Wait, what is this?

$329-349, as usual with x700 SKUs.
Why did they even do that after someone fed him ALL the Rome details.

3700X HYPE TRAIN
30% MORE IPC
200% MORE THREADS
100% MORE COARS
25% MORE NIGGAHURTZ
50% LESS TDP
THE 8700k/9700k JEW NIGGERS ARE ALL GON

>30% MORE IPC
Hahaha, that's funny, user.

>THE 8700k/9700k JEW NIGGERS ARE ALL GON
kek, it's anudda 7700k shoah all over again
>buy top of the line 9900k $500 CPU to compete with $300 CPU
>4 months later AMD releases a better CPU for the same price but doubles everything.
fools and their money are soon parted.

ofc it's funny, ipc means instructions per clock.
so, in worst case where you have the same clock, same μarch but double the cores, IPC simply doubles.
so, if there are even 10% improvements per core, we are talking about 210% IPC increase.

But since the different cores are on different clock domains, it should be a priori obvious that IPC refers to a single-threaded metric.

Man, Intfails just can't cope lol

ipc means one thing
Instructions Per Clock.
Assume a single core 4-issue superscalar CPU that has IPC=4
Using the same single core on a 4-core CPU with added caches to avoid any performance decreases due to cache misses across the more complex chip and keep the clock the same, a program that scales 100% will get IPC=16

IPC on multicore systems depends
1st on the number of cores, number of instruction issue per core and how frequently you get cache misses and mispredictions
2nd on the program that runs on the CPU. if the program is able to run on single core thread, then the IPC is X, if the program can use 2 cores, then it's roughly 2*X.

running single core programs in order to determing IPC per Core is stupid, because on e.g. 2700x you get a 1 core utilization and you give it the chance to have 16MB cache for itself, so you actually skew the results by giving it an enormous pool of fast memory to use.
If you use the same CPU with a program that scales 100% on all cores of the 2700x and you divide the results by the number of cores, you just fuck up everything because your results are skewed due to cache coherence, e.g. you share data across cores but you want to measure the single core perf with the proper cache misses, and this "IPC" is valid for that specific program.
Bulldozer for example had a terrible IPC on programs that had many and wide floating point operations, but had more IPC even by 4790k when it came to integer operations.
The greatest part is that those "tech" websites and "tech" youtubers haven't taken even a single class of comp. arch, so they feed the consumers with uneducated bullcrap.

Isn't it 8MB per CCX, no matter what?

16MB per CCX in Zen2.

the 8MB L3 cache is present on each CCX.
When you get 2 CCXs in ryzen 1000 and 2000, core from the first CCX can get data from the L3 cache of the second CCX.
the greatest example is the 1500X with the 16MB cache.

I am talking about the 1000 and 2000 series of ryzen.

>When you get 2 CCXs in ryzen 1000 and 2000, core from the first CCX can get data from the L3 cache of the second CCX.
That's very slow.
You should treat Zeppelin as 2*8MB NUCA setup.

>That's very slow.
it's still a couple of orders of magnitude faster than fetching data from the dram.
prefetching data and using the second ccx's cache for not so frequent data usage(which can be determined at a certain point by the compiler) it can reduce significantly the throughput of the core.

>it's still a couple of orders of magnitude faster than fetching data from the dram.
No, not really, HA sits at IMC so you're still doing a roundabout trip around the chip.
Again, treat it as 2*8MB NUCA setup.

WHY ARE YOU SO HAPPY ABOUT THIS YOU LOSERS

INTEL IS JUST GONNA RELEASE 10900K AND THAT WILL BE THE END OF THIS MEME

check those charts at the bottom of the page.
anandtech.com/show/12625/amd-second-generation-ryzen-7-2700x-2700-ryzen-5-2600x-2600/3

It clearly shows the latency skyrockets after 8192KB so yeah, goes EXACTLY as I've said it.

Cause intel having actual competition stops them from being the greedy jews theyve been these past 10 years

>Assume a single core 4-issue superscalar CPU that has IPC=4
But the entire point is that a 4-issue superscalar CPU doesn't get IPC=4. That's exactly why it's interesting to compare IPC between architectures, because it's a metric of the efficiency of the core, and that's why it's only really interesting as a single-core metric.

That's not to say that your point about a single-threaded program getting all shared resources to itself isn't valid, but that's also why most people measure IPC and thread scaling as two separate metrics.

ok, I didn't know that you couldn't read the chart.
The chart shows the latency between l1 l2 and l3.
check pic related to get a spherical view on l3 vs dram latency

Attached: RangerLatencyChart.jpg (919x670, 241K)

Literally not happening unless Intel has a new node up and running like tomorrow
Intel’s already at its limit with the 9900k which is a heavily binned and aggressively clocked CPU on a very mature process and it’s thermals put it on the edge of being uncoolable with off the shelf cooling solutions.
They can’t do anything without something being sacrificed, if they add more cores they have to lower the clock due to power/thermal limitations. If they increase the clock speed well that just means less CPUs available and increased costs

The page you've linked has no such chart, moron.

fuck

Attached: kek.png (476x202, 10K)

>efficiency of the core
the efficiency of the core in any modern CPU is ~20% of what the throughput you get.
Why? Hazards, speculation, if-conversion, predication, e.t.c.
so yeah, the efficiency is shit, 80% of the instructions that are executed are not used because the CPU runs as much code as possible and uses whatever is valid in the end.
>most people measure IPC
I explained above that IPC is different on the same CPU when you run different loads.
read this AGAIN
>Bulldozer for example had a terrible IPC on programs that had many and wide floating point operations, but had more IPC even by 4790k when it came to integer operations.
I personally run gnu radio demodulation examples on the lab's 4790k and on a friend's 8350 and the 8350 was way ahead when demodulating integer data.
Also, it consumed less than the overclocked 4770k sold as 4790k.
I gave you a picture that's easier to digest, because you can't read the page I gave you.

I'm drooling and my erection is about to split like an over-ripe banana.

IT'S OVER INTEL IS FINISHED AND BANKRUPT

Will this work with my 1800X motherboard?

>inb4 that was the plan all along, just like 32C TR
>inb4 VRMs like in the ASrock X370 Taichi weren't overkill at all

backwards compatibility is guaranteed but you might miss some features not able to run with a simple bios upgrade, due to h/w missing from your chipset.

Even if this is true the chips will somehow be shittier than Intel in games. It's always been that way and always will be. This isn't going to be slamdunk for amd in all computer tasks. It just ISNT GOING TO HAPPEN.

Attached: download (1).jpg (227x222, 6K)

My motherboard is a Asrock X370 Taichi. Do you think it will support this CPU well?

no one cares if some underdeveloped gaymes can't utilize a multi-core system.
If it ISNT GOING TO HAPPEN, why is Intel is playing catch-up with Ryzen after every release?
Wasn't 4cores enough for everything?
what's after the 250Watt 9900k at $500?
a 500Watt 16c32t 10900k at $2000?... just to beat 3800x at 125Watt?

You'll be missing meme features at most

Maybe they will be good who knows.

Intel isn't play catch up fool. Their arch is better, their per core performance is better, they only lagged on core count and that shows with their only deficit to amd being ONLY in certain workstation tasks. 4 cores isn't enough even for gaming and intel is paying the price for it. Now that intel has similiar core counts amd gets demolished in everything. When intels new cpus come out it will be the same situation we are in now.

AMD will be better till intels new cpus come out. The question is how long of a window will AMD be superior.

You know it, i know it, EVERYONE knows it.

how much are you getting paid, mr pepe?

Intel is basically at their limit of 14nm, the last two generations have been Intel cannibalizing Xeon SKUs to compete with Ryzen and clocking to the moon to maintain the edge in gaming performance. They can't push clocks any higher and there's no way they can maintain their margins and compete with a ~$500 16-core with a 4.4GHz all core boost. For reference, the closest SKU they have to those specs has a 4.4GHz two core boost and is priced at ~$2000.

Intel has spent the last half decade trying to bring their 10nm to market, but they were overly ambitious and couldn't get yields to any kind of reasonable level. So now they're gutting it and they'll have "10nm" to market later this year or early next year, and it will be a side-grade, at best, from 14nm++. It will be at least another year or two before they reach performance parity on 10nm with their current process.

Quite the delusional block of text considering a 7700K is equaling a 2700X in performance and there are half a dozen Intel processors performing above the 7700K.

Excuse me, 7980XE is nearly $2000, the 7960X is actually closer to $1750. Still, that's 3.5x the price of the Ryzen competitor. Also bear in mind these processors require a more expensive X299 motherboard. You do get quad channel memory, which is a plus, but for that kind of money you could easily get a 2950X and still have money for a nice GPU.

>Intel isn't play catch up fool
so they moved from 4 cores to 6 cores to 8 cores on their $350 parts in less than 2 years because they care about (YOU) but they stayed nearly a decade on 4cores.
>Their arch is better
Can you point the areas where it's better?
Right now the only thing I see is that they have a shitload of sec. holes and they are copying bulldozer's cache architecture on their 9xxxs and hedt chips.
>their per core performance is better
that's why I said that they only excel on poorly coded old engines
>they only lagged on core count
they lag on interconnects, high latency on many-core cpus
they lag on node
they lag on power consumption
they lag on caches, on multi-cpu configuration and a shitload of other cpu related staff.
they also lag on engineering, fabricating something at double or triple the cost of the competition and at 1/10nth of the yields is not acceptable for someone the size of intel.
>4 cores isn't enough even for gaming
It never was. even 5 years ago gaming and streaming wasn't viable on 3770ks or 4770ks. Bulldozer had better performance on gaming+streaming even against the 3820
>The question is how long of a window will AMD be superior.
just a couple of years when intel finally gets their euv equipment working from ASML
By then TSMC and GF will be at 5nm

>You know it, i know it, EVERYONE knows it.
that sounds desperate, just like intel's move from 14nm to 22nm on many new and current chipsets.

>I explained above that IPC is different on the same CPU when you run different loads.
Of course, noone ever pretended anything different, rather the opposite. But that doesn't mean that there aren't composite IPC metrics that are somewhat useful.
>so yeah, the efficiency is shit, 80% of the instructions that are executed are not used
That's actual bullshit. Yes, some instructions are thrown out due to invalid speculation, but you'll have to find extremely speshul workloads to get it up to 80%. In particular, for most inner loops (which make up the vast majority of actual instructions in most programs) are usually predicated perfectly except for the last iteration. Also, many times the instruction thrown out due to invalid speculation are still useful, as they prefetched data into the cache that is used anyway by the correct path.

A 7700k might hold up in gaming or single/lightly threaded workloads. Coffee Lake and beyond are what I was referring to. The 9900k basically has zero OC headroom, even with watercooling, and beyond 8 cores clocks start dropping off pretty fast. Intel is doing everything they can with 14nm++ until they can get 10nm out and they had already stated the original 10nm design would be inferior to 14++, so the replacement "10nm" certainly will be as well.

>copying bulldozer's cache architecture on their 9xxxs and hedt chips
One of the main drawbacks of Bulldozer's cache was that it was writethrough, and that's not something Skylake is doing on any cache level. What part of the cache architecture are you even talking about?
>that's why I said that they only excel on poorly coded old engines
They excel on the vast majority of single-threaded tasks. *Mostly* due to frequency, but not merely. Also, what does it even really matter if the engine is "poorly coded" or not, if Skylake still runs the same code faster?

Ok, I see. Excuse my ignorance.

joke's on you, i don't play games because i'm on linux so i'm happy

I've never seen anyone so objective on Jow Forums on any subject regarding hardware, holy crap!

Quick question: I already have a Z170 build but a shitty CPU in it. Do I want to switch to Ryzen for gaming on 2560x1080@60Hz or do I just buy a 7700K?

>more cores and higher clocks than amd themselves said to expect
>its too good to be true so it must be real
LOL its like you newfag plebbuttors have never seen a die shrink. Its not going to magically allow you to double to cores while increasing the clocks by 30% at the same tdp

Attached: thot2.jpg (1080x1350, 1.26M)

If you don't want to have massive buyer's remorse it would behoove you to wait a week to see how CES and the Zen 2 leaks pan out

>useful
IPC is useful when you want to learn the bottlenecks of your CPU.
e.g. run some integer heavy program, IPC=2
then run some floating point heavy program, IPC=1.2
then run a program with complex if statements and large loop blocks, IPC=0.4
that's how you build CPUs, you study common programs, you check that % of branches you have, you check the strides you do in memory addresses, you check the % of arithmetic operations and you design your CPU around those metrics.
IPC will tell you where your CPU bottlenecks.

>That's actual bullshit
I know that Jow Forums is uneducated and haven't ever done profiling on CPUs and it's my fault for losing my time here explaining even the basic shit.
software.intel.com/en-us/vtune-amplifier-help-clockticks-per-instructions-retired-cpi
...but I give (YOU)s a break because this is advanced comp. arch. fields and you have to have at least a M.Sc. in comp. sci. and take classes on comp. arch. to learn them.
go back to ricing desktops.

2600x -> 3700x

Would I be retarded
All I would need to swap is the chip, my mobo is 470x should be supported. All I'll miss out on is pcie 4.0.

>IPC is useful when you want to learn the bottlenecks of your CPU.
I'm very well aware, thank you. I've been doing IPC- and other related measurements and microarchitectural optimizations in my own programs.
>software.intel.com/en-us/vtune-amplifier-help-clockticks-per-instructions-retired-cpi
The fact that you're even linking this article as proof of your number-of-discarded-instructions vs. number-of-retired-instructions is just proof of your own lack of education, as it has nothing at all with discarded instructions to do.

Yes.

>Wait, what is this?
This is pure antisemitism.

Yes it would. Would be like upgrading from a 4970k to a 6700k

Thanks for the input you fucking faggot go back to t_d on plebbit

Yes ok but HOW MUCH IS IT
If they go maximum Jew because they know Intel is behind right now then they can fuck right off. I want that 8 core processor to be no more than £180 now that the 6 cores are shifting down. The previous 6 cores were sub £180.

I don't see how. If the 7700K equals a 2700X in gaming, I don't see how buying a Zen 2 cpu and board would be a better investment. I do not want to spend more than I have to, I don't care if a Zen2 CPU will be better than the 7700K if it will come at a higher price with the additional purchase of a motherboard.

>Would I be retarded
Not at all. The only caveats I could think of would be:
>power delivery on your current board; wait for reviews to see how much power the 3700x actually needs for various turbo levels and whatnot and match that to your board
>as you said, pci-express, but you may perhaps also be missing out on memory frequency on your current board, but if you aren't planning to upgrade your memory anyway, that's probably not an issue

OYYY VEYYYYYYYYYYYYY

Are this zen things any good for multiple virtual machines or they are still shit compared to intel?

Attached: 1307998052511.png (284x193, 56K)

If you literally don't care even if Zen 2 is substantially better than the 7700K due to the cost of a new motherboard, why did you even ask if you should switch to Ryzen?

Then why the fuck did you bring up Ryzen like you were considering purchasing it if getting a new mobo is giving you the vapors?

>still shit
They never were shit.

Last time I used AMD was in the FX era and they WAS shit for VMs

I literally replied to a 7700K vs. 2700X conversation. I want to know if I get the same fucking performance gaming or not. I don't give a flying fuck about EVEN MORE performance. Got it?

I know youre a newfag but all this 1488 shit didnt happen on gee until 2016 when you came here from t_d. Go back

Yes, but then they weren't "zen things".

instructions executed minus instructions retired divided will give you the discarded instructions.
I don't have a Jow Forums grade meme-tier picture to show you with pretty colours and shit, it's either vtune's wiki or some 25 page paper.
The wiki seems already too much for you.

>all this 1488 shit didnt happen on gee until 2016
You know how I know you didn't come here before 2016?

>divided
typo. just skip it.

Please about how every thread in 2013 had oyy vay kikes btfo posted in it other than the few delusional fx owners trying to cope because i can assure you that never happened

>instructions executed minus instructions retired
Again, you have no idea what you're talking about. The number of "instructions executed" as you put it literally *is* the number of retired instructions. There may be some room for definitions here (which is why one normally doesn't speak of "executed" instructions whenever precision is required, but rather of retired instructions), but that's certainly the normal definition. The article you linked certainly doesn't redefine it in any other way.

I'm waiting, do I get 2700X performance out of a 7700K in gaming at 2560x1080@60FPS, yes or no? Or do you only respond if you want to complain, being useful is not your MO? Fuck I hate this place so much...

>another underpinning in games chip
wew its nothing

If the leaks are accurate Zen 2 will provide a significant upgrade over the 2700X or the 7700k at a similar price. Wait a week, see if the leaks are accurate, and if they are, go for a 3600X.

When he says ‘poorly coded’ he means not utilising multiple cores/threads, so they aren’t running the same code faster, since 2/3s of the Ryzen cpu’s Cores and running any core at all, because older programmes are poorly coded for modern hardware, multi core systems are the future (although I’d say in ARM not x86, I’ll bet you any money that in the next 3 years MacBook uses ARM) and the code will catch up, even newer video games are getting far more core utilisation than they used to.

There's been Jow Forums and Jow Forums crossposting since at least 2011, which is when I started browsing Jow Forums. I've been here since a year or two before that.

the page I linked is the tool that is used to check the efficiency of the executed code vs the useful code.
do you know anything about predicates?
every fucking instruction is executed, when the compiler has done the if-conversion, by the cpu and when it reaches the one instruction that depends on one of the predicates it executes it and discards the other instructions based on the rest of the predicates.
I am not taking about flushing.

>When he says ‘poorly coded’ he means not utilising multiple cores/threads
I know that's what he means, but running single-threaded code is also a very relevant measure of the goodness of a CPU.

I see, thank you. If the 7700K and 2700X are equivalent in performance for my use case and buying a 3600X would cost basically the same as buying a 2700X just with more performance it really is worth waiting for.

You realize Zen 2 will use the same boards and socket as Gen 1 Ryzen, right? And that the 2700x won't be much cheaper than its Zen 2 equivalent if the leaks are true? To answer your poorly worded question, IF the leaks are true, the 7700k will perform worse in games than most of the Zen 2 lineup, though probably not by a whole lot.

Yea, excuse my shitty phrasing, it's 2:26 AM in Hungary over here, and we don't speak English by default.

AMD "leaks" are Intelkike false flags to get people unrealistically hyped and turn the actual announcement into a disappointment even if it ends up being a massive improvement over Zen+. They're trying to take the wind out of AMD's sails.

Attached: 1464375613055.png (200x200, 26K)

>do you know anything about predicates?
Yes, it's a very common term and doesn't commonly refer to any specific concept in machine language, unless you're talking about predicated non-branch instructions, but it doesn't sound like you are (since those don't "discard" following instructions), and also they make up a very small minority of instructions in modern programs (being mostly useful on in-order architectures rather than out-of-order ones).
>the page I linked is the tool that is used to check the efficiency of the executed code vs the useful code.
vTune is only one of several programs used to capture CPU performance counters. My own tool of choice is the perf tool on Linux. Your definitions are still imprecise, however: There's no commonly agreed-upon meaning for your terms "executed code" and "useful code".

AMD = Another Massive Disappointment
Even if the leaks are real (they aren't), Intel will one up them right away

ehww that name sounds awfull. besides i don t they they can add more ++++ to the 14++++++++++++++++++++++ architecture

could we get another miserable PR statement from intel like the 'glued together desktop dies' for EPYC?

Wont 16 cores be terrible with dual channel memory?

the 32C threadripper sucks ass on quad channel

Attached: 1504839114482.jpg (500x500, 22K)

so you are saying that AMD is 1 step ahead of intel? so you are saying shintel needs several months to come out with something that is better than AMD?

>implying t_d cucks are not good goys

The 32C threadripper sucks for reasons unrelated to memory channels or even the die topology
Pic related is a 32 Core TR vs 32 Core Epyc
The Epyc machine is fully populated with 8 channel memory

Attached: Screenshot_3.png (721x585, 62K)