Zen2 IPC increase

How will Intel recover? They've been stuck on within a 1.5% IPC increase each year for years. They're going to be on both an IPC and node disadvantage for the foreseeable future.

Attached: vivaldi_2018-10-16_22-58-11.png (901x400, 39K)

Other urls found in this thread:

youtube.com/watch?v=l5-gja10qkw
phoronix.com/scan.php?page=article&item=ubuntu1810-initial-server&num=1
phoronix.com/scan.php?page=article&item=amd-epyc7601-2p&num=1
twitter.com/SFWRedditImages

AMD still can't beat 1.2% yields, though.

Attached: 1536116000846.png (630x828, 653K)

Isn't that in the range people were expecting?

Either way, a similiar IPC increase in gaming should close the gap to Intel while higher overclocking ability due to TMSC's 7nm process should pull AMD ahead. Things are looking great.

Attached: excited loli.jpg (437x471, 52K)

My Wop friend here, great guy, has a spotty track record for leaks. He has had someone at a big OEM give him legit ES data before, and some early performance figures for some GPUs.
Given how conservative this is I'd be willing to say its probably accurate. Getting more than 10% IPC uplift in a range of various scientific workloads isn't anything astounding.
That couple with the clock speed increase is more than enough to make Zen2 a winner.

Suppose its only 7% ipc (per same clock) and 7% increase in clock speed.

That would give us 4.5 Ghz OC for most Zen2 cpu (similar to 4.2Ghz) and 4.7/4.8 ghz or so for top 30%.

Wait

13% IPC or 13% increase per core performance?

IPC is instructions per cycle and has nothing to do with clock speed(mhz)

If its 13% IPC and lets say 8%(4.3 to 4.65ghz) in clock speed we could be looking at 20% higher performance in single and multi core performance

Im scared
Scared for intel

Attached: TheFuck.gif (480x354, 1.48M)

If the memory controller is fixed to allot for 3600mhz+ memory we could see even larger gains

When is DDR5

>7% instructions per clock (per same clock)

>stuck on within a 1.5% IPC increase each year for years
In their dreams. They've been stuck with a 0% IPC increase per year for over three years now. Coffee Lake is just another Skylake do-over with more cores and higher clocks. The 6700K is as good as Intel's latest clock for clock.

Retweeted

>instruction per clock

Attached: hyouka disgust.jpg (1280x720, 75K)

We could see OC headroom skyrocket compared to 14 LPP since its unlikely that TSMC's 7nm process has the same voltage/clock scaling wall.
A base clock of 4ghz for an 8 core part with a 95w TDP isn't exactly much of a departure from the performance GloFo's 12nm line delivers. Performance at iso power should be 30% or more higher than 12nm, and the 2700x hits 3.7ghz base at 105w.

Even if clocks were limited by some nuance of the process, we should see TDP drop, perf/watt shooting way up.
Given the information about an engineering sample showing up with a 4ghz base clock and 4.5ghz turbo, I'm not exactly worrying about potential clocks. If an ES is hitting that then final silicon will have more legs than that.

For 3 years?
They have had like 3% IPC gain in 8~ years
Most of the performance gains of newer CPUs is due to higher clocks or newer instruction sets not available to older chips
actual hard IPC has not changed much

Btw
If these gains are true and the chips hit 4.6ghz this would be comparable performance to a 5.1ghz ryzen 2xxx

Attached: 1525127970971.jpg (4771x859, 771K)

Zen 2 has *POTENTIALLY* huge gaming improvements, I really hope AMD pulls through for it. Intel has been bigger jews than ever before and really need a steel toed boot in their ass.

The 10-15% IPC boost is good and puts them pretty close to Intel's IPC. But, Zen 2 is supposed to have big memory controller improvements and improved infinity fabric, meaning they can finally have much faster RAM speeds. Since Zen scales so much with memory, that could boost gaming performance a ton.

On top of that, if the rumors about 6 core CCX's is true, that's 2 more cores that don't have to communicate over the infinity fabric.

Their 7nm design supposedly will also be a big jump in cache sizes which is again a massive boon for overall performance. Usually pre-prod silicon has memory clocks at the normal rated speeds and sections of cache disabled so it could be even better.

Honestly why would anyone, outside of gaymers. really care about these news?

I though a clock hz WAS a cycle?

when the fuck is zen 2 coming out, nobody cares about performance without a release date

Zen 2 would rape intel if their base die is 6 core die instead of 4 core. Intel buyers would be caught pants down. If AMD actually does this, its dead end for Intel.

It's really a "clock cycle", so IPC can be cycle or clock and both are 100% correct.

Servers market is experimenting with new EPYC cpus. IPC increase with Zen2 means they are even more comfortable with it.

Stop with the anti-semitism

No 6 core CCX, there is updated IMC from Cadance supporting faster DIMMs without the hassle.
AMD will announce details at CES. EPYC2 chips have already sampled to some big vendors, they'll ship more throughout the year. Full release is early next year.

>They have had like 3% IPC gain in 8~ years
That isn't true though, and there's no need for the pointless hyperbole. With both chips locked to 3.5GHz, a 6700K gets a 12% higher single-threaded score in Cinebench than a 4790K, for example. Whereas there's fuck all difference between a 6700K and an 8700K. Intel's problems are laughable enough without having to invent pretend ones.

Intel made slow, but steady progress in terms of IPC from Sandy Bridge to Skylake, even it was never anything mindblowing. Now they don't even have that and can only keep cranking clocks and faking TDPs to stay on top.

Some people use CPUs for shit besides gaming dude

Attached: RyzenSkylakeIPC.png (1276x713, 349K)

Ram speed equalized? Show me.

>4790K
Thats a DDR3 chip compared to a DDR4 chip

memory speed is indifferent to IPC but it does effect overall performance

Intel ring bus design wont scale well past 6 cores
The 9900k will have higher intercore latency than the 8700k as a result of having an additional jump for the core to core communication.
The 5.0ghz turbo(possibly higher) and increased core count will allow it to perform at least as good as the 8700k even with the latency increase as it will be slight.
But going to 10 cores and beyond it will not work hence why intel went to the mesh interconnect for the skylake X chips

Attached: RyzenSkylakeIPCGame.png (1276x719, 559K)

10-15% is realistic on a brand new design like zen, respin it with some easy fixes that were slowing it down. It keeps getting harder to find improvements over the generations, which is why core has completely stalled at skylake.

AMD is at least 3 generations ahead of intel right now at least for the server side of things.

Intel will/must switch to small chips "glued" together or they will not be able to compete on a price front at this small of a manufacturing node.
AMD is 3 generations into the design where intel is at zero.
But go look at a die shot of a Q6600
Intel has been on this road before

Attached: Q6600.jpg (998x759, 86K)

bingbus scaled up to 24 cores, although i think that was the double bingbus broadwell, so technically 12

It was the range people expected, yes, yet you still have delusional people thinking the 3700X won't completely BTFO the i9-9900k.

The price of the i9-9900k probably has less to do with Intel's yields, and more to do with one last desperate cash grab before no one else is buying their CPUs.

Bits And chips has been one of the most accurate ones, still.

Look up what IPC means.

That can't be true at all unless the 4790k had shitty RAM in such a test.

Hardly scaling when the uncore power consumption is through the roof and the performance isn't scaling well.
Bingbus stops making sense after 8 cores. It can be used for 10, like the 6950X, but the downsides become more obvious.

>bits and chips
have literally never been right about their rumormill bullshit, literally worse than wccftech

RIP in peace intel

Attached: 1539272491104.jpg (800x600, 63K)

>Bits And chips has been one of the most accurate ones, still.
cite one thing they 'leaked' that turned out to be true

I'm not saying it was perfect, still done decent up to those core counts, mesh is worse perf/w at low core counts, high idle consumption.

but w/e just an observation

>picture

what

Attached: 1525884902029.jpg (374x347, 39K)

>cite one thing they 'leaked' that turned out to be true
Disagree and you're a homo.

Attached: firefox_2018-10-17_01-06-34.png (932x1021, 1.44M)

did you post the wrong image or

No they posted that and it's true.
Nvidia's best way to advertise the new RTX cards would be a nude Dawn.

ok so you're just retarded

I think the might be right though.

Common core + retarded "teachers", user.

Attached: __sakurauchi_riko_love_live_and_love_live_sunshine_drawn_by_pas_paxiti__477bdb048188a52f4655020ed4f9 (1150x1266, 120K)

>When is DDR5
not till 2020-21 dork
besides 3200+ DD4 will pretty much be the same as DDR5 initially in terms of speed (higher freq but looser timings) the only gain should be throughput so your vidya won't get better unless that vidya is a megafactory in factorio

Quite believable.

With the known layout of Zen1 most (sane) people speculated at best/at most a 15-17% IPC increase for Zen2.
Zen2 is legitimately the Sandy Bridge jump from Nehalem.
11-15% IPC increase with a 10-20% clock increase at same or lower TDPs.

Gonna wreck the market yet again.
Way to go AMD!

>we should see OC headoom skyrocket
Define skyrocket because I don't consider 10-15% a "skyrocket"-ed increase

Attached: ptb.jpg (273x184, 11K)

DDR5 will bring lower latency than DDR4, and the bandwidth increase is enormous. Early low power DIMMs were hitting 5500mhz, back in July Samsung stated they had standard power DIMMs at 6400mhz using just 1.1v.

Its absurdly fast, its even easier on the IMC, total latency decreases, density is drastically higher as well. Its better on absolutely every front. Its likely that mature DIMMs will reach effective speeds around 7500mhz, thats 120GB/s for two memory channels.
For the enterprise market having that increase in bandwidth for ECC even at 5500mhz is going to be fucking incredible.

Oh great so my 2020 mainstream 8 core 4.5Ghz CPU is still going to have the same bandwidth per-core as we've had for the last 5 years.

Even at an extremely modest 7% clock speed increase, 3700X is sure to have higher single core performance than the i9-9900k.

15% would be a 650MHz increase. How's that not "skyrocketed"?

Have a source on lower latency? From what I've seen, a ~20% true latency increase is expected. Going from the realm of 8ns peak to 9.7ns peak.
Meanwhile with DDR3 you could get down to around 7.5ns.

DDR5 will be incredibly for smartphones and servers because it's much higher bandwidth per watt, but from what I've seen it may be a regression for games.

The CCX bandwidth bottleneck mostly goes away at 2933MHz memory frequency. But I heard it is being increased.

I heard the IF ratio is going to be unlocked and we will finally have the long rumored 2x IF>IMC speed that Zen actually needs to win against Intel on all those benchmarks they currently lose.
I hear a lot of magical fairy wishes.

15% is not a skyrocket you clown. It's, being by percentage, relative, which means a "skyrocket" would be something unbelievably impressive.

>3700x is sure to have higher ST-IPC than a 9900k
3700x won't be competing with the 9900k, now will it?

15% increase in clock speed over a generation is a "skyrocket" in relative terms, you clown, as Intel was giving 3-5% increases per year.

>3700x won't be competing with the 9900k, now will it?
Yes, it will be. Intel has no 10nm CPU release except maybe a paper launch a year from now.

0/10 try being more subtle next time

Naw champ, RAM latency essentially hits a hard wall of about 8.5ns in the real world. This has held true since at least DDR2. I do not expect DDR5 to make this any different.
If DDR5 can double rates at the same latency we should all be sorely unimpressed,
Because all DDR generations have achieved this (once mature) from the outset.

Kind of a badly worded problem, cutting a board into into two pieces is one cut, 3 pieces is 2 cuts

The best DDR3 kits produced which were actually sold in volume, not the absolute highest binned cherry picked dies sold in $400+ kits, were 2400mhz cl10. Most of them however were lose timings.
Thats 8.33ns first word latency. That was the peak of DDR3 right before DDR4 came out and production decreased.

Some of the best DDR4 actually sold at a sane price now is 3600mhz cl15 giving it a first word value latency of the same 8.33ns, but they have consistent timings, and the spec of DDR4 improves inter DIMM communication over DDR3.

DDR5 at 6400mhz is hitting that with mid to high 20ns range, and even at 30ns it would yield a first word value of 9.3ns. Real world performance is going to be heavily in DDR5's favor from all the changes to the DRAM itself and controllers for the spec.

>But go look at a die shot of a Q6600
>Intel has been on this road before
goes through the FSB and ATE SHIT after the security patches.
Yeah no, Intel have no tech for this shit currently.

you respond to me as if I'm not on your side.
it hurts a little, and pains more the older I get

I only want you to remove yourself from some emotionally idyllic fantasy
And get back down on the ground with the rest of us.

It's better than OK to be realistic. Realistic is something I think our modern world is sorely devoid of.

intel will probably pull something out of their ass in the next 3 years or so, they're betting big on their reprogrammable GPU so it's more than likely going to get baked into whatever CPU they are working on and bring something like ray-tracing or path-tracing to the general public

>cutting a board into into two pieces is one cut, 3 pieces is 2 cuts
And if one cut takes 10 minutes then those two cuts will take 20 minutes. The student is correct.

Reminder that this is Instructions per Clock, so with the increased clock speeds of the new process, the improvement might be impressive.

Are you saying that... Threadripper competes with the i9-9900k instead? Or what? So the 3700X just won't have competition?

I was realistic with Zen+, and it wound up being better than anyone expected (due to having some IPC improvements from microcode and a slightly better IMC reducing memory latency, and being more than simply a die shrink). Was better than expected.

20% performance increase per core with Zen2 is very very conservative. That's all it needs to win in every way. Per core, multicore, gaming, etc. It's likely it'll be even better than that, though.

why is AMD's stock in the shitter if zen2 is going to kick so much ass?

were we really talking about core versus core performance or are you just trying to save nonexistant face on a bullshit message board?

youtube.com/watch?v=l5-gja10qkw

Ryzen stutters in games and is a buggy piece of shit that can't be used reliably in anything. Enjoy premiere pro crashing, but hey, at least you can run synthetic benchmarks all day long I guess lmao.

What are you even talking about on the server side of things? EPYC 7601 fucking destroys Intel's processors.

phoronix.com/scan.php?page=article&item=ubuntu1810-initial-server&num=1
phoronix.com/scan.php?page=article&item=amd-epyc7601-2p&num=1

>But go look at a die shot of a Q6600
>Intel has been on this road before
That is not even remotely comparable to the Infinity Fabric.

Because now is the time to buy you fucking muppet

Oh, you said 3 generations AHEAD, sorry. I thought you said "behind".

My bad.

that only happens for shitty games that aren't multithreaded correctly

>i-it's s-someone else's f-fault!

Yeah, the AMDfag slogan. Meanwhile, intel just werks while AMD shit brings nothing but headaches. But I guess Premiere Pro is a "shitty game" now lol.

This is entirely a question that is poorly phrased. It's trying to teach fractions where 2/2 * 10 = 10 and 3/2 * 10 = 1.5.

Both answers are actually correct, depending on how you frame the question. Just goes to show that you need to ask questions correctly.

AMD is pure shit.

I have never seen it, and been gaming on ryzen since day one. So since you posted no data my word is just as good as your's

>Both answers are actually correct
No they aren't. The student was right.

You mean when ryzen will stop freezing my entire pc? Kys. t. R5 2600 unusable for basic office work paired with a 500gb ssd

Doesn't happen on my Ryzen 6 core nor for anyone else I know.
You have bad RAM or something, most likely.

Incomparably smoother than the i5 I had previously.

You are fucking missing the point of my post, brainlet.

>This is entirely a question that is poorly phrased. It's trying to teach fractions where 2/2 * 10 = 10 and 3/2 * 10 = 1.5.
Granted, I did typo "1.5" which should be "15", but the point still stands. You can clearly tell that the teacher is attempting to teach fractions and that the question is poorly phrased.

In other words, with a poorly phrased question, which answer is correct depends on how you frame the question.

Attached: brainlet2.jpg (800x450, 41K)

It's not poorly phrased.

To break a board into 3 pieces, you need to first break it in two then break one of the halves into two.
So it takes twice as long to break it into 3 pieces.

You and the teacher are just retarded.

I have a 2700x, and never had a freeze. Maybe it's not the CPU.

Does shilling for intel pay well?

>shitter
>went from $2 to $26
Are you dumb?

>hurrr call someone a brainlet win argument.

You can say the question is poorly phrased all you want. The way it's written the student is correct. It has nothing to do with framing, It can only be interpreted one way in an objective reality.

Now if the question has said It took her 10 minutes to saw two pieces _off_ of a board (thus implying 2 cuts in the first sentence making 3 pieces). Then you and the teacher would be correct, but it doesn't say that.
The question uses the word 'into' twice. It's completely internally consistent, and not vague at all.

It maybe wasn't what the author intended, though based on the look, it's from a text book so I suspect the teacher's just a moron, much like yourself.

run memtest

>To break a board into 3 pieces, you need to first break it in two then break one of the halves into two.
>So it takes twice as long to break it into 3 pieces.
That depends entirely on the context of the question in the first place. If the context is homework or a test that pertains to fractions then the question is poorly phrased.

>You and the teacher are just retarded.
The teacher is correcting people's homework/test answers from a list of definitive answers that whoever wrote the question in the first place calculated.

The intent of the question does matter a lot. The student isn't wrong, but it is unlikely that the teacher is. We don't know the context the question was framed in because the picture doesn't tell you, but the intent clearly shines through based on the correction.

>You can say the question is poorly phrased all you want. The way it's written the student is correct. It has nothing to do with framing, It can only be interpreted one way in an objective reality.
You assume that the context of the question was NOT a chapter dealing with fractionals or a test for fractionals.

>It maybe wasn't what the author intended, though based on the look, it's from a text book so I suspect the teacher's just a moron, much like yourself.
I'm arguing for the intent itself, that's what I'm saying. In other words, the question is misleading and poorly phrased, but frame it in a context of fractionals and you begin to understand what the intent was. It's VERY clear from the correction what the intent was, and again, the teacher is probably correcting from a list of definitive answers that the author of the question wrote.

I completely agree that the question itself directly suggests the answer is 20, but that is why context matters.

Attached: cuts.png (1454x841, 145K)

It might be the CPU. Have you set "Power Supply Idle Control" to "Common Current Idle" in UEFI?

The C6 sleep lockup problem is quite well documented, and the problem is CPU power delivery when it needs to wake up from C6 sleep. The issue isn't the CPU itself, but rather that the power supply can't maintain power delivery to the CPU when it's that idle.

>I completely agree that the question itself directly suggests the answer is 20, but that is why context matters.
>suggests
It outright states that the situation is such that the result is 20 min. It's the only logical answer and as such any other drawn from it is objectively wrong. If the question is intended to teach division it is faulty for that purpose, but it is unambiguous as it is written.

Now you're just nitpicking on words. I agree with you that the answer is 20 for how the question is phrased, but my point is that the question itself is poorly phrased because what it asks and what it intends to ask are two entirely different things.

If just consumer software implemented IPC it would be nice.

It would have been easier if thy didn't ditch QPI or whatever they had in the i7 920-era. There's something about the Nehalems that made them feel faster than they were.

Alter the question as little as you can to make it support the other solution.

There have been NO IPC improvements from Intel, since the architecture and die size hasn't changed since at least Skylake.
Any kind of small differences are margin of error stuff.

holy shit. we got a live one here.

Someone else already did it for me. See >Now if the question has said It took her 10 minutes to saw two pieces _off_ of a board (thus implying 2 cuts in the first sentence making 3 pieces).

I noticed it and that's not an answer. You can't just throw the word "off" somewhere in there to fix it. How much do you have to alter this exact worded question to make it mean what you want it to mean? How many characters have to be added, removed or replaced? In total, not just from the last part. The first part would make no sense then.
>It took Marie 10 minutes to saw a board into 2 pieces. If she works just as fast, how long will it take her to saw another board into 3 pieces?
My point is there's nothing unclear about it and it wasn't a typo. You have to make a different question to get it fixed. Also I don't know if it needs to be pointed out but you have multiple people replying here.

>My point is there's nothing unclear about it and it wasn't a typo.
It doesn't have to be a typo. The question is poorly phrased and the issues run deeper than just a single word.

The teacher is most likely correcting from a list of definitive answers written by the author of the question in the first place. It is clear from the teacher's correction that the intended answer to this question is based on fractions, where 2/2 * 10 = 10 and 3/2 * 10 = 15.

Here's an attempt at fixing it:
>It took Marie 10 minutes to saw 2 pieces off a board. If she works just as fast, how long will it take her to saw 3 pieces off a board?

>RRRRRREEEEEEEEE EVERYONE WHO SAYS ANYTHING EXCEPT FOR WHAT I WANT THEM TO SAY IS PART OF MY POLITICAL OPPOSITION
I'm just trying to offer an explanation for why it's fucked up you moron.

>How to derail a thread
I hope you are happy with yourself

>a board into 2 pieces
21 characters removed
>2 pieces off a board
20 characters added
>another board into 3 pieces
27 characters removed
>3 pieces off a board
20 characters added

Characters removed: 48
Characters added: 40
Total changes: 88
Total characters in the question: 143

This is a totally different line now which was my point. Since it can't be an accidental typo, it can only be either on purpose or a brain error. A teacher shouldn't just accept this and mindlessly read from an answer book, they should focus and read the answer with some thought. This question cannot create the answer that shows in the answer book so people shouldn't be corrected into being incorrect. This teacher is teaching wrong. That's even worse than not teaching.

Ass burgurs GTFO!

The most irritating part of this pic is the paste on the smaller side of the bread. Who the fuck does that?

>This is a totally different line now which was my point.
We understand each other here, yes, but *my* point was that the question was incorrectly phrased in the first place.

>A teacher shouldn't just accept this and mindlessly read from an answer book, they should focus and read the answer with some thought. This question cannot create the answer that shows in the answer book so people shouldn't be corrected into being incorrect.
I agree with you completely here, yes. Part of the teacher's job is to understand their students and their answers and either correct their logic or recognise a flaw like this one. Sometimes the list of definitive answers is also wrong, and I've run into such cases when checking my own answers to math problems (although it isn't 3rd grade maths like this question here), which the teacher should notice when correcting someone's answers.