Zen is faster than Broadwell-E clock for clock

[Atmos]

16 minutes ago, Dabombinable said:

With the shit Intel, AMD and Nvidia have pulled in the past, do not believe the benchmark at all and wait until Zen is in full production and reviewed.

" what they're claiming is a higher ipc "

" they never claimed "

"  if this cpu is capable ..."

Never said any of it was true, was just correcting a misconception.   I agree though, we do need to wait to see third party reviews of the hardware.

 

 

15 minutes ago, zMeul said:

that CPU has a base clock of 3.2Ghz and a boost clock of 4Ghz

AMD's benchmarking was done at 3Ghz, by literally gimping the Intel CPU - lower freq translates into lower IPC

 

so what is AMD trying to prove here, that their ZEN based CPU has same IPC (or better) than a i7 6900K?

because that's false and it's a marketing move to distract from the real performance of ZEN - ZEN isn't at Broadwell-E level

You seem to not understand what IPC is, or even actually means. Single threaded performance is based off of two things mainly; thread clock speed, and thread ipc.

IPC means the amount of instructions a thread can perform per clock. IPC does NOT change with clock speed; single threaded performance does though, because it is based off of clock speed, and ipc of that thread.

 

If this cpu that amd was showcasing was run at the same 3.2-4ghz as a sotck 6900k, no matter the clock speed, it would be faster so long as there is no bias in the testing software; because it has a higher ipc, it can complete more instructions per clock, per thread.

For simplicities sake, lets say that the 6900k has an ipc of 4, it can complete 4 instructions every cycle, and its running at 4ghz, or 4,000,000,000 cycles/second... meaning it can complete 16,000,000,000 instructions in one second. This theoretical amd cpu however has an ipc of 5, and is running at 4ghz, meaning its capable of completing 20,000,000,000 instructions per second. Because the base ipc of the thread is faster on this theoretical amd cpu, it will always have better performance when compared clock per clock, thread per thread.

 

Once again... Because amd decided to release this data like this, and present it in this fashion, almost definitely means it will not release to US at those clock speeds, but slightly slower. And if this data is correct, then yes it would mean that when run at the same speed, it performs slightly faster, but when both run at their stock speeds it would fall a little ways behind, but they're not saying that it WONT do that; that's not what this release was saying at all.

[Morgan MLGman]

19 minutes ago, zMeul said:

that CPU has a base clock of 3.2Ghz and a boost clock of 4Ghz

AMD's benchmarking was done at 3Ghz, by literally gimping the Intel CPU - lower freq translates into lower IPC

 

so what is AMD trying to prove here, that their ZEN based CPU has same IPC (or better) than a i7 6900K?

because that's false and it's a marketing move to distract from the real performance of ZEN - ZEN isn't at Broadwell-E level

You people are unnecesarily bashing AMD. Why in the world would you say that they "gimped" a 6900K?

 

If your statement was correct, they would have tested both in cinebench and said "Our Summit Ridge Zen 8c/16t CPU is faster than Intel's Broadwell-E 6900K! We're awesome" and "forgot" to mention that they downclocked the 6900K to 3gHz for those tests.

 

13 minutes ago, zMeul said:

no, you are missing the point

because if you downclock, you get lower IPC and that is getting in the range of previous generation of Intel CPUs, like Haswell-E, for example - and that is what AMD needs to hide

 

The point of this test was to showcase IPC improvements made in Zen, that it's really 40% faster IPC-wise than Excavator. Their test CPU was at 3gHz, the only Broadwell-based CPU that features 8 cores and 16 threads is the 6900K. To showcase IPC improvements, they HAD to dowcnlock it to the same levels (provided that they were unable to reach 6900K clocks with the engineering sample they showcased).

 

Zen is still not a finished product. They've probably got only engineering samples now. They might not overclock well, maybe they just couldn't get it to 4gHz for example? Have you even thought about that? Or have you gone straight to hating on AMD as usual?

 

Besides, IPC does not get lower with clockspeed. IPC is IPC, clock speed is clock speed. Higher clock speed = more performance, but IPC stays the same, at least AFAIK.

[ThinkWithPortals]

8 minutes ago, zMeul said:

no, you are missing the point

because if you downclock, you get lower IPC and that is getting in the range of previous generation of Intel CPUs, like Haswell-E, for example - and that is what AMD needs to hide

Nope. Lowering clockspeed, IIRC, doesn't decrease IPC. The amount of instructions per cycle a CPU can carry out (IPC) is a constant, but underclocking reduces the number of cycles that can be carried out in a certain span of time.

IPC remains the same, it's the amount of cycles that changes with clockspeed.

[BrightCandle]

1 minute ago, ThinkWithPortals said:

Lowering clockspeed, IIRC, doesn't decrease IPC. The amount of instructions per cycle a CPU can carry out (IPC) is a constant, but underclocking reduces the number of cycles that can be carried out in a certain span of time.

IPC remains the same, it's the amount of cycles that changes with clockspeed.

It might increase slightly, CPU's don't scale with clockspeed perfectly due to the memory latency and bandwidth not increasing proportionately, decreasing the clockspeed will produce marginally higher IPC.

[Lawliet93]

Some people don´t even know, what IPC means... let me tell it to you simply, so that you can actually understant. Per core performance=IPC*Clock speed. An i7 6900k @4GHz has the same IPC as an i7 6900k@3.0GHz, but the 4GHz one has better single core perf. because of higher clocks...

[zMeul]

@ThinkWithPortals @Atmos @Paragon_X

by decreasing the clock you increase the response time of the transistor gate - and that leads to decreased IPC

 

if I'm wrong, I'm wrong, but from what I read so far I shouldn't be

[ThinkWithPortals]

Just now, zMeul said:

 

Can you drop me a link to what you're reading?

[zMeul]

Just now, ThinkWithPortals said:

Can you drop me a link to what you're reading?

right now?! I've read this shit long time ago not now

[ThinkWithPortals]

Just now, zMeul said:

right now?! I've read this shit long time ago not now

Sorry, I misunderstood, I thought had just read that somewhere.

[Morgan MLGman]

4 minutes ago, BrightCandle said:

It might increase slightly, CPU's don't scale with clockspeed perfectly due to the memory latency and bandwidth not increasing proportionately, decreasing the clockspeed will produce marginally higher IPC.

I don't think it's correct. CPUs don't scale with clockspeed perfectly, true, but I don't think it's related to IPC. IIRC it's true that you get a smaller performance increase when overclocking from 5 to 5,2gHz, instead of 3gHz to 3,2gHz, though within the range of 3gHz (AMD's test setups) and the Turbo clock of the 6900K (4gHz) the difference should be very, very marginal. AFAIK it starts decreasing much more significantly above 4,5gHz.

[zMeul]

8 minutes ago, ThinkWithPortals said:

Sorry, I misunderstood, I thought had just read that somewhere.

this should be "easily" (not really) replicated on a CPU by running a predetermined set of instructions on a single thread with lower clock (base clock) then run it again at boost clock

[xgn]

58 minutes ago, Nena360 said:

AMD claimed a FX-8370 4.3GHz gets 638, yet mine gets 680...

The intel cpu is quadcore. You cannot compare octa core vs quad. There are reasons why you would use quadcore over octa core.

[Morgan MLGman]

4 minutes ago, zMeul said:

this should be "easily" (not really) replicated on a CPU by running a predetermined set of instructions on a single thread with lower clock (base clock) then run it again at boost clock

You said this: "AMD lowered the clockspeeds to show the Zen CPU have higher IPC than Broadwell-E, it wouldn't be true at higher speeds and it would come around Haswell levels", this DEFINITELY ISN'T TRUE. If it beats Broadwell at 3gHz, it's gonna beat Broadwell at 4gHz, 5gHz and so on.

 

My bet goes to this: They needed to downclock the 6900K because their Engineering Samples of the 8c/16t Summit Ridge Zen CPU are not capable of hitting 4gHz. That 1. doesn't mean Broadwell would beat it at 4gHz, it wouldn't. and 2. doesn't mean that the final product won't be capable of such overclocks, as Zen is still not a finished product.

 

1 minute ago, xgn said:

The intel cpu is quadcore. You cannot compare octa core vs quad. There are reasons why you would use quadcore over octa core.

And yet the FX was and still is, cheaper than the 4690K

[Atmos]

11 minutes ago, zMeul said:

@ThinkWithPortals @Atmos @Paragon_X

by decreasing the clock you increase the response time of the transistor gate - and that leads to decreased IPC

 

if I'm wrong, I'm wrong, but from what I read so far I shouldn't be

I can find no mention of anything like that in any of my resources. Theoretically, yes, that could introduce latency issues, but they would be so minuscule that it would barely shift performance to one side or the other.  The only thing I can find that even resembles anything like that is a paper written back in '99, hardly up to todays technology.

[Prysin]

1 hour ago, zMeul said:

what did you expect?! it's AMD who we're talking about

no benchmark done by AMD should be trusted

lol fanboy much?

 

28 minutes ago, zMeul said:

no, you are missing the point

because if you downclock, you get lower IPC and that is getting in the range of previous generation of Intel CPUs, like Haswell-E, for example - and that is what AMD needs to hide

 

also, Kabilake is just around the corner and will probably see an earlier release than ZEN - that's another IPC increase over Broadwell-E

 

so ... who's AMD kidding!?

no, you are  again just acting as a mahoosive troll, spreading bullshit intentionally.

 

Lower clocks does not lower IPC

IPC stands for "Instructions per clock/cycle"... meaning this is how much you can do PER MHz... If you lower clocks you get lower CORE PERFORMANCE. But IPC is a fixed constant and does not change with clocks.

[Guest]

Nice. Will definitely be hopping over to Zen when it's released. 

 

 

[Prysin]

5 minutes ago, xgn said:

The intel cpu is quadcore. You cannot compare octa core vs quad. There are reasons why you would use quadcore over octa core.

depending on the workload, FX8 is either a "quad" or a "octa".

 

Under integer workloads, it is a octa core. Under floating point/vectorized workloads its a quad core. This is due to the unique way the current FX linup is designed.

[zMeul]

12 minutes ago, Morgan MLGman said:

You said this: "AMD lowered the clockspeeds to show the Zen CPU have higher IPC than Broadwell-E

I haven't said that at all 

what I said is that AMD lowered the Intel's CPU clock to make ZEN look like it has similar IPC

[SLAYR]

26 minutes ago, zMeul said:

@ThinkWithPortals @Atmos @Paragon_X

by decreasing the clock you increase the response time of the transistor gate - and that leads to decreased IPC

 

if I'm wrong, I'm wrong, but from what I read so far I shouldn't be

The difference would be so minimal it would not even matter, not even 5%, if you graph cinebench single threaded results vs clockspeed for pretty much any cpu it is a very linear graph.

[zMeul]

14 minutes ago, Atmos said:

I can find no mention of anything like that in any of my resources. Theoretically, yes, that could introduce latency issues, but they would be so minuscule that it would barely shift performance to one side or the other.  The only thing I can find that even resembles anything like that is a paper written back in '99, hardly up to todays technology.

it's not theoretical , it's actual practice

IPC is a theoretical number because in real life a CPU architecture is depends on various variables, it's at the mercy of the OS and the HW scheduler in that CPU - no two runs will yield same result

[Morgan MLGman]

1 minute ago, zMeul said:

I haven't said that at all 

what I said is that AMD lowered the Intel's CPU clock to make ZEN look like it has similar IPC

This is the same. Looking at the showcase, Zen has higher IPC than Broadwell. And my point still stands, you're wrong here because if Zen is faster at 3gHz, it will still be faster at 4gHz and 5gHz than Broadwell.

 

Both architectures are SMT, what makes you think that the Broadwell CPU will magically scale better?

[DocSwag]

This seems like a legit benchmark so I'm assuming from this zen has similar ipc to broadwell but is a bit behind skylake. 

 

Problem is we still have no idea about tdp or clock speeds. What if zen can reach 5ghz? What if it can only reach 3.5? Dang it amd we need information!

[zMeul]

6 minutes ago, Morgan MLGman said:

This is the same. Looking at the showcase, Zen has higher IPC than Broadwell. And my point still stands, you're wrong here because if Zen is faster at 3gHz, it will still be faster at 4gHz and 5gHz than Broadwell.

O'really!?

so how come AMD hasn't tested their ZEN at the base clock of the Broadwell-E CPU? that would've been a really compelling case for ZEN

 

Mohamed goes to the mountain or the mountain goes to Mohamed?

in this case, AMD brought the mountain to ZEN - why?! the answer is obvious

[awesomeness10120]

I'll believe it when I see it.

[Prysin]

2 minutes ago, zMeul said:

O'really!?

so how come AMD hasn't tested their ZEN at the base clock of the Broadwell-E CPU?

 

Mohamed goes to the mountain or the mountain goes to Mohamed?

i this case, AMD brought the mountain to ZEN - why?! the answer is obvious

because it is an engineering sample? Because as it stands the current production setup still need tweaking to perfect the chip for final consumer release? Because tweaking a litography process takes more then a month, aka the time that we have know about ZEN engineering samples at ~3GHz (it takes 3 months from you change something in manufacturing process, to you see the effects of the change in actual products. That is based on Intel and TSMC manufacturing schedules. I find it very hard to believe that GLOFO, which are known for their issues, is able to change their process faster then Intel or TSMC.

 

I would actually bet you money that they could run that ZEN ES at 3.6GHz baseclock, which is what Broadwell-e runs at, and the results would still be within 3-5% margin of error. AKA well within the "Broadwell esque" performance that they are showing here.

 

AMD showed us ONE run in blender. If they had done multiple, i bet that the two CPUs would trade blows. Maybe ZEN is 1-2% faster then Broadwell. Just maybe, but even then that is ENOUGH. Even if Broadwell-E ran at 4Ghz and ZEN at 3GHz, it would still show performance that would be GOOD ENOUGH to be a competing product.