Intel Skylake / Kaby Lake

Page 544 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
https://forums.anandtech.com/thread...s-out-page-501.2428363/page-539#post-38993092

I was quoting this user when i posted the old reply that you quoted, if you gonna quote me at least have an idea of what or to who im talking to.

Alright then. But . . .

dont try to damage control it.

Who's damage controlling? Someone called out Intel for "anti-consumer" tactics, and the only recourse here is to blame some other company for the same thing? One that's easily-avoided? Cmon now.

The fact is that nobody has ever had to buy AMD. It's easy to stay away from their products if that is your intent. Granted it was a bit harder back in the P4D days, but Intel still sold the majority of chips and was better-represented in OEM systems. Lots of people found out how to avoid AMD.

Non-competitive, poorly-priced products, which you have the choice not to buy, doesn't exactly constitute "anti-consumer".

Things like excluding Optane support from Kaby Lake Pentiums, and leaving mainstream Skylake out for no apparent reason despite no additional hardware requirements, on the other hand, can be termed as "anti-consumer".

That is probably not the worst thing Intel has ever done, and it is easily avoided.
 
  • Like
Reactions: Drazick

csbin

Senior member
Feb 4, 2013
838
351
136
CFL 8700K(?) @5G vs 1700X

CineBench 15 Multithreaded Results:http://www.anandtech.com/bench/CPU/1603


38cdex7g9ibz.png
 
  • Like
Reactions: TheF34RChannel

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
So if I read this correctly, a STOCK 1700x is within 96% of the score of a 5 ghz overclocked 8700k ??? ( 4271/4467 = 95.6%)

Not quite sure what are You expecting? Rendering is the perfect workload for Ryzen core as long as it hasn't got vectors longer than 128bit. So 5ghz / 1.33 => 8 core (that has 1.33 more cores than 8700k) only needs ~3.75Ghz clock to match. And that is around where 1700x is in turbo?

Video encoding in x264 or x265 would make much more interesting case for comparison: Intel would drop clocks hard due to AVX offset, AMD would hit nasty penalties due to longer vectors tying execution units.
 

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136

95W TDP at stock versus 5GHz OC @ 1.415V+? Performance may be good, but that's throwing efficiency out the window.

OC TDP ~= 95W TDP * (5000MHz / 4000MHz) * (1.415V / 1.200V)^2
= 165W

So if I read this correctly, a STOCK 1700x is within 96% of the score of a 5 ghz overclocked 8700k ??? ( 4271/4467 = 95.6%)
And would use less than 2/3 the power to perform the same work, per my back of the napkin calculations above.
 
  • Like
Reactions: Drazick and psolord

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
Not quite sure what are You expecting? Rendering is the perfect workload for Ryzen core as long as it hasn't got vectors longer than 128bit. So 5ghz / 1.33 => 8 core (that has 1.33 more cores than 8700k) only needs ~3.75Ghz clock to match. And that is around where 1700x is in turbo?

Video encoding in x264 or x265 would make much more interesting case for comparison: Intel would drop clocks hard due to AVX offset, AMD would hit nasty penalties due to longer vectors tying execution units.
I have been HEARING that Intels chips have better IPC, but it seems like they are tied by your number, but using 2/3rd's the power by @IEC's calculations. Very good showing for AMD IMO.
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
20,839
3,174
126
you cant really use that coffee lake benchmark.
i have said this many times, cpu's which show 0000 are mostly inhouse pre vendor release.
They are on the most infant of steppings the cpu has.

Usually after a few more stepping revisions the cpu gets its brand code... in this case the 8000 series.
So a new stepping can do wonders on a cpu.

Look at Q6600 G0, and the i7 920 D0.


Now that 7920X is a different story tho...
That you can take as white paper, as that is a vendor ES, and will act / behave near launch specs.
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
I have been HEARING that Intels chips have better IPC, but it seems like they are tied by your number, but using 2/3rd's the power by @IEC's calculations. Very good showing for AMD IMO.

Umm, I don't think anyone in these forums is doubting that AMD is within 10-15% of general computing IPC. There are some outliers like AVX2 workloads for Intel or Cinebench style workloads that love AMD's 4xFP pipes, but in general they are within 15% of Skylake.

Sadly this great IPC showing can't undo clock deficit, and that is where things get ugly. If Intel's 6 core is competing with 8 core in rendering and also has 30-40% ST advantage due to much higher clocks...
 
  • Like
Reactions: rgba and Sweepr

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
I have been HEARING that Intels chips have better IPC, but it seems like they are tied by your number, but using 2/3rd's the power by @IEC's calculations. Very good showing for AMD IMO.


Cinebench is a well known best case bench for Ryzen, it doesn't reflect most client workloads. And of course Intel has a much better IPC, at least their Mainstream part. Do you try to deny that because of Cinebench? You are cherry picking in this case.
 

TheF34RChannel

Senior member
May 18, 2017
786
309
136
95W TDP at stock versus 5GHz OC @ 1.415V+? Performance may be good, but that's throwing efficiency out the window.

OC TDP ~= 95W TDP * (5000MHz / 4000MHz) * (1.415V / 1.200V)^2
= 165W

Overclocking and efficiency don't belong in the same sentence. As soon as you do the former the latter goes out the window. I thought that was pretty much self explanatory?
 
  • Like
Reactions: mikk

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
Umm, I don't think anyone in these forums is doubting that AMD is within 10-15% of general computing IPC. There are some outliers like AVX2 workloads for Intel or Cinebench style workloads that love AMD's 4xFP pipes, but in general they are within 15% of Skylake.

Sadly this great IPC showing can't undo clock deficit, and that is where things get ugly. If Intel's 6 core is competing with 8 core in rendering and also has 30-40% ST advantage due to much higher clocks...

Actually Zen has a solid roadmap to address the weakness with clocks. We already have Zen on 14nm+ (Pinnacle Ridge) lined up for Q1 2018. I think 14nm+ will bring a good 12-15% higher clocks . So 4.5 -4.6 Ghz should be the avg OC on those chips. Then we have 7nm Zen 2 with microarchitectural improvements. Zen 2 CPU is being made using GF 7LP process which is designed for 5 Ghz operation. Coffeelake won't be out till October . Thats 6 months after Zen launched. Zen on 14nm + in Q1 2018 should cut that clock deficit against Coffeelake by roughly 50% . Zen 2 on GF 7LP vs Icelake on Intel 10+ in H1 2019 will be when things get really interesting. AMD is most likely to go for a 6 core CCX and have a 12 core Ryzen 2 with IPC improvements and capable of OCing to 5 Ghz.
 

R0H1T

Platinum Member
Jan 12, 2013
2,582
162
106
The scores for Threadripper on GB have been floating around for a while. Either they are not real or GB is only reading half the cores. An 1800X scores around that amount by itself on the multi-core part. http://browser.geekbench.com/v4/cpu/search?q=1800x
GB4 doesn't scale well past 8~12 cores, check any Skylake Xeon results & you'll see the same trend. Though the TR's FP score, ST & MT, is really low which could imply a number of different things.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
Cinebench is a well known best case bench for Ryzen, it doesn't reflect most client workloads. And of course Intel has a much better IPC, at least their Mainstream part. Do you try to deny that because of Cinebench? You are cherry picking in this case.
Ok so let's take a few single-threaded consumer workloads, not including things which usually run multi-threaded but may be restricted to a single-thread by the user, like encoding. I guess these are fairly representative, outside of corner cases like emulation and legacy games:

87579.png


7740X is 23% faster than 1600X, but 675ms isn't anything earth-shattering. Say you open 100 of these types of PDFs in a day - what does it amount to - 67.5s in a day? Meh.

Suppose you're a consumer who has a YouTube channel doing gaming performance reviews. This would be relevant:

87580.png


Again, comparing the 7740X and 1600X, the former is 16% faster. So you do 10 benchmarks which is 10X the data processing, and the 7740X saves you only ~90s? Some huge drawback if you have the 1600X I guess.

Now to some MP3 conversion:

embed.php


What happened to the the extra clock speeds and IPC?
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
What happened to the the extra clock speeds and IPC?

In the first two benchmarks even a mediocre clocked i5-7600 beats all AMD Ryzen offerings quite comfortably and in the last benchmark 7700k with only 4 cores is able to beat AMDs fastest 8 core offering and you are still asking what happened with the extra IPC and clock speed - LOL. Thanks for the proof that Cinebench is a best case bench for AMD, but we already knew that. By the way are you still in denial mode that Intel can't do Coffeelake with 12 MB L3? You were very silent about this lately. What happened?
 

nvgpu

Senior member
Sep 12, 2014
629
202
81
To admin, please remove Ryzen posts from this thread, you specifically told them not to post about Ryzen here but they keep doing it, including the mod Markfw.



You have an issue with the moderation? Make an MD thread.
Mod callouts are absolutely not allowed. But you knew that already.


esquared
Anandtech Forum Director
 
Last edited by a moderator:

TheGiant

Senior member
Jun 12, 2017
748
353
106
Does anybody actually consider price ? or efficiency ? Everybody used to down AMD for efficiency, but now that AMD is better we hear nothing of it. Or how AMD are so much less expensive in general.
Well we will see efficiency with 6C CFL- someone should test the voltage needed to reach 3,9GHz and then the real power consumption.

The high freq chips from intel are beyond the optimal point...imo
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
To admin, please remove Ryzen posts from this thread, you specifically told them not to post about Ryzen here but they keep doing it, including the mod Markfw.
I did not start the benchmarks and comments, I just responded to them.
 
  • Like
Reactions: Drazick