AMD Ryzen 7 2700X and Ryzen 5 2600X

[CAT-THE-FIFTH]

That's an interesting question, because looking at that set of benchmarks the 2700X trounces the 8700K at 1080p compared to what is shown here. Either way I'm also wondering if AMD has held back a 2800X deliberately while waiting for the standard Intel response.

Great looking CPU for the price though, I've really got to think about upgrading, memory and gpu pricing is making me hold off though.

Some of the improvements are really large though.

DAW performance has improved massively over Ryzen 1:

https://techreport.com/review/33531/...pus-reviewed/7








The Ryzen 5 2600X is now consistently ahead of the Core i5 8400,and looking at some of those scores,I suspect you would need a Core i5 8600K probably overclocked to match it.

Also,in FO4 which is probably one of the worst games for Ryzen,even if Intel is still ahead a 14% to 16% increase in minimums is really a big improvement,especially as Bethesda has not patched the game AFAIK for Ryzen.

[Xlucine]

So once precision boost overdrive hits (at some unspecified date), we could see even more performance under load?

The 6C/12T 2600X basically matches the 6C/12T 8700k in every other benchmark on that page, apart from the slightly weird and unrepresentative PiFast holdout (unless someone actually wants a CPU to run PiFast on). That's despite the clocks being higher on the Intel part, if anything.

I'm not saying it's not fine to include old synthetic benchmarks, but I think far too much attention is paid to it as the single threaded Hexus benchmark, even in conclusions. It's probably about as representative of real-world performance of any of the Geekbench sub-benchmarks for example.

Combined with the laptop results (matching intel core-for-core and watt-for-watt), I'm starting to suspect intel chips come with pi preloaded to 3m decimal places to they can just regurgitate it in benchmarks. The 8700k and 2600X have the same nominal base clock, so either precision boost blows turbo boost out the water or something isn't right with pifast

[scaryjim]

... Combined with the laptop results (matching intel core-for-core and watt-for-watt), I'm starting to suspect intel chips come with pi preloaded to 3m decimal places to they can just regurgitate it in benchmarks. The 8700k and 2600X have the same nominal base clock, so either precision boost blows turbo boost out the water or something isn't right with pifast

Three things really.

1) Intel's top turbo bin is way over AMD's - the 8700k has a 10% boost clock advantage over the 2700X, and almost 20% over the 2600X. It also has a pure IPC advantage. It should be winning purely single-threaded benchmarks hands down.

2) Precision Boost 2 is pretty phenomenal. From page 3 of this review, discussing the 2600X result: "With good cooling the new Precision Boost 2 algorithm keeps the all-core speed at 4.1GHz". That's only 100MHz down from the maximum boost clock...

3) AMD's SMT implementation is more efficient than Intel's.

Once you add 2 and 3 together - higher all core clocks and more efficient SMT - it's not surprising that they can make up the IPC difference once you apply a heavily threaded load. AMD's processors are built to be threading monsters.

[watercooled]

Three things really.

1) Intel's top turbo bin is way over AMD's - the 8700k has a 10% boost clock advantage over the 2700X, and almost 20% over the 2600X. It also has a pure IPC advantage. It should be winning purely single-threaded benchmarks hands down.

PiFast is an outlier, other single threaded benchmarks e.g. Cinebench don't show that big of a difference and more in line with boost clock differences as you'd expect.

https://www.tomshardware.co.uk/amd-r...-34307-10.html

Take some of the single-threaded numbers from this article:
Lame is actually faster on the 2700X than the 8700K.
POV-Ray is faster, but only 536 vs 622 (lower is better) so ~16% faster
Cinebench is faster, 197 vs 181 (higher is better) so ~9% faster

PiFast is a whole 44% faster

y-cruncher is understandably faster due to its heavy use of Intel's wider AVX2 units, and unless I'm very much mistaken, PiFast is a pretty old benchmark and far predates AVX2 instructions.

[scaryjim]

Cinebench seems well optimised for Ryzen, interestingly. POVRay puts the 8700k > 15% ahead of the 2700X despite a < 10% boost clock difference. iirc PiFast uses one of the older floating point instructions (SSE? MMX even?! plain x87!?!) that Intel either still supports in hardware or has particularly good emulation of.

And that's kind of the issue - IPC isn't a very good term for general CPU performance as the type of instruction matters. ISTR that Cinebench 11.5 gives Intel a bigger lead than R15 does, for instance. Guru3D's 2700X review does an "IPC" test, but it only uses Cinebench R15, which seems to use instructions that favour Ryzen. There is no perfectly representative benchmark out there.

PiFast has one big advantage - you can use it to compare > 15 years of processors via Hexus reviews. And ultimately it tells us exactly what we already know - Intel are still faster in a straight line, as demonstrated over a wide range of ST benchmarks across dozens of reviews. The actual degree of difference is largely academic, as there are very few real world programs that are heavily ST bound now....

[lumireleon]

With such numbers we will soon hear of Intel Partner Program (IPP). ~ What you're seeing is a $229 processor keep up with a rival $349 one.

[DanceswithUnix]

PiFast has one big advantage - you can use it to compare > 15 years of processors via Hexus reviews. And ultimately it tells us exactly what we already know - Intel are still faster in a straight line, as demonstrated over a wide range of ST benchmarks across dozens of reviews. The actual degree of difference is largely academic, as there are very few real world programs that are heavily ST bound now....

In a world of 64 bit processors which on the PC mean a minimum of SSE2, I don't think PiFast tells us anything because I *think* it is 32 bit and I *think* it is x87 fpu but therein lies a massive problem, I have tried googling in the past for some description of this code and come up blank so I can't *know* what it is doing.

x87 was never a good idea even when new; whoever came up with the idea of doing floating point on a stack was an idiot.

All useful modern PC workloads are AMD64 instruction set using recent SSE/AVX instructions. I either need a proper analysis of what the code is doing, the ability to build it myself so I can do that analysis myself, or the benchmark should be thrown out for the worthless fossil that it is.

POVray would seem a way better benchmark **if* you are interested in floating point throughoput. Intel is very strong on floating point and have been for some time almost to an absurd level, so an i7 should rule the roost on POVray.

But for most people, single threaded really means all those badly written lightly threaded games. Cat's example of FO4 seems a good one, it sucks less on Intel systems. WoW has always been similar though getting better over time. Other games aren't so bad, Elite for example on my machine sees 8 cores all with some spare capacity but all well utilised so it can be done.

Personally I just checked the Phoronix benchmarks for kernel code compile times because that's pretty close to my workload.

[lumireleon]

.....All useful modern PC workloads are AMD64 instruction set using recent SSE/AVX instructions.....Is Intel and many others paying AMD in royalties for using their X64 instruction set? or how does the dark world of instruction sets work?

[3dcandy]

Is it just me because it appears that AMD have come good with a great cpu, great supporting chipsets et al and other things like apathy and no need and other component prices mean this is all just a bit meh!

[DanceswithUnix]

Is Intel and many others paying AMD in royalties for using their X64 instruction set?

Intel and AMD have extensive cross license agreements in place, so they don't pay each other for CPU tech.

What the likes of VIA pay these days I have no idea, but Intel pretty much squashed them like a bug years ago so it doesn't much matter.

Besides, Intel are only using AMD64 because Microsoft forced them to. If Intel had their way, we would all be paying through the teeth for Itanium instruction set machines by now with no chance of a second supplier for competition.

[azrael-]

.....All useful modern PC workloads are AMD64 instruction set using recent SSE/AVX instructions.....Is Intel and many others paying AMD in royalties for using their X64 instruction set? or how does the dark world of instruction sets work?

Intel and AMD have extensive cross license agreements in place, so they don't pay each other for CPU tech.

What the likes of VIA pay these days I have no idea, but Intel pretty much squashed them like a bug years ago so it doesn't much matter.

Besides, Intel are only using AMD64 because Microsoft forced them to. If Intel had their way, we would all be paying through the teeth for Itanium instruction set machines by now with no chance of a second supplier for competition.

Also, we shouldn't forget that Intel's processors do not use AMD64. They use EMT64T, conceived and developed by Intel and only coincidentally 100% compatible with AMD64.

(The previous statement may contain traces of sarcasm.)

[Phage]

Confused: is the Precision Boost 2 a function of the chip or the X470 board ?

[watercooled]

Cinebench seems well optimised for Ryzen, interestingly. POVRay puts the 8700k > 15% ahead of the 2700X despite a < 10% boost clock difference.

Which is more in line with reality - there might be a difference with some applications favouring the way one uArch works over the other, but it's not nearly the vast chasm that PiFast shows.

iirc PiFast uses one of the older floating point instructions (SSE? MMX even?! plain x87!?!) that Intel either still supports in hardware or has particularly good emulation of.

So it's not really representative then if it's using a deprecated codepath where something as simple as different compiler flags could change everything.

And that's kind of the issue - IPC isn't a very good term for general CPU performance as the type of instruction matters. ISTR that Cinebench 11.5 gives Intel a bigger lead than R15 does, for instance. Guru3D's 2700X review does an "IPC" test, but it only uses Cinebench R15, which seems to use instructions that favour Ryzen. There is no perfectly representative benchmark out there.

Exactly my point. My issue is that PiFast is being used as the sole single-threaded performance test and even the conclusions reference it as a standard for single-threaded performance which is misleading at best.

PiFast has one big advantage - you can use it to compare > 15 years of processors via Hexus reviews.

Which is fine from an academic point of view.

And ultimately it tells us exactly what we already know - Intel are still faster in a straight line,

It really doesn't.

as demonstrated over a wide range of ST benchmarks across dozens of reviews. The actual degree of difference is largely academic, as there are very few real world programs that are heavily ST bound now....

To a point, but the difference between this and the relatively minor (sometimes negligible) difference shown in other benchmarks is more than academic.

Please don't get me wrong, I'm not just picking on this because it shows Intel CPUs as being better than AMD's CPUs, it's the sheer scale of the difference which is unlike so many real-world applications that it borders on being meaningless.

In a world of 64 bit processors which on the PC mean a minimum of SSE2, I don't think PiFast tells us anything because I *think* it is 32 bit and I *think* it is x87 fpu but therein lies a massive problem, I have tried googling in the past for some description of this code and come up blank so I can't *know* what it is doing.

I've found that myself. I believe SuperPi made use of mostly x87 and IIRC PiFast was a substantial improvement over that, but there's really nothing out there (that I can find) about what sort of code it's made up from.

All useful modern PC workloads are AMD64 instruction set using recent SSE/AVX instructions.

y-cruncher is an example of a program to hugely improve performance of other Pi calculating applications and includes several binaries and a dispatcher for different uArchs. The latest versions make good use of the wider FMAs with AVX2 and show substantial improvements for modern Intel uArchs; an area where Zen doesn't manage the same throughput for obvious reasons. The results from this benchmark at least makes sense.

POVray would seem a way better benchmark **if* you are interested in floating point throughoput. Intel is very strong on floating point and have been for some time almost to an absurd level, so an i7 should rule the roost on POVray.

Again, it's modern and the results actually make some sense.

[3dcandy]

Confused: is the Precision Boost 2 a function of the chip or the X470 board ?

Chip - but the boards are better balanced to help with power delivery etc. You can also overclock on x470 Putting a Ryzen 2 in an x370 board is still fine though

[CAT-THE-FIFTH]

Confused: is the Precision Boost 2 a function of the chip or the X470 board ?

Both - you need a Ryzen 2 CPU and a 400 series motherboard for it to work in,as the 400 series boards on average apparently can deliver more current for short periods.

[Phage]

Hmmmm. I won't bother if that's the case. Is there a definitive answer to this ?

Just found this

https://youtu.be/YSMXbbw2B8Y