays Intel and AMD chips.
You're thinking gaming. I don't think that's what the other guy is referring to.
Even though I am a strong AMD supporter I think the first gen Ryzen 7 which maxes out at 4-4.1 Ghz will probably lose to a mainstream Coffeelake 6C/12T CPU built on Intel 14++ process (12% faster transistor performance than 14+ which Kabylake is built on) even for multithreaded workloads as I expect a 5.2 Ghz OC easily given the significantly higher transistor performance..