Ryzen vs Coffee Lake

juanrga said:
Can I change the congratulations by a requirement?

The requirement is that here and thereafter people stop giving only CineBench scores for RyZen and ignoring any other benchmark where RyZen looks worse. To get he performance of a chip, any chip, one has to quote a collection of benchmarks, including games.
Click to expand...

No, absolutely not.
 
Honestly, I really very much enjoy my Ryzen systems. However, if I had stuck with my FX 8350 build for my day to day machine that I do not use at home, I would have been perfectly fine and saved myself some money as well.
 
Why not just test both AMD and Intel at the same clocks to see the real IPC of both designs?
 
I would rather see the test 4ghz vs 4ghz. max Ryzen clocks vs intel 4ghz. I think that would be a good test to see, but I honestly expect Intel to still have the IPC advantage.
 
Brackle said:
I would rather see the test 4ghz vs 4ghz. max Ryzen clocks vs intel 4ghz. I think that would be a good test to see, but I honestly expect Intel to still have the IPC advantage.
Click to expand...

3.5GHz vs 3.5GHz test have been given here. 4GHz will only make things worse for AMD.
 
legcramp said:
I apologize for calling you juanrga's little bitch the other day.
Click to expand...

I'm just saying, stuff like Cinebench doesn't really show how good stuff is anymore as well. Maybe a few other other benchs like Handbrake or something, kinda gives the idea a lot closer and not that far off. You're Mom Man Lol
 
juanrga said:
3.5GHz vs 3.5GHz test have been given here. 4GHz will only make things worse for AMD.
Click to expand...
We all know that Intel enjoys a huge clock speed advantage over AMD and many seem to think that is IPC but on average from what I have seen....IPC is 3-5 percent with clock speed being greater for one side than the other especially in gaming not the huge number being bandied around here.
 
Batboy88 said:
I'm just saying, stuff like Cinebench doesn't really show how good stuff is anymore as well. Maybe a few other other benchs like Handbrake or something, kinda gives the idea a lot closer and not that far off. You're Mom Man Lol
Click to expand...

Unless, of course, Intel was showing up as doing better at it across the board, then it would still be a good benchmark. :rolleyes: These benchmarks are good because the are generic or are at least supposed to be.
 
ManofGod said:
HAPPY NEW YEAR AMD! May your clocks go high and your temps remain cool. :)
Click to expand...

You're Clocks will go High if you get a good Binned chip. Cooling is minimal Problems on that end as well. Vs this end it's a lot of just ok and average chips and they all could use delid and change that damn tim, it must be crap. sure want to run Hot alright.
 
quasar56 said:
We all know that Intel enjoys a huge clock speed advantage over AMD and many seem to think that is IPC but on average from what I have seen....IPC is 3-5 percent with clock speed being greater for one side than the other especially in gaming not the huge number being bandied around here.
Click to expand...

No one is confounding clocks with IPC.

When a 6-core CFL is ~12% faster than 8-core Zen on Blender, each CFL core is ~50% faster and correcting for the clocks we obtain ~18% IPC gap.

Perf_Zen = 8 * IPC_Zen * Freq_Zen

Perf_CFL = 6 * IPC_CFL * Freq_CFL
= 6 * (1.18 * IPC_Zen) * (1.27 Freq_Zen)
= 8.99 * IPC_Zen * Freq_Zen
= 1.12 Perf_Zen


In favorable benches as CineBench (favorable to Zen), the IPC gap is 11%. Average IPC gap on Windows is 10--20%. It increases to 24% under linux.
 
Last edited:
In 2018 seems like nobody seems to understand that 22-33% clock boosts alter scores a 3.6ghz coffee lake does not beat a 1700 at 3.4ghz in blender. Since perf is largely affected by clockspeed you are not actually doing clock vs clock but rather leveraging a 6 core at 4.8 against 8 at 3.4 or 3.6.

In the same way a coffee lake at 4.8ghz beats a low clocked all core skylake x 8 core CPU or a 10 core Broadwell. You just repeat the obvious.
 
Batboy88 said:
I'm just saying, stuff like Cinebench doesn't really show how good stuff is anymore as well. Maybe a few other other benchs like Handbrake or something, kinda gives the idea a lot closer and not that far off. You're Mom Man Lol
Click to expand...

Cinebench is modeled on a real animation and cgi software suite, its very realistic and very scaleable, but not many can afford it
 
I almost bought an 8700k last week but the MC sale for the 1800X at $299 was too good to pass up. Doesn't really make a difference to me for games at 1440p, my GPU is the limiting factor anyways but I was able to get the chip and board for almost the same price as the 8700k alone.

I've worked with Ryzen several times already so I know what to expect, probably try out the 9700k when it comes out at the end of this year.
 
ManofGod said:
Unless, of course, Intel was showing up as doing better at it across the board, then it would still be a good benchmark. :rolleyes: These benchmarks are good because the are generic or are at least supposed to be.
Click to expand...

Unless, of course, Intel was showing up as doing better at it across the board, then it would still be a bad benchmark. :rolleyes: Remember all that people that wrongly pretended CineBench was "Intel-biased" by using the Intel-complier. Now, magically, the same version of Cinebench is no more biased for that people.

We don't work with suppositions here. We test RyZen with different benchmarks and see that Cinebench is not "generic" anymore, but only an outlier case that disagrees with many other benchmarks:

Even though it has two fewer cores than the Ryzen 1800X (a CPU that costs a hefty £437), the 8700K comes in faster in many production workloads. It's four seconds quicker in Blender at stock, and 11 seconds quicker when overclocked. It's faster at Handbrake video encoding too, and miles ahead in 7-Zip's synthetic benchmark, which tends to favour clock speed even in multithreaded mode.

It's only in PovRay and Cinebench that 1800X comes out on top—and only then by a small amount.
Click to expand...
 
Last edited:
OrangeKhrush said:
In 2018 seems like nobody seems to understand that 22-33% clock boosts alter scores a 3.6ghz coffee lake does not beat a 1700 at 3.4ghz in blender. Since perf is largely affected by clockspeed you are not actually doing clock vs clock but rather leveraging a 6 core at 4.8 against 8 at 3.4 or 3.6.

In the same way a coffee lake at 4.8ghz beats a low clocked all core skylake x 8 core CPU or a 10 core Broadwell. You just repeat the obvious.
Click to expand...

So 2018 starts with the same excuses than 2017. Again the excuse that i7-8700k is running above the max turbo, whereas the 1800X is running below the base clock.

And of course continued ignoring of clock-for-clock benchmarks showing that the average IPC gap goes from ~15% under windows to ~30% under linux.

OrangeKhrush said:
Cinebench is modeled on a real animation and cgi software suite, its very realistic and very scaleable, but not many can afford it
Click to expand...

The problem with CineBench is not if it is realistic or not. The problem is that it doesn't represents average behavior. RyZen runs better under CineBench than under other workloads.

Using only CineBench to favor AMD is so biased as using only Audacity to favor Intel.

clock-cb15-2.png


clock-audacity.png


To get the average IPC gap between Intel and AMD CPUs, we have to use a collection of different workloads and take the average of all them. This is obvious. And the only people rejecting this elementary idea is people that don't want the real IPC gap to be reported. People that only want CineBench scores because Zen looks better. This is pretty evident to me as well.
 
Last edited:
juanrga said:
So 2018 starts with the same excuses than 2017. Again the excuse that i7-8700k is running above the max turbo, whereas the 1800X is running below the base clock.

And of course continued ignoring of clock-for-clock benchmarks showing that the average IPC gap goes from ~15% under windows to ~30% under linux.



The problem with CineBench is not if it is realistic or not. The problem is that it doesn't represents average behavior. RyZen runs better under CineBench than under other workloads.

Using only CineBench to favor AMD is so biased as using only Audacity to favor Intel.

clock-cb15-2.png


clock-audacity.png
Click to expand...

Awesome, Intel is faster at MP3 encoding. Now we know and will adjust our world view to account for this monumental achievement.
 
Kaos_Drem said:
Only difference here is one is a synthetic benchmark whereas the other is an actual workload. While sure nobody really does 128kbps MP3 anymore, it is still actually doing work rather than just arriving at a predetermined destination faster.
Click to expand...

Synthetic benchmarks, when well-made, can represent a collection of actual workloads. This is not the case with CineBench and RyZen. RyZen generally run worse under actual workloads than under Cinebench. Of course, we can still use Cinebench to complement a collection of actual workloads and synthetic benchmarks. What we cannot do is to give only CineBench scores and ignore anything else.
 
Dermac said:
Awesome, Intel is faster at MP3 encoding. Now we know and will adjust our world view to account for this monumental achievement.
Click to expand...

I am not really surprised that you ignored completely my point. The funny part is that rest of world knows Intel is faster in many more workloads than MP3 encoding and sales reflect that. CoffeeLake is #1 in sales across the world and beating RyZen even at Mindfactory. ;)
 
juanrga said:
I am not really surprised that you ignored completely my point. The funny part is that rest of world knows Intel is faster in many more workloads than MP3 encoding and sales reflect that. CoffeeLake is #1 in sales across the world and beating RyZen even at Mindfactory. ;)
Click to expand...

Cool story bro
 
for me all that matter is gaming performance. and amd is way behind there. my 5820k is stronger per core then amd still today, and mine cpu is 2 years already. and look to live atleast 2 more years as a top end cpu. hardly is any point to replace with newest intelol. more logical to add another 1080ti realistically, and it would cost the same if to replace cpu,mobo and ram.
 
juanrga said:
I am not really surprised that you ignored completely my point. The funny part is that rest of world knows Intel is faster in many more workloads than MP3 encoding and sales reflect that. CoffeeLake is #1 in sales across the world and beating RyZen even at Mindfactory. ;)
Click to expand...

You talk of all the suites and how cinebench favours AMD but excuse piss poor synthetics like lame as being one that favours intel. Even per gamersnexus extensive re review of the 8700K per blender AMD holds its own against higher clockspeed and in both blender and cinebench the 8 core AMD is using less power to do the workload. Intels power usage onhigh stress loads is scary, the 7980/60 ramping over 500w to threadripper 240w.

Cinebench and blender are just fine and it would perhaps favour AMD's workhorse architecture which is rather economic

As for market talk, it's well established that you have no idea what you are talking about so I'll just skirt around that clusterfuck
 
Brackle said:
Well with the current security flaw that has now been uncovered. It is possible that Intel will receive a 30-35% performance hit for fixing the issue. Now I don't know how it would affect gaming, but this is something to add to the discussion. IF intel takes a 30% to fix the issue, then that would make Ryzen even more competitive.

https://www.hardocp.com/news/2018/01/02/massive_intel_hardware_bug_may_be_on_horizon
Click to expand...

That would make it superior. Plus new Ryzen chips are due to come out as well, that would be a nightmare for Intel. Cause if I had a Intel chip I would be demanding a refund for that kind of performance loss.
 
The bug seems to be more towards enterprise and datacenter systems rather than general home users, though it would be interesting if the hit is as much, that would hurt a little.
 
OrangeKhrush said:
You talk of all the suites and how cinebench favours AMD but excuse piss poor synthetics like lame as being one that favours intel.
Click to expand...

LOL I wrote "Using only CineBench to favor AMD is so biased as using only Audacity to favor Intel." but you pretend that I am excusing Audacity to favor Intel.

Did you consider a visit to the optician? Or are the red glasses what makes you cannot read properly others' messages never?

OrangeKhrush said:
Even per gamersnexus extensive re review of the 8700K per blender AMD holds its own against higher clockspeed and in both blender and cinebench the 8 core AMD is using less power to do the workload. Intels power usage onhigh stress loads is scary, the 7980/60 ramping over 500w to threadripper 240w.
Click to expand...

And another spin. It is proven you wrong about IPC, then you spin to performance. It is proven you wrong about performance, then you spin to power. If we start discussing power, you will soon spin to anything else.

OrangeKhrush said:
As for market talk, it's well established that you have no idea what you are talking about so I'll just skirt around that clusterfuck
Click to expand...

Data is here out. ;)
 
Brackle said:
Well with the current security flaw that has now been uncovered. It is possible that Intel will receive a 30-35% performance hit for fixing the issue. Now I don't know how it would affect gaming, but this is something to add to the discussion. IF intel takes a 30% to fix the issue, then that would make Ryzen even more competitive.

https://www.hardocp.com/news/2018/01/02/massive_intel_hardware_bug_may_be_on_horizon
Click to expand...

The patch performance penalty must be 0--5% for consumer, and the penalty affects both AMD and Intel CPUs.
 
juanrga said:
LOL I wrote "Using only CineBench to favor AMD is so biased as using only Audacity to favor Intel." but you pretend that I am excusing Audacity to favor Intel.

Did you consider a visit to the optician? Or are the red glasses what makes you cannot read properly others' messages never?



And another spin. It is proven you wrong about IPC, then you spin to performance. It is proven you wrong about performance, then you spin to power. If we start discussing power, you will soon spin to anything else.



Data is here out. ;)
Click to expand...

Yeah data is out, like a distributors month to month actual sales showing Ryzen outselling intel in net for their roll over, of course there are millions of distributors that can probably share the same data instead of the rather small sample provided by amazon and all those dubious penny pinchers. I can probably get 1000 people who buy through distributor before i find anyone that uses Amazon or Newegg, granted that outside the US those are practically useless to order through unless you want to pay double the cost on import, customs and VAT. Locally from people I know in the game of reselling Ryzen is selling very well.
 
juanrga said:
The patch performance penalty must be 0--5% for consumer, and the penalty affects both AMD and Intel CPUs.
Click to expand...

Well if AMD is accurate in stating that their products does not have this bug, then you can safely set the -nopti boot option and have normal performance, at least in linux.
We currently have no information on how microsoft is going to deal with this.
 
juanrga said:
The patch performance penalty must be 0--5% for consumer, and the penalty affects both AMD and Intel CPUs.
Click to expand...

Performance penalty is 50% in some workloads - like file I/O. SSD users will see a huge hit. 0% in other stuff (like encoding), no data on gaming or rendering yet. AMD does not have the issue, so no performance penalty unless they stupidly apply the intel fix for AMD processors (which would be unlikely).

https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

page file isolation should not be enabled for AMD processors:

https://lkml.org/lkml/2017/12/27/2
 
Verado said:
Well if AMD is accurate in stating that their products does not have this bug, then you can safely set the -nopti boot option and have normal performance, at least in linux.
We currently have no information on how microsoft is going to deal with this.
Click to expand...

One AMD engineer claims AMD CPUs aren't affected. But other devs disagree. This is the reason why all x86 CPUs are being treated as vulnerable today. If tomorrow it is confirmed that some CPUs aren't affected modifications to the kernel can be made for those CPUs.

bobzdar said:
Performance penalty is 50% in some workloads - like file I/O. SSD users will see a huge hit. 0% in other stuff (like encoding), no data on gaming or rendering yet. AMD does not have the issue, so no performance penalty unless they stupidly apply the intel fix for AMD processors (which would be unlikely).

https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

page file isolation should not be enabled for AMD processors:

https://lkml.org/lkml/2017/12/27/2
Click to expand...

The phoronix link that you give shows there is no performance lose for the system with the Toshiba SSD in the FS-Mark, so your generic "SSD users will see a huge hit" isn't true. Also you are only mentioning a synthetic disk I/O that doesn't represent most real-life workloads. From the review: "While applications mostly limited to user-space activity should see minimal change (if any) in performance."

I know the second link; it is the comment from the AMD engineer that claims there is no problem. What he says hasn't been confirmed, and, one of the responses in the second link states the obvious: "This is a rather wide class of issues and I would rather not just hard-code it in a way that we say one vendor has never and will never be affected."
 
juanrga said:
One AMD engineer claims AMD CPUs aren't affected. But other devs disagree. This is the reason why all x86 CPUs are being treated as vulnerable today. If tomorrow it is confirmed that some CPUs aren't affected modifications to the kernel can be made for those CPUs.



The phoronix link that you give shows there is no performance lose for the system with the Toshiba SSD in the FS-Mark, so your generic "SSD users will see a huge hit" isn't true. Also you are only mentioning a synthetic disk I/O that doesn't represent most real-life workloads. From the review: "While applications mostly limited to user-space activity should see minimal change (if any) in performance."

I know the second link; it is the comment from the AMD engineer that claims there is no problem. What he says hasn't been confirmed, and, one of the responses in the second link states the obvious: "This is a rather wide class of issues and I would rather not just hard-code it in a way that we say one vendor has never and will never be affected."
Click to expand...

OK, users with NVME SSD drives will see a big hit - ie anyone who shelled out for super fast IO will see it crushed. If you have an old SATA SSD or a spinner, you won't see much of an issue, if at all, unless reading a bunch of small files.

Compiler times for some stuff was hurt badly, Postgres and Redis hit as well. I run a bunch of spark stuff, which I'd imagine will see the hit, especially if run on top of hadoop (ie, loading a bunch of small files from an SSD). ML algorithms probably hit, especially stuff that uses graphs. Having my load times double when booting or running stuff off of my nvme drives would piss me off.

So far no indication that there's an issue on any AMD processors.
 
I guess this kind of makes the original thread title irrelevant, especially if intel performance drops 5%, then coffee lake is back to skylake performance......
 
juanrga said:
https://www.computerbase.de/2018-01/intel-cpu-pti-sicherheitsluecke/
https://www.hardwareluxx.de/index.p...er-sicherheitsluecke-im-prozessor-design.html

Average performance hit on i7s running patched Windows is nearly zero.
Click to expand...

You're hopeless. How about you post some of your precious 720p gaming graphs that take the GPU bottleneck out of the picture and get back to us before you claim it is "nearly zero" (specifically in gaming). AC Origins takes a hit at 1080p on low graphics mode of about 3% and no hit on 1080p on high graphics where likely the GPU is the bottleneck in the computerbase article. All three of the hardwareluxx 1080p gaming tests show a comparable 3-5% hit without a lot of detail on their setup. If you start arguing that it is "practically zero," some people are going to argue that the current 10% at 1080p over Ryzen and post-patch 5-7% is just as "practically zero" in gaming and you shouldn't complain.
 
kirbyrj said:
You're hopeless. How about you post some of your precious 720p gaming graphs that take the GPU bottleneck out of the picture and get back to us before you claim it is "nearly zero" (specifically in gaming). AC Origins takes a hit at 1080p on low graphics mode of about 3% and no hit on 1080p on high graphics where likely the GPU is the bottleneck in the computerbase article. All three of the hardwareluxx 1080p gaming tests show a comparable 3-5% hit without a lot of detail on their setup. If you start arguing that it is "practically zero," some people are going to argue that the current 10% at 1080p over Ryzen and post-patch 5-7% is just as "practically zero" in gaming and you shouldn't complain.
Click to expand...

Just curious, which is your excuse for the 1%, 0%, 0.6%,... performance hit measured with applications?
 
You must log in or register to reply here.
Back
Top