https://gifimage.net/wp-content/uploads/2018/11/eric-cartman-sweet-gif-8.gif

I'm happy with my b450 + r5 3600x until the next socket ( am4 + ddr5 ) appears.

Alessio1989:

games on cpu side (physics, IA, or just simple things like camera projections , etc), every multimedia software like enconding and decoding audio, pictures and videos, scientifics software.. oh the calculator.. yeah very limited, specially since the x87 FPU is no more used outside legacy 32-bit software or ad hoc intrinsics calls on 64-bit software. PS: the SIMD instructions cover both FP and integer operations.

^^ THIS, Entirely this. If they get 50% improvement on FPU or anywhere close to this, SO MANY things are just going to STOMP all over and SMEAR intel right off the charts. Not everything mind you, but that would really be SWEET as I do a lot of modding for BeamNG.Drive and that's ALL physics based and 100% slams the FPU. My Ryzen 3700x not only costs less than my old 4790k did new in 2014, it also runs absolute circles around it. Sure the single-core speed (if newer instructions are not used) isn't that big of a boost, but the content creation on here is PHENOMENALLY better. Sometimes as much as 4~5x the speed such as file zip or unzip operations, working with graphics, converting models, and running lots of traffic in BeamNG Drive. It's an amazing difference currently for me, not to mention that what 50% more FPU performance would do. That'd mean up to 24~30 vehicles VS the 15~21 I can currently run (at 40~20fps average respectively). All that stuff above is so true I would even call it an understatement. Even YOUTUBE or NETFLIX uses the FPU - all the encoded video you view on the web uses it. All the H.265/H.264 stuff is all FPU based and heavy multi-media extension use (which again is fpu based). The future is physics in games. If that's what they deliver, or even 25~30% more fpu performance, I'll buy one the day it's available. Hooray for competition!

Yeah SIMD performance is extremely important these days as more and more software uses it. In fact, I would say SIMD is a large part of how x86-64 CPUs have become so dominant over the years, especially in HPC and database. Both AMD and Intel design their CPUs around heavy use of SIMD instructions. If Zen 3 has a 50% boost in FPU, they may have added a third AVX2 unit. If it were AVX-512, the performance gain would be higher I think.

Alessio1989:

specially since the x87 FPU is no more used outside legacy 32-bit software or ad hoc intrinsics calls on 64-bit software.

Just because doing simple floating point math does run through the SSE register these days does not really make it proper "SIMD" (Single Instruction Multiple Data, there is no "multiple data" when just using it as a replacement for x87) and is unlikely to get significantly faster, and while some games do use SIMD, in-depth use is still rather rare. In any case, as you may have noticed I did acknowledge that SIMD improvements are nice, but making up some average IPC from specialized improvements is just pure marketing. "Average IPC increase of 20%", of which a huge chunk is FPU improvements, does that mean an average application is going to run 20% faster? Not very likely. Some may run no faster at all, and some may benefit exceptionally well from that 50% FPU. Just present the details, don't create fake hype with numbers like "20% extra performance", AMD is already suffering from overhyping every new generation.

bobblunderton:

Even YOUTUBE or NETFLIX uses the FPU - all the encoded video you view on the web uses it. All the H.265/H.264 stuff is all FPU based

To be correct, Video encoding/decoding is based on Integers, not Floating Point. If you use a software decoder, it'll still make good use of SSE/AVX, but with Integer units, not Floating Point. But, unless your system has a very old graphics card for some reason, you are far more likely to be decoding video on the GPU instead.

@AATT why? @3600 im getting 50GB/s. you need more?

- Increased IPC - Checked ! - Increased clocks - Checked ! (even if just a bit) - Unified cache (so little to no thread jumping penalty) - Checked ! - Even more power efficient - Checked ! If people considered 3000 series as good, the 4000 will blow their socks off !

Will it be on am4 socket,?

fry178:

@AATT why? @3600 im getting 50GB/s. you need more?

To really make use of wider vectors like AVX2 and AVX-512, you need lots of memory bandwidth. So no, 50 GB/s isn't nowhere near enough.

wavetrex:

- Increased IPC - Checked ! - Increased clocks - Checked ! (even if just a bit) - Unified cache (so little to no thread jumping penalty) - Checked ! - Even more power efficient - Checked ! If people considered 3000 series as good, the 4000 will blow their socks off !

So what about memorylatency? One of the things that have great impact in performance in many FPS games.

Fox2232:

That's still enough for marketing purposes...

i agree with that!

schmidtbag:

Intel still wins or is on-par in cases where latency matters, which for the most part is games, particularly those that are CPU-bound. That also in turn makes Intel a more ideal choice for robotics, though, I personally would advocate for ARM over Intel when it comes to robots, since ARM has even better latency (especially when it comes to inter-core communication) and lower power consumption. AMD wins where total duration and performance-per-watt matters (which is basically everything else not yet mentioned). EDIT: Oh yeah Intel also wins at AVX512, but... that's pretty niche.

In 3 out of 100 let me say tests, is not enough to say intel still beats AMD. Not to mention that AMD now have 16/32 core/thread CPU, while intel is still on 8/16 (i did not include here XEs and threadrippers, there is difference even bigger for AMD). I am very subjective (biggest AMD fanboy ever in history), but i waited many Years so i can claim that AMD is kicking Intels butt right now 🙂 - in desktop segment.

Loophole35:

Adobé as well. For now it like the frequency that the Intel parts can run at. But really it’s that and games. Intel will likely lose that after the end of 2020 then they will be fully in catch up mode.

Yes, i forgot this, thats true, but this is mainly because most Adobe plug-inns are using only 1 core.

What I'd like to know is if I'd be able to put zen 3 CPU into my current X370 mobo, but that is not up to AMD to answer...

mohiuddin:

Will it be on am4 socket,?

This will be on a x670 chipset PCIE 4.0 and the last AM4 socket cpu from AMD.

If we are dreaming can we also say it will work in a x370 board.

butjer1010:

In 3 out of 100 let me say tests, is not enough to say intel still beats AMD. Not to mention that AMD now have 16/32 core/thread CPU, while intel is still on 8/16 (i did not include here XEs and threadrippers, there is difference even bigger for AMD). I am very subjective (biggest AMD fanboy ever in history), but i waited many Years so i can claim that AMD is kicking Intels butt right now 🙂 - in desktop segment.

3 out of 100 is an exaggeration and you know it. For the record, I agree AMD is in fact kicking Intel's butt and in most cases they win in ways that actually matter. So, AMD is currently the obvious choice. I would even recommend them over Intel for most gaming systems despite Intel being ever-so-slightly better at it, because the FPS difference is indecipherable and if you really care that much about a few FPS, you're trying to compensate way too much for your e-peen. Despite all of this, I prefer to remain objective, and AMD isn't a complete winner yet.

nevcairiel:

Just because doing simple floating point math does run through the SSE register these days does not really make it proper "SIMD" (Single Instruction Multiple Data, there is no "multiple data" when just using it as a replacement for x87) and is unlikely to get significantly faster, and while some games do use SIMD, in-depth use is still rather rare. In any case, as you may have noticed I did acknowledge that SIMD improvements are nice, but making up some average IPC from specialized improvements is just pure marketing. "Average IPC increase of 20%", of which a huge chunk is FPU improvements, does that mean an average application is going to run 20% faster? Not very likely. Some may run no faster at all, and some may benefit exceptionally well from that 50% FPU. Just present the details, don't create fake hype with numbers like "20% extra performance", AMD is already suffering from overhyping every new generation.

Most of the time a CPU spend is waiting for load and store operations into system memory. Solve this and you will be one of the richest person of the world (you could have better luck trying to purge object oriented design in the entire world in favour of data oriented though!). But anyway, since most of consumer real-word CPU-demanding software rely on SIMD, FPU improvements (whatever AMD PowerPoint team meant) are always welcomed. As for integer operations, nobody cares about them any-more in consumer software.

Alessio1989:

Most of the time a CPU spend is waiting for load and store operations into system memory. Solve this and you will be one of the richest person of the world (you could have better luck trying to purge object oriented design in the entire world in favour of data oriented though!). But anyway, since most of consumer real-word CPU-demanding software rely on SIMD, FPU improvements (whatever AMD PowerPoint team meant) are always welcomed. As for integer operations, nobody cares about them any-more in consumer software.

Did you forget about caches?