News for hackers. Stuff that matters.
Re:Intel 9th Gen Review by IvyMike on Friday October 19, 2018 @02:32PM (#18258871)
I'm comparing the i7 here to the i7 in my current computer, built in April 2011(!)

https://ark.intel.com/compare/186604,52214

Kinda though I would see a lot bigger differences. There was a time in my life when I expected I would have to build a new computer every two years. Weird.

Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @02:38PM (#18258915)
The i7-2600k doesn't even have 256-bit busses on the inside. The i7-2600k barely begun to implement AVX (and doesn't even implement AVX2).

The i9-9900k is way better. 3x 256-bit execution ports, 64-byte (512-bit) messages going around the Ringbus, DDR4 RAM (double memory bandwidth).

The raw clocks ignore the huge advancements in IPC that exist between Sandy Bridge and Coffee Lake. And even then, the 5GHz clock on the i9-9900k makes it completely incomparable to compare the ancient Sandy Bridge with the modern Coffee Lake.

Re:Intel 9th Gen Review by coldtea on Friday October 19, 2018 @03:18PM (#18259215)
>The i9-9900k is way better. 3x 256-bit execution ports, 64-byte (512-bit) messages going around the Ringbus, DDR4 RAM (double memory bandwidth).

All those don't matter if they hardly make a 10-15% difference in everyday use, and require specialized code to take advantage of.

Re:Intel 9th Gen Review by celrod on Friday October 19, 2018 @04:46PM (#18259925)
Much of what I do is vectorized, so the more fma units and the side wider, the faster. So avx512 > avx2 with two 256 bit units > avx2 with two 128 bit units.

But my work (statistics) seems poorly represented in benchmarks and the community at large. Ryzen is in the later group, and gets rave reviews.

While most press I've seen talking about avx512 has been negative. Yet I get a lot more flops out of the 10-core i9 7900x than the similarly priced 16-core Threadripper 1950x in simulations.

A big part of that is probably doing most of my work in Julia, where because the code you run is compiled locally the first time a function/argument type combination is called (a) code will be generated for all your CPU's features, and (b) it is easy to encourage aggressive optimizations to vectorized code by the compiler, eg by parameterizing arrays by their bounds.

R benchmarks will not see benefits at all, as much of R's math libraries are terribly unoptimized -- the default BLAS doesn't even use any runtime dispatch! (Thankfully it's easy to change, but the benchmarks all reflect that reality -- and most people's personal experiences too).

Similar likely applies to many NumPy benchmarks, but depending on the distributor of those binaries, the situation likely isn't better. Either generic unoptimized binaries, or shipping with something like Intel MKL which, while the fastest linear algebra library for Intel processors, dispatches to slow code paths for AMD processors -- a very unfair choice for benchmarking. I'm less less familiar with NumPy though, so perhaps there's a booth fair and well optimized distribution.

I fear the future will move against my use cases.

Re:Intel 9th Gen Review by marmaduke on Friday October 19, 2018 @05:53PM (#18260353)
If you want fat vector units you should be on a GPU.
Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @06:01PM (#18260403)
Yes, in principle. A dedicated GPU will always be "fatter" in SIMD than a CPU.

But a CPU has latency benefits. Most noticeably, it takes only a few clock cycles for an Intel CPU to transfer data from its RAX register -> AVX vector registers, compute the solution, and transfer it back to the "scalar world".

This level of data-transfer is done within single-digit nanoseconds.

In contrast, any GPU memory transfer takes hundreds-of-nanoseconds to microseconds... an order of 100x to 10,000x slower than transferring between Scalar-code and AVX-code on a CPU.

----------

So GPUs are good for bulk compute (big matrix multiplies, Deep Learning, etc. etc.)... but I'm definitely interested in the low-latency uses of SIMD.

For example: Consider a SIMD-based Bloom Filters to augment a Hash Table (or any data-structure, really). Multi-hash systems like Cuckoo-hashes have also been successfully implemented with CPU-SIMD acceleration.

You can use CPU-accelerated SIMD to accelerate CPU-based algorithms. And CPUs certainly can benefit from fatter, and better designed, SIMD instruction sets.

Re:Intel 9th Gen Review by PeCaN on Friday October 19, 2018 @06:32PM (#18260581)
> it takes only a few clock cycles for an Intel CPU to transfer data from its RAX register -> AVX vector registers, compute the solution, and transfer it back to the "scalar world".

unfortunately, it takes millions of cycles for the CPU to switch to the higher-power state necessary for AVX instructions, so your latency really isn't that much better.

the radfft guy did a good writeup https://gist.github.com/rygorous/32bc3ea8301dba09358fd2c64e0...

Re:Intel 9th Gen Review by celrod on Friday October 19, 2018 @09:19PM (#18261167)
Yes. My code is mostly AVX, so regardless of down-clocking I see substantial performance boosts.

But I've found getting into GPU computing extremely difficult. Admittedly, part of that may be having bought an AMD Vega graphics card instead of NVidea (not a fan of vendor locking, and AMD is building an open source stack with ROCm/HIP).

My code does lots of different things, many small, but given a billion iterations, it can add up. If you're running a Monte Carlo simulation over millions of different data sets, fitting each iteration with Markov Chain Monte Carlo, with a model that has a few long for loops and requires (automatic) differentiation and some number of special functions... It's not hard to rack up slow execution times out of small pieces.

Part of the problem is I need to actually find a GPU project that's accessible for me, so I can gain some experience, and learn what's actually possible. How I could actually organize computation flow. It's easy with a CPU. Break up Monte Carlo simulations among processors. Depending on the model being fit, as well as the method, different computations can vary dramatically in width. But optimizing is as simple as breaking up all those computations into the appropriately sized AVX units (and libraries + compilers auto-vectorization will often handle most of that!), so wider units directly translate to faster performance.

Part of the problem is I don't really know how to think about a GPU. Can I think of the Vega GPU as a processor with 64 cores, a SIMD width of 64 (64 * 64 = advertised "4096), and more efficient gather loads/scatter store operations?

If there were a CPU like that, compiler + library calls and macros (including libraries you've written or wrapped yourself) go a long way so you can quickly write well optimized code.

I really need to dedicate the time to learn more. My 7900x processor is about 4x faster at gemm than my 1950x, but 10x slower than the Vega graphics card that cost less money. I see the potential.

To start, I need to find a GPU project that lets me really get experimenting and figure out how programs can even look.

Is there an Agner Fog of GPGPU?

Re:Intel 9th Gen Review by marmaduke on Saturday October 20, 2018 @04:50AM (#18262420)
If the MC simulations vary only in the values of parameters (as opposed to memory access or execution path through an if/else), then they are ideal for GPU. You write a kernel which looks like it’s doing one of those simulations, and call it on arrays on parameters, and you’re in business.

Concretely, I’ve used this for parameter sweeps of large systems of ODEs and SDEs.

Re:Intel 9th Gen Review by jabl on Saturday October 20, 2018 @03:44AM (#18262249)
Did you see https://news.ycombinator.com/item?id=18249651 about Gpu programming with Julia?

You still need some understanding of gpus, but at least you can write your code in Julia rather than CUDA or OpenCL.

Re:Intel 9th Gen Review by dnautics on Saturday October 20, 2018 @04:44AM (#18262401)
As a longtime user of Julia I'm convinced that the julia approach to gpus is wrong -- for being insufficiently Julian. I would love for the gpus to be registered as independent GPU nodes accessible by the Distributed module, to which you dispatch async compute tasks.
Re:Intel 9th Gen Review by scns on Saturday October 20, 2018 @01:39AM (#18261927)
There was an Intro to gpuprogramming in Julia post on hn one or two days ago
Re:Intel 9th Gen Review by dragontamer on Saturday October 20, 2018 @01:33AM (#18261918)
> Part of the problem is I don't really know how to think about a GPU. Can I think of the Vega GPU as a processor with 64 cores, a SIMD width of 64 (64 * 64 = advertised "4096), and more efficient gather loads/scatter store operations?

From my limited experience, that's the wrong way of looking at things.

SIMD is good at looking at "converged" instruction streams, and bad at "divergent" instruction streams.

"Converged" is when you've got your bitmask (in AMD / OpenCL: the scalar bitmask register) executing all 64-threads as much as possible. This means all 64-threads take "if" statements together, they loop together, etc. etc.

"Diverged" is when you have a tree of if-statements, and different threads take different paths. The SIMD processor has to execute both the "Then" AND the "Else" statements. And if you have a tree of "if / else" statements, the threads have less-and-less in common.

-------

That's it. You try to group code together such that as much code converges as possible, and diverges as little as possible. It helps to know that the work-group size on AMD can be anywhere from 64 to 256 (so keep your "thread groups" together as large as 256 at a time).

The OpenCL compiler will automatically translate most code into conditional moves and do its best to avoid "real if/else branches". But as the programmer, you just have to realize that nested if-statements and nested-loops can cause thread-divergence.

---------

Matrix maths are the easiest for SIMD, because of how regular their structure is. When you get to "real" cases like... well...

> My code does lots of different things, many small, but given a billion iterations, it can add up. If you're running a Monte Carlo simulation over millions of different data sets, fitting each iteration with Markov Chain Monte Carlo, with a model that has a few long for loops and requires (automatic) differentiation and some number of special functions... It's not hard to rack up slow execution times out of small pieces.

Okay, well that's why its hard to program properly on a GPU. Because they're all diverging. Unless you figure out a way to "converge" these if statements, it won't run on a GPU correctly.

Chances are, if you have a million-wide Monte-carlo, a lot of those threads are going to be converging. Can you split up the steps and "regroup" tasks as appropriate?

Lets say your million threads run into a switch statement:

* 106,038 threads will take path A.

* 348,121 threads take path B.

* 764 threads take path C.

Etc. etc.

Can you "regroup" so that all your threads in pathA start to execute together again? To best take advantage of the SIMD-architecture?

Think about it: a gang of 64 may have thread #0 take PathA, thread#1 take PathB, thread #2 take PathA, etc. etc. So it all diverges and you lose all your SIMD.

But if you "rebatch" everything together... eventually you'll get hundreds of threads ready for PathA. At which point, you gang up your group of 64 threads and execute PathA all together again, taking full advantage of SIMD.

SIMD is an architecture that allows "similar" threads to run at the same time, in groups of 64 or more (up to 256 threads at once). And they truly execute simultaneously as long as they are all taking the same if/else branches and loops. The real hard part is designing your data-structures to maximize this "convergence" of threads.

Something like Chess would practically be impossible to SIMD on a GPU. Its just impossible to handle how divergent the typical chess analysis engine gets due to the precision of chess board positions. But something like Monte-Carlo surely will have hundreds, or thousands, of threads taking any "particular execution path". So I'd have hope for a SIMD-based execution on Monte-Carlo simulations.

Re:Intel 9th Gen Review by Filligree on Friday October 19, 2018 @06:31PM (#18260571)
Using AVX instructions clocks down your CPU by a lot, though. It isn't usually worth it unless you're running mostly AVX.
Re:Intel 9th Gen Review by esmi on Saturday October 20, 2018 @03:47PM (#18265051)
> So GPUs are good for bulk compute (big matrix multiplies, Deep Learning, etc. etc.)

Maybe. Sparsity of the matrix in question matters a lot. Matrixes with very little sparsity, like say an image, do well. Others may not. It’s more like GPUs do well on algorithms with predictable branching.

Re:Intel 9th Gen Review by ip26 on Friday October 19, 2018 @10:35PM (#18261408)
People aren't benchmarking avx512 because very little is coded to use it yet. You are a small niche.

The future will probably include more avx512, but not yet.

Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @04:48PM (#18259947)
If you're running anything CPU-intensive these days, you are definitely going to use those AVX cores.

Video Editing, Video Games, Graphics, 3d Modeling, Photoshop. Even Stockfish Chess uses new instructions (not SIMD: but the Bit-board popcnt and pext / pdep instructions) to accelerate chess computations.

At a bare minimum, AVX grossly accelerates memcpy and memset operations. (Setting 256-bits per assembly instruction instead of 64-bits per operation is a big improvement). And virtually every program can benefit from faster memcpy and faster memsets.

"Standard Software" isn't written to be very fast. But anything that's even close to CPU-bound is being upgraded to use more and more SIMD instructions.

Re:Intel 9th Gen Review by Keyframe on Friday October 19, 2018 @06:13PM (#18260480)
In theory, yes. I was looking to build a new rig, I do a lot of video. I'm running a 2697v2 and 3930k in another (both built around the same time, around ~5-6 years ago). To be honest, I can't really tell much of a difference in using those for those heavy video editing and fx work vs threadripper and newer Xeons I'm using at other places. I'm not saying they're not faster, but there's not much of a noticable, if at all, difference in using those machines.
Re:Intel 9th Gen Review by the8472 on Friday October 19, 2018 @05:45PM (#18260301)
> At a bare minimum, AVX grossly accelerates memcpy and memset operations.

Not necessarily, for large operations and depending on processor generation a simple rep stos/movsb will be simpler (no alignment requirements) and saturate your memory bandwidth just as well as any AVX sequence will with less icache pressure.

Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @06:06PM (#18260439)
I wasn't aware of ERMSB "Enhanced Rep MOVSB". Thanks for the tip.

Seems to be a feature in Ivy Bridge and later, which happens to be around the time AVX2 started.

Re:Intel 9th Gen Review by eloff on Friday October 19, 2018 @06:29PM (#18260565)
To be fair, it's issuing a long sequence of microcode ops under the hood using 256bit ops.
Re:Intel 9th Gen Review by CamperBob2 on Friday October 19, 2018 @05:20PM (#18260154)
At a bare minimum, AVX grossly accelerates memcpy and memset operations. (Setting 256-bits per assembly instruction instead of 64-bits per operation is a big improvement). And virtually every program can benefit from faster memcpy and faster memsets.

How often are memcpy and memset CPU-bound, though?

Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @05:34PM (#18260227)
Whenever its in L1 or L2 caches, which on Intel, have 64-byte-per-clock bandwidth. (96-byte-per-clock bandwidth to L1 cache, 64-byte-per-clock theoretical L2 bandwidth... dropping to ~29 sustained)

L3-cache on Skylake architectures has 18-bytes-per-clock sustained, which is still greater than 128-bits per clock bandwidth to your L3 cache.

Soooo... I'd say roughly for any memset or memcpy smaller than 256kB or so... you're going to benefit to an AVX-based memset or memcpy.

At ~8MB or so, where you're hitting L3 cache, it probably is a benefit but not much of one. The important thing is Skylake only has one store unit, so you can only write once per clock cycle.

So... do you write one 64-bit value, or do you write one 256-bit (32-byte) value?

----------

To be fair: I'm pretty sure every compiler's default settings today outputs SSE-based memcpy and memsets (128-bit). So AVX doubles that to 256-bit.

Re:Intel 9th Gen Review by keldaris on Friday October 19, 2018 @07:20PM (#18260780)
The issue - and one that a compiler often can't really reason about very well - is that in using AVX (and especially AVX-512) you incur other performance costs due to throttling the CPU down to lower frequencies and having additional latency issues. Deciding when it's actually worth doing so for a memcpy is, in general, a very hard problem.
Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @07:27PM (#18260814)
This is the problem when I have a Threadripper can't can't see these issues, lol. One day I'll get a proper AVX512 machine and try this stuff for myself.

Still, it seems like the documentation elsewhere says that AVX 128-bit doesn't cause any clocking issues. So AVX512 (applied to 128-bit) should still lead to good speeds without any clocking problems.

Re:Intel 9th Gen Review by keldaris on Friday October 19, 2018 @08:17PM (#18260991)
Right, if you stick to 128 bit registers, you avoid most of the issues. Does that get you much benefit for a memcpy over an SSE implementation on a SKX CPU? I don't think so, but I haven't looked into that particular problem, so maybe I'm missing something obvious here.
Re:Intel 9th Gen Review by celrod on Friday October 19, 2018 @09:53PM (#18261262)
Simple test, `memcopy.c`:

  void memcopy(double* restrict a, double* restrict b, long N){
    for (long n; n < N; n++){
      a[n] = b[n];
    }
  }
You can check assembly with:

  gcc -march=skylake-avx512 -O2 -ftree-vectorize -mprefer-vector-width=128
   -shared -fPIC -S memcopy.c -o memcopy128.s

  gcc -march=skylake-avx512 -O2 -ftree-vectorize -mprefer-vector-width=512 
  -shared -fPIC -S memcopy.c -o memcopy512.s
to confirm they use `xmm` and `zmm` registers, respectively. (The 512 version also uses xmm to finish off a remainder. Ideally, it would use masked load/stores, but I haven't seen any auto-vectorizers actually do that.

I'm using gcc 8.2.1. I believe the option `-mprefer-vector-width` was added recently.

Now I compiled both into shared libraries (drop the `-S` and choose an appropriate file name), and benchmarked from within Julia.

  julia> using BenchmarkTools, Random

  julia> memcopy128!(a, b) = ccall((:memcopy, 
  "/home/chriselrod/Documents/progwork/C/libmemcopy128.so"),
  Cvoid, (Ptr{Cdouble},Ptr{Cdouble},Clong), pointer(a), pointer(b), length(a));

  julia> memcopy512!(a, b) = ccall((:memcopy, 
  "/home/chriselrod/Documents/progwork/C/libmemcopy512.so"), 
 Cvoid, (Ptr{Cdouble},Ptr{Cdouble},Clong), pointer(a), pointer(b), length(a));

  julia> b = randn(32); a = similar(b);

  julia> b = randn(32); a = similar(b);

  julia> all(a == b)
  false

  julia> memcopy128!(a, b); all(a == b)
  true

  julia> randn!(a); all(a == b)
  false

  julia> memcopy512!(a, b); all(a == b)
  true

  julia> @btime memcopy128!($a, $b)
    7.517 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    5.277 ns (0 allocations: 0 bytes)

  julia> b = randn(64); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    10.980 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    7.300 ns (0 allocations: 0 bytes)

  julia> b = randn(100); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    15.060 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    14.031 ns (0 allocations: 0 bytes)

  julia> b = randn(200); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    33.867 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    19.546 ns (0 allocations: 0 bytes)

  julia> b = randn(400); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    53.923 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    28.376 ns (0 allocations: 0 bytes)

  julia> b = randn(800); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    98.277 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    57.960 ns (0 allocations: 0 bytes)

  julia> b = randn(2_000); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    232.138 ns (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    133.229 ns (0 allocations: 0 bytes)

  julia> b = randn(20_000); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    3.688 μs (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    2.759 μs (0 allocations: 0 bytes)

  julia> b = randn(200_000); a = similar(b);

  julia> @btime memcopy128!($a, $b)
    112.598 μs (0 allocations: 0 bytes)

  julia> @btime memcopy512!($a, $b)
    117.437 μs (0 allocations: 0 bytes)
The advantage wasn't very impressive here, but it persisted for vectors of length 20,000. At 200,000 we saw the memory bottleneck hit in force, when it busts the per core L2 cache. The 7900x's per core L2 cache is 1048576, which translates to 131,072 doubles.
Re:Intel 9th Gen Review by keldaris on Saturday October 20, 2018 @05:25AM (#18262497)
This is a nice benchmark, thanks for doing it, but the point I was making is somewhat orthogonal to this.

Of course there's a performance advantage from using 512 bit registers for a memcpy - but a memcpy is rarely a major performance bottleneck by itself and is usually surrounded by other code. Unless that code is also AVX-512, you've just made it slower by optimizing the memcpy. My point was that a compiler can't usually decide whether it's worth making the optimization in light of the broader context.

The other point was whether using AVX-512 while sticking to xmm registers is faster than just using xmm SSE/AVX code. I don't have an AVX-512 capable machine at the moment, perhaps you'd like to check if your 128 bit version is any faster than just doing "gcc -march=skylake -O2 -mprefer-vector-width=128" (thereby retaining the microarchitecture optimizations, but sticking to AVX2)?

Re:Intel 9th Gen Review by spatulon on Saturday October 20, 2018 @06:51AM (#18262686)
Here are your C examples for easy viewing: https://godbolt.org/z/5j-QUj
Re:Intel 9th Gen Review by rys on Friday October 19, 2018 @05:09PM (#18260084)
I don’t know about the other software types, but games are not typical users of AVX (at client runtime anyway, the tooling used to make the games is different).
Re:Intel 9th Gen Review by blihp on Friday October 19, 2018 @03:16PM (#18259209)
If you roll up all of the 'huge advancements' (IPC, clock speed, all of it) over the course of 7 years, that's probably in the neighborhood of what we used to call the 'annual upgrade.'

Ancient? Yes. Incomparable? No.

Re:Intel 9th Gen Review by dragontamer on Friday October 19, 2018 @03:46PM (#18259450)
The Pentium 4 Netburst architecture was relatively stagnant between the years 2000 and 2004. "Prescott" (2004) was still "Netburst" but Prescott includes AMD64 so I'd say that's a major change.

Still, its reflective in the naming. Netburst, Willamette, Northwood, Prescott, and Tejas were all called "Pentium 4", and lasted from a time period from year 2000 to 2006.

Sure, there were a few additions: Hyperthreading I guess. But otherwise, the main difference in this timeframe were the 250nm to 90nm scaling (which got "free" GHz upgrades back then without any effort)

In short:

> what we used to call the 'annual upgrade.'

You're exaggerating. Progress has slowed but Willamette (year 2000 1.5GHz) vs Northwood (year 2002 2GHz) Pentium 4s weren't that dissimilar.

Prescott (2004) is even more amusing, as it hit 3.8GHz (Intel Pentium 4 570J). But boy, oh boy... you really can't compare Prescott 3.8GHz against Sandy Bridge (2011, 3.6GHz) or Coffee Lake (2018, 3.6GHz ), even if the base clocks are all close.

Re:Intel 9th Gen Review by blihp on Friday October 19, 2018 @03:59PM (#18259562)
And luckily we had AMD then, as we do now (only recently), to pick up the slack. Sure I'm exaggerating a bit, but no more than you are to say 'incomparable'. Intel upgrades for the last 7 years have been modest at best. If you need a new machine, sure they'll make a fine replacement. But hardly a reason to upgrade year over year (or even every 2-3 years) for performance and/or functionality alone.

I've been sitting on an old i7 Sandy Bridge precisely because Intel hasn't been offering me a compelling reason to upgrade. I'm probably going to finally do it next year... but it won't be to another Intel system.

Re:Intel 9th Gen Review by tracker1 on Friday October 19, 2018 @04:45PM (#18259915)
My 5yo i7-4790K is to this day very fast (gtx 1080, and nvme 2 years ago). May upgrade next year for more threads/vms though.
Re:Intel 9th Gen Review by ksec on Friday October 19, 2018 @11:50PM (#18261676)
1. Sandy Bridge to Skylake combined makes barely a double digit gain in IPC, ~roughly 14%. I hardly called that a "huge advancements" in IPC over 7 years time.

2. We have already proved should the SandyBridge Overclocked to 4GHz+ it would have perform faster in some cases than Skylake Uarch with same Clock Speed. Remember SandyBridge was 32nm.

3. "Netburst, Willamette, Northwood, Prescott, and Tejas " You made it sounds as if there were many generation. All of them are part of Netburst uArch and Willamette was the first Gen, Tejas was cancelled. So there is only "Willamette, Northwood, Prescott". Each with much improved Clock speed, better implementation of HT, better memory controller and cache read improvement, and finally AMD64, all translate to substantial performance improvement, regardless of its power usage.

Netburst may not be a good uArch, but it definitely brings many improvement, the writing was on the wall in 2004 and we got Core 2 in 2006. And we have been stuck with Quad Core Desktop CPU despite we have had over 4x transistor density improvement.

Re:Intel 9th Gen Review by mrb on Friday October 19, 2018 @10:52PM (#18261456)
«The i9-9900k is way better...»

You forgot the most important: i9-9900k has TWICE the number of cores as i7-2600k.

Re:Intel 9th Gen Review by emddudley on Friday October 19, 2018 @03:08PM (#18259139)
CPU speed improvements have slowed down compared to the early 00s, but there's still a big difference in performance between 2011 and today.

On CPU Benchmark the i7-2600K scores an 8449 [1], while the i7-8700K scores a 15970 [2]. The i7-8700K is the predecessor to the i7-9700K reviewed in the article.

[1] https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-2600K...

[2] https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-8700K...

Re:Intel 9th Gen Review by user5994461 on Friday October 19, 2018 @03:22PM (#18259238)
The real score is 1942 against 2704. Not even a 50% progress in almost a decade. It's disappointing.

The score you gave is the multi threaded benchmark that is just higher for having 6 cores rather than 4, that regular applications don't use.

Re:Intel 9th Gen Review by sireat on Saturday October 20, 2018 @06:48AM (#18262680)
Another point is that 2600k was good for OC and 4.8GHZ OC were pretty common(not sure about 5Ghz).

Newer intel CPUs turbo higher at stock (you could say that OC is sort of built in) so single threaded performance looks better than it really is.

Now 32nm at 4.8Ghz would be quite power hungry but that is not the issue here.

Pretty much all the progress Intel has achieved in the last 7 years since Sandy Bridge has been in achieving the same performance for less wattage focusing on mobile use cases.

(on server side Xeons have increased the core count substantially but again single threaded performance IPC has stagnated)

Thus mobile i5-8250U gets to claim that it does about the same in 15W that 2600k did at 95W.

Remember we are still on the same Skylake Arch from 3-4 years ago, just a few minor process tweaks on 14++++nm plus obviously 2x the number of cores.

Re:Intel 9th Gen Review by cix_pkez on Friday October 19, 2018 @04:07PM (#18259610)
Some do. But in those cases, you might be better served with Ryzen.

If you're playing games, you'll get better performance with this i9.

But if you're playing games, you're probably not CPU-bound anyway.

I'd be interested in what niches people here work in and what CPUs they look for because of their unique needs.

Re:Intel 9th Gen Review by tracker1 on Friday October 19, 2018 @04:46PM (#18259928)
In all the truly multithreaded benchmarks the Ryzen and ThreadRipper chips really punch above their weight (price).
Re:Intel 9th Gen Review by leetcrew on Friday October 19, 2018 @04:15PM (#18259674)
the people who care the most about frames per second are playing esports titles that don't demand much from the GPU. counterstrike is still CPU bound in most cases.
Re:Intel 9th Gen Review by EugeneOZ on Friday October 19, 2018 @11:53PM (#18261686)
Esport is not about graphics quality at all. Try Assassins Creed - it will eat all of your CPU cores.
Re:Intel 9th Gen Review by hoorayimhelping on Friday October 19, 2018 @05:07PM (#18260071)
>But if you're playing games, you're probably not CPU-bound anyway.

We used to say this, but I don't believe it anymore. A lot of games these days make heavy use of simulation and physics engines. Games like KSP that perform continuous physics calculations are absolutely CPU-bound.

Re:Intel 9th Gen Review by tomwilson on Friday October 19, 2018 @11:34PM (#18261601)
You will also be cpu bound often when going for extremely high refresh rate 1080p. E.g. 240hz.
Re:Intel 9th Gen Review by kevindqc on Friday October 19, 2018 @04:23PM (#18259736)
WoW is ST CPU-bound :( until next patch!
Re:Intel 9th Gen Review by xoa on Friday October 19, 2018 @04:26PM (#18259765)
Feature differences and bandwidth can still be a significant issue though, and could become more so as raw single thread general CPU performance increases slow down while the surrounding ecosystem continues to improve. More PCIe lanes, more special purpose instructions, more features like ECC and IOMMU support becoming universal standards would all be nice outcomes to see.

This could represent a real additional long term opportunity to AMD as well, because historically Intel has been pretty fond of creating profitable product line segmentation via only allowing certain features on certain chips. And they've been mediocre about boosting interconnect capacity as well. Perhaps AMD doesn't necessarily need to match them as well in single thread performance if they can get the absolute difference down to a few percent, but be much friendlier then Intel in terms of not artificially restricting useful extra CPU features? I wonder if that becoming an area of competition again could bring some nice energy back to the market?

It's been something that's personally making me reconsider at least. I used to stick to Intel more despite price differences because of their performance advantages, but now I look at the rest of the ecosystem and start to calculate adding a few GPUs, NVMe storage, 10G networking, possibly other accelerator/utility cards and future interconnects, and the PCIe lanes seem to get sucked up a lot faster then they used to. With Ryzen/Epyc AMD is offering a full 64 lanes it looks like vs Intel's 24 here, and even with the latest Xeon I think it's now up to 48 lanes from 32? The difference there isn't nothing if overall system performance starts to depend less on just the CPU then on the overall ability to move data around and keep various more special purpose accelerators fed.

Re:Intel 9th Gen Review by RowanH on Friday October 19, 2018 @03:21PM (#18259234)
My Overlocked 6600k vs Ryzen/1800x at stock speeds is night and day difference in "real world" performance. Even though general clock speeds are not that much different - the Ryzen machine is the one I want to use.

Desktop productivity improves dramatically with a bunch more cores. Multiple browser tabs, background services, code editors open, VM's open... you can throw a lot of things at the Ryzen machine before you start noticing it starting to slow down.

On the 6600 machine it doesn't take much before you notice it chugging.

Re:Intel 9th Gen Review by tracker1 on Friday October 19, 2018 @04:48PM (#18259941)
I'm curious about the amount of memory as well as the drive performance... if you went from 8GB on an HDD to 16+ with an SSD or NVME, huge difference.
Re:Intel 9th Gen Review by nine_k on Friday October 19, 2018 @04:39PM (#18259876)
Is the amount of RAM the same?
Re:Intel 9th Gen Review by FractalParadigm on Friday October 19, 2018 @04:59PM (#18260016)
Not the OP but I upgraded from 6600K -> 1800X and honestly they're not even in the same ballpark performance-wise. When you come across software that can take advantage of multiple threads it's not even a contest, the 1800X just walks away laughing.

One example I can think of off the top of my head is gaming and transcoding (Plex server) at the same time. Even with QuickSync enabled and running, transcoding HEVC->x264 would regularly stutter and drop frames while a game is running, while also affecting the games framerate. The 1800X (without hardware accelerated video transcoding) just plows through it like it's nothing. I've even had it transcode 10-bit HDR video while playing Forza Horizon 4, not a single loss of framerate in either the game or the video.

The extra cores/threads are nothing to scoff at when software can take advantage of them.

My upgrade consisted of only the board and CPU; same RAM, GPU, PSU, SSD, etc.

Re:Intel 9th Gen Review by RowanH on Friday October 19, 2018 @07:00PM (#18260703)
Same amount of ram, different speeds (2133 vs 3000). I attribute the performance largely to the additional cores... the machine has that processor power sitting ready and waiting.
Re:Intel 9th Gen Review by sangaya on Friday October 19, 2018 @06:33PM (#18260583)
Check out this benchmark comparison: https://cpu.userbenchmark.com/Compare/Intel-Core-i9-9900K-vs...

It tells you more than the spec sheet comparison.

Re:Intel 9th Gen Review by kbenson on Friday October 19, 2018 @07:06PM (#18260719)
It's better than indicated in GP, but as an indicator of how good modern CPUs are in comparison to older ones, this page page overstates the case. The older CPU is 4-core but 8-thread (hyperthreads), the newer one is 8-core and 8-thread (no hyperthreads). The multi-core benchmarks are much higher for the newer CPU because it has twice as many cores, so that is inflating the overall score quite a bit. That said, it's not as simple as halving that score, as hyperthreads don't scale quite the same way, so it depends on how the multi-core test is formulated.

Edit: Oh, wow, as taspeotis points out it's not even the same two CPUs in the original Intel Ark comparison (which I was working off of from memory when noting cores and threads). The i9-9700k is actually an 8-core and 16-thread (hyperthreads) CPU, so it is actually literally twice as many cores and threads. It's still better performance, but not nearly as lopsided as it even seemed before, given those resources.

Re:Intel 9th Gen Review by taspeotis on Friday October 19, 2018 @07:23PM (#18260794)
GP said “I'm comparing the i7 here to the i7 in my current computer” and you linked to an i9 vs i7 comparison...
Re:Intel 9th Gen Review by 0-_-0 on Saturday October 20, 2018 @05:26AM (#18262498)
53% single core clock improvement over almost 8 years.

Transistor count has increased from 1B to about 7B, which is roughly in line with Moore's law though...

Re:Intel 9th Gen Review by overcast on Friday October 19, 2018 @03:09PM (#18259141)
Still rocking my LGA1366 Motherboard from 2008 with a Xeon X5675 @ 4.0ghz that I "upgraded" from an I7920 for $90 on ebay , 16GB of RAM, SSD, and an nvidia 1070 as the gaming PC. Handles everything at 2560x1600 60fps+ ultra quality(AA generally not necessary). Really don't quite understand the need to go crazy on computer hardware these days. Upgrading your GPU every 2-3 years seems plenty.
Re:Intel 9th Gen Review by ragequitta on Saturday October 20, 2018 @06:53AM (#18262693)
There is no way sort of some magical overclocking that a 1070 runs everything at 60 fps+ at that res. It has trouble at 1440p with many games. Note the minimum fps (which is the key factor):

https://www.tomshardware.com/reviews/nvidia-geforce-gtx-1070...

Re:Intel 9th Gen Review by overcast on Saturday October 20, 2018 @11:49AM (#18263792)
I can provide ALL the proof you need on my Asus STRIX 1070. I run everything at 2560x1600 synced to 60fps on my 30" monitor with Ultra settings. Those benchmarks are all enabling AA, which is just wasteful and not required at such resolutions. So yes, with AA, below the 60fps target. But right now my library of 418 Steam games, all run rock solid at those framerates. Metro, Batman, BF1/4/5, Witcher, Alien Isolation, Fallout, GTA, Bioshocks, whatever I throw at it.

Though the point of this discussion was the CPU not being critical anymore. If you need more FPS, buy a more expensive video card, not a CPU.

Re:Intel 9th Gen Review by tinco on Saturday October 20, 2018 @04:36AM (#18262379)
I'm rocking the 920 with 6gb still. I have been hesitating to upgrade to the Xeon. I have always been so salty that I jumped in just before the EFI boat, I feel my motherboard is so archaic. Is the performance jump to the Xeon very noticeable?
Re:Intel 9th Gen Review by overcast on Saturday October 20, 2018 @11:52AM (#18263807)
I think the 920 was fine, but the Xeon granted two more cores, and higher overclocking frequency for me, plus it was only $90. So more future proofing, as games take more advantage of multithreading. Increasing your RAM definitely should be on the list though.
Re:Intel 9th Gen Review by Brockenstein on Friday October 19, 2018 @03:06PM (#18259128)
Well those stats don't really highlight most real-world performance differences. If your expectation is that performance is tied to clock speed and you wanted to see 8 Ghz by now, then sure that's a disappointment, that's not reality though.

And besides what we're seeing now and why CPU's are viable for much longer is they got extremely powerful and user demands for more power win general computing kind of tapered off. I mean nowadays you can run multiple programs, a browser with dozens of tabs, multiple other utilities and background task pretty easily. You can run way more stuff then a person can reasonably multi-task. You have to get into some pretty specific workloads to really utilize all the power a modern CPU has and be wanting more.

And yeah the performance difference between generations has been pretty meager, but over 5, 6, 7 generations those modest improvements do add up. Going form a 2600k to a 8700k or 9700k is a big leap. But if you're not desperate for a lot more power you just might not care about 50% more performance.

Re:Intel 9th Gen Review by wil421 on Saturday October 20, 2018 @07:06AM (#18262709)
I have a Kaby Lake i7 and my last build was a i7-920 (Bloomfield). The old chip is a beast but the power consumption is a lot more. One thing I use is Intel QSV and the intel graphics chip. I don’t need a GPU at this time and it’s nice to have the option.
Re:Intel 9th Gen Review by Quequau on Saturday October 20, 2018 @06:45AM (#18262678)
I am currently using an early 2008 MacPro and I too had a time in my life when I expected I would have to build a new computer every two years.
Re:Intel 9th Gen Review by lotyrin on Saturday October 20, 2018 @07:30PM (#18266017)
What are you comparing, because 50% more cache and double the memory and PCIe bandwidth is actually a really big deal?