Ryzen vs Coffee Lake

[OrangeKhrush]

Do you even read the crap you post?

"At stock, the Intel system with fewer cores is faster, but the AMD system gets faster when you overclock it!"

What about when you overclock the Intel system? Or are you not going to compare apples to apples? Is your need to 'counter-troll' so demanding that you cannot present an honest argument?


Further, with respect to memory, it's not about bandwidth with Ryzen- it's all about latency. By increasing memory clockspeed, you overclock the CPU's infinity fabric, which results in lower inter-die latency, and by using lower latency memory, you decrease memory latency on top of that, therefore lowering system latency.

Your best bet for RAM on Ryzen is the DDR4-3200 C14 stuff, which is stupid expensive, when on Intel it mostly doesn't matter, and it'll still be faster than Ryzen. If anything, AMD needs to get a massaged stepping out that isn't so dependent on nice RAM.

Maybe you are the one who doesn't read, but hardly surprising given your stance.

A 1700 stock at 3.4ghz all core is basically as fast as a stock 8700K with its 4.4ghz all core, so in a non clock vs clock the Ryzen is already impressive, then you look at the overclock scores and the Ryzen 1700 at 3.9ghz is beating the overclocked 8700K compellingly.

 

[IdiotInCharge]

3.4 --> 3.9, 15% increase.
4.4 --> 5.0, 14% increase.

'Compellingly'

Sustained boost clocks between parts- and I'd bet more 8700k parts will track closer to 5.0GHz than R7's will clock to 3.9GHz- would tell the story, and that story is not 'beating compellingly', in either case.

What we can say, is that in some compute workloads, the R7 stock vs. stock or OC vs. OC, can put out similar numbers to an 8700k. We can also say that in any single-threaded or single-thread dependent workload, the 8700k will leap ahead of the R7.

That's honest.

 

[OrangeKhrush]

3.4 --> 3.9, 15% increase.
4.4 --> 5.0, 14% increase.

'Compellingly'

Sustained boost clocks between parts- and I'd bet more 8700k parts will track closer to 5.0GHz than R7's will clock to 3.9GHz- would tell the story, and that story is not 'beating compellingly', in either case.

What we can say, is that in some compute workloads, the R7 stock vs. stock or OC vs. OC, can put out similar numbers to an 8700k. We can also say that in any single-threaded or single-thread dependent workload, the 8700k will leap ahead of the R7.

That's honest.

Again neither of those are clock vs clock and in almost all instances Ryzen is going to be significantly lower clocked against the 8700K, so it is producing performance which at the lowest end is matching a 8700K and when running a 3.9ghz OC it is faster against a overclocked 8700K. AMD's strong SMT implementation off sets single thread dependence and in work loads which I utilse the parallelism is exceptional.

On single thread side on a clock vs clock the 8700K has about 10-12% advantage though a year ago that number was around 60% given how much AMD caught up and how Ryzen represents "worst case scenario" even the worst case is punchy with intel who didn't move the marker all that much.

 

[IdiotInCharge]

Again neither of those are clock vs clock and in almost all instances Ryzen is going to be significantly lower clocked against the 8700K, so it is producing performance which at the lowest end is matching a 8700K and when running a 3.9ghz OC it is faster against a overclocked 8700K. AMD's strong SMT implementation off sets single thread dependence and in work loads which I utilse the parallelism is exceptional.

On single thread side on a clock vs clock the 8700K has about 10-12% advantage though a year ago that number was around 60% given how much AMD caught up and how Ryzen represents "worst case scenario" even the worst case is punchy with intel who didn't move the marker all that much.

So you're going to insist on distorting the facts as much as possible?

I'm a proponent of Ryzen where the workloads and costs make sense. Right tool for the job. Unfortunately, unless you go out of your way to present workloads that are niche (or outmoded like Cinebench!), the case for Ryzen for most consumers is fairly slim as consumer workloads benefit more from increased single-thread performance.

 

[OrangeKhrush]

So you're going to insist on distorting the facts as much as possible?

I'm a proponent of Ryzen where the workloads and costs make sense. Right tool for the job. Unfortunately, unless you go out of your way to present workloads that are niche (or outmoded like Cinebench!), the case for Ryzen for most consumers is fairly slim as consumer workloads benefit more from increased single-thread performance.

That would be true if it wasn't being bought but the difference is that Ryzen is being bought and in instances Threadripper. I use FL studio and will see a massive gain going to a more parallel load.

As for niche, intel has thrown out niche benches for a very long time even comparing non clock vs clock as a form of mindshare marketing, we saw how they tried to create artificial loads in their secret bunker epyc test, and even that looked bad.


Anandtech once did a 3ghz comparison at the time with Kaby, SKL, HSW, RYZEN, IVY and Sandy

ST:

7700K - 127.33
6700K - 128
8700K - 127.65
4770k - 120
1600- 124.16
1700 - 120.81
4790K - 125
3770K- 112
2600K - 108

*all scores based off anandtechs database scores clocked down to 3ghz
** ST scores rated off max single core turbo

MT:

8700K - [email protected]
1600 - [email protected]
1700 - [email protected]

Converted:

8700K - 977.44
1600 - 1034
1700 - 1254.70

From a synthetic stand point the only real limitation to AMD is the node, but on a clock vs clock perspective AMD is very close, to close for comfort. If one is to distort what is easily determinable the 8700K's MT score is concerning, if the clocks no longer compensate the lake of SMT implementation, the 8700K keels off and is soundly beaten by a 1600. Clockspeed is being used to hide no gains, Intel is going for the bang 5ghz down the pipes as this is the only thing that is saving them.

The real world limitation on Ryzen comes to which loads are sensitive with the infinity fabric, it is also not general.

On performance, being AMDs first draft on a new uarch it has been far better than expected across all loads.

 

[IdiotInCharge]

Anandtech once did a 3ghz comparison at the time with Kaby, SKL, HSW, RYZEN, IVY and Sandy

I looked. Link please.

The real world limitation on Ryzen comes to which loads are sensitive with the infinity fabric, it is also not general.

Anything with a multi-thread workload and latency sensitivity is sensitive to the performance of the infinity fabric. That's almost everything, including workloads that are also sensitive to single-thread performance.

On performance, being AMDs first draft on a new uarch it has been far better than expected across all loads.

Here, we're talking about the base architecturem I can agree: with the caveat that AMD must decouple infinity fabric performance from memory controller speed and latency for their future SKUs.

 

[OrangeKhrush]

I looked. Link please.



Anything with a multi-thread workload and latency sensitivity is sensitive to the performance of the infinity fabric. That's almost everything, including workloads that are also sensitive to single-thread performance.



Here, we're talking about the base architecturem I can agree: with the caveat that AMD must decouple infinity fabric performance from memory controller speed and latency for their future SKUs.

Ask Juanrga it was once his favourite Anandtech post, alternatively you can just do math very simple to calculate.

I am sure AMD tech is fully aware of what Ryzen's limitations were and how to improve them, maybe we will be in for something surprising, unlinking IF is a very easy change however they may give rise to other issues.

 

[IdiotInCharge]

I am sure AMD tech is fully aware of what Ryzen's limitations were and how to improve them, maybe we will be in for something surprising, unlinking IF is a very easy change however they may give rise to other issues.

There has to be a reason for them setting up Ryzen with a linked internal bus, and they had to know what the tradeoff would be. This leads me to moderate your point, though I do agree in principle.

 

[juanrga]

Eitherway the result difference is negligable, memory doesn't affect performance all that much, and is only noticeable in memory synthetics.

It is trite that a buyer of a Ryzen setup will almost certainly opt for best performance possible and in this case given that Ryzen does benefit more from added bandwidth I don't think anyone other than a casual builder will put anything less than 3000mhz in, for intel it is inconsequential, I have never bought anything faster than the IMC rated.

I have also found that with Ryzen you can actually get better performance from timings, I can make a 2600mhz kit beat a 3000mhz kit. It is something very few overlook, it is not about mhz with memory.

And once again, you ignore my point and discuss about something irrelevant to my point. You are doing on purpose, so I copy and paste my former reply to you.

And here you do again. You ignore my point and then answer about something else...

I am not discussing throughput neither memory bandwidth. I am discussing latency and interconnect speeds: Infinity fabric, mesh, and ring.

Feel free to copy and paste this response if you do it again.

Yes a 6 core CFL beats a 8 core <snip>

As demonstrated before CB is not representative of real-world performance. CineBench is cherry picked benchmark.

Blender? Here you have a collection of results with the six-core 8700k beating the expensive eight-core 1800X on Blender

With all your fanboying, I would be more concerned that AMD don't stumble upon a high clocking node, if they do, it will be very bad. Going from Bulldozer to Zen the gap has closed a lot.

In the Bulldozer epoch, a 8 thread FX CPU could beat (or at least match) an 8 thread i7 SB in multithread workloads.

In the Zen epoch a 16 thread R7 CPU is being humiliated by a 12 threads CFL in multithread workloads. Ouch!

Take last benchmark. The i7 8700k is 12.2% faster than the R7 1800X, but the RyZen has 33% moar cores. So each CFL core is 50% faster than Zen core.

50% faster!

 

[juanrga]

Ask Juanrga it was once his favourite Anandtech post, alternatively you can just do math very simple to calculate.

Has he to ask me about something is only in your imagination?

 

[OrangeKhrush]

Has he to ask me about something is only in your imagination?

Don't be bashful, it was the one you used to post all the time, then claimed 12% is margin of error.

 

[OrangeKhrush]

And once again, you ignore my point and discuss about something irrelevant to my point. You are doing on purpose, so I copy and paste my former reply to you.



Feel free to copy and paste this response if you do it again.



As demonstrated before CB is not representative of real-world performance. CineBench is cherry picked benchmark.



Blender? Here you have a collection of results with the six-core 8700k beating the expensive eight-core 1800X on Blender

In the Bulldozer epoch, a 8 thread FX CPU could beat (or at least match) an 8 thread i7 SB in multithread workloads.

In the Zen epoch a 16 thread R7 CPU is being humiliated by a 12 threads CFL in multithread workloads. Ouch!

And you again can't seem to grasp the concept of clocks being vastly different. If you go by official stock clocks at 4.3ghz that is a 23% difference in clocks and if you go with the MCE benches that is a 32% differential. You want to do a comp on a non clock vs clock basis.....okay chump.

 

[juanrga]

And you again can't seem to grasp the concept of clocks being vastly different. If you go by official stock clocks at 4.3ghz that is a 23% difference in clocks and if you go with the MCE benches that is a 32% differential. You want to do a comp on a non clock vs clock basis.....okay chump.

Clocks are only part of the history...

 

[OrangeKhrush]

Clocks are only part of the history...

when wanting to work out IPC of different uarches you compare on a clock vs clock basis. At this point you are making yourself look pretty silly, can't even do equal comps. Yeah its really not hard to see a 20-30% faster CPU perform better since clock speed affects performance a lot more than cores ibid the 8700K at 3ghz is as fast as a 1600

 

[juanrga]

when wanting to work out IPC of different uarches you compare on a clock vs clock basis. At this point you are making yourself look pretty silly, can't even do equal comps. Yeah its really not hard to see a 20-30% faster CPU perform better since clock speed affects performance a lot more than cores ibid the 8700K at 3ghz is as fast as a 1600

If one core is 50% faster than the other, and the clock difference between both cores is only 27%, then it is pretty obvious that the rest of the difference in performance is coming from higher IPC. It is trivial, isn't?

 

[OrangeKhrush]

If one core is 50% faster than the other, and the clock difference between both cores is only 27%, then it is pretty obvious that the rest of the difference in performance is coming from higher IPC. It is trivial, isn't?

But in reality they are not ie: 3ghz a 8700K is ~ 1600 clock vs clock and equal threads, this probably helped by the 10%~ IPC advantage intel has but in MT it is very far from your 50% crap

 

[juanrga]

But in reality they are not ie: 3ghz a 8700K is ~ 1600 clock vs clock and equal threads, this probably helped by the 10%~ IPC advantage intel has but in MT it is very far from your 50% crap

You continue very confused. 50% is the performance gap, NOT the IPC gap. The IPC gap can be obtained as follows

1.50 / 1.27 = 1.18

i.e. 18% IPC gap.

 

[OrangeKhrush]

You continue very confused. 50% is the performance gap, NOT the IPC gap. The IPC gap can be obtained as follows

1.50 / 1.27 = 1.18

i.e. 18% IPC gap.

Seem to be incorrect but we will keep that your secret information, secret enough to keep to yourself and never share with anyone. I think at this point you just make up numbers.

 

[Meeho]

So, all MCE enabled benches should have an [OC] label?

 

[juanrga]

So, all MCE enabled benches should have an [OC] label?

MCE is like XFR. So it depends of the criteria. I consider that neither XFR nor MCE require a "[OC]" label because both are stock built-in enhancements. People with different criteria must want both XFR and MCE to be labeled as [OC].

What we cannot do is what ADF want. The ADF want to disable MCE on Intel chips, but benchmark RyZen only with the highest interconnect OC possible.

 

[Meeho]

How is MCE stock? It's less stock than running the memory at its rated speed.

 

[kirbyrj]

MCE is like XFR. So it depends of the criteria. I consider that neither XFR nor MCE require a "[OC]" label because both are stock built-in enhancements. People with different criteria must want both XFR and MCE to be labeled as [OC].

What we cannot do is what ADF want. The ADF want to disable MCE on Intel chips, but benchmark RyZen only with the highest interconnect OC possible.

The difference is that XFR is a function of the CPU while MCE is a function of the motherboard. They are not the same. If you would have said XFR and Turbo Boost are the same, I would be more inclined to agree.

 

[Nightfire]

So now we shiften from IPC to per core performance.

Guess the IPC battle was running out of steam.

Garbage benchmark when you have a 1700 and 1600x running the same time.

Btw, you can't just stack percentage on top of each other. This is basic math my friend.

30% less cores plus 20% faster plus 50% less latency = 100% better performance.

100%!!!!

 

[Nightfire]

The Hot Hardware review had the 8700k get a 1521 in CB.

What does that mean? It was running over 4.6 ghz all core.

(1521/6)/55 = 4.61

 

[juanrga]

How is MCE stock? It's less stock than running the memory at its rated speed.

MCE and XFR are stock built-in enhancements.

The difference is that XFR is a function of the CPU while MCE is a function of the motherboard. They are not the same. If you would have said XFR and Turbo Boost are the same, I would be more inclined to agree.

XFR and MCE aren't exactly the same technology, but both are closely related. From that review that OrangeKrush likes to mention again and again:

Intel and AMD both do something similar to the above: With the 8700K, Intel uses Turbo Boost with different frequencies as dependent on thread engagement, where single-core utilization boosts the highest (4.7GHz) and six-core utilization boosts the lowest (4.3GHz, over 3.7GHz base). AMD employs XFR on Ryzen onward, or Extended Frequency Range, and leverages its boosting also on a per-thread level. With lower thread engagement applications, depending on if it’s Ryzen or Threadripper, AMD can add an additional +100MHz to +200MHz to the boosted speed. When more threads are engaged, the boost is lower (in compliance with stability and lower voltage).

What we’re demonstrating today is the impact of multi-core enhancement – a feature present on both AMD and Intel boards – and how the feature can cause confusion in the user base.

 

[juanrga]

Btw, you can't just stack percentage on top of each other. This is basic math my friend.

30% less cores plus 20% faster plus 50% less latency = 100% better performance.

100%!!!!

LOL, no one is stacking the percentages.

6C CFL is 12.2% faster than 8C Zen so

1.122 * 8/6 = 1.496

aka CFL core is 50% faster than one Zen core. This is basic math.

 

[Hagrid]

LOL, no one is stacking the percentages.

6C CFL is 12.2% faster than 8C Zen so

1.122 * 8/6 = 1.496

aka CFL core is 50% faster than one Zen core. This is basic math.

You mean intel math.

 

[Nightfire]

MCE and XFR are stock built-in enhancements.



XFR and MCE aren't exactly the same technology, but both are closely related. From that review that OrangeKrush likes to mention again and again:

Not even close! Who wrote that garbage?? XFR is similiar to turbo boost as Kirby mentioned.

MCE is a factory overclock. NOTHING ELSE! The 8700k is getting clocked up to 4.7 ghz all core, not 100-200 mhz mentioned. Good luck to the poor bastard that doesnt have adequate cooling.

 

[juanrga]

XFR is similiar to turbo boost as Kirby mentioned.

MCE is a factory overclock. NOTHING ELSE!

And XFR is a factory overclock. NOTHING ELSE!:

In AMD CPUs XFR, or eXtended Frequency Range, is an automated overclocking feature in AMD's Zen CPU architecture first introduced on AMD's Ryzen CPUs.

https://www.custompcreview.com/wiki/xfr/

Because Ryzen in effect self-overclocks thanks to its eXtended Frequency Range (XFR) technology, I found overclocking my chip really didn’t do a huge amount to the Ryzen 7 1800X’s scores.

http://www.trustedreviews.com/revie...nce-benchmarks-conclusion#6Y0yu05bW7sMXeHr.99

 

[ManofGod]

AMD is definitely doing 100% better than they were during the FX era. It is also helping to kick Intel into high gear, considering that the fully paper launched their 8700 and 8600 series of processors. Without Ryzen, the 7700K is what would still be the king of Intel's world in the mainstream arena.

 

[noko]

So in a nutshell, when a Ryzen 8 core and Intel 8700K CPUs are used at full capacity - Intel in the end does not look better at all. Considering current pricing I can see why folks would just go with AMD plus future options look way better as well.

 

[juanrga]

AMD is definitely doing 100% better than they were during the FX era. It is also helping to kick Intel into high gear, considering that the fully paper launched their 8700 and 8600 series of processors. Without Ryzen, the 7700K is what would still be the king of Intel's world in the mainstream arena.

Without the 7700k, AMD would be still selling Piledriver FX. So all RyZen users would be thankful to Intel.

So in a nutshell, when a Ryzen 8 core and Intel 8700K CPUs are used at full capacity - Intel in the end does not look better at all. Considering current pricing I can see why folks would just go with AMD plus future options look way better as well.

Benchmarks and sales show exactly the contrary.

 

[noko]

Techspot has a data point for Blender Gooseberry Benchmark for the 8700K at 2949 seconds:
https://www.techspot.com/review/1497-intel-core-i7-8700k/page2.html

I ran the blender Gooseberry Benchmark in the background while the 1080 Ti was Mining and I was using the internet and got 2808 seconds on my 4ghz configuration, about 10% faster over their 8700K. First run, second run the setup stage would probably be less time required but have not done that yet. For full CPU type workloads I do not see any kind of significant advantage in the 8700K. I do see some advantage for smaller type or more restrictive threaded type applications for the 8700K but for me it is not significant enough to make a difference.

 

[ManofGod]

Without the 7700k, AMD would be still selling Piledriver FX. So all RyZen users would be thankful to Intel.

Nope, just nope.

 

[juanrga]

Techspot has a data point for Blender Gooseberry Benchmark for the 8700K at 2949 seconds:
https://www.techspot.com/review/1497-intel-core-i7-8700k/page2.html

I ran the blender Gooseberry Benchmark in the background while the 1080 Ti was Mining and I was using the internet and got 2808 seconds on my 4ghz configuration, about 10% faster over their 8700K. First run, second run the setup stage would probably be less time required but have not done that yet. For full CPU type workloads I do not see any kind of significant advantage in the 8700K. I do see some advantage for smaller type or more restrictive threaded type applications for the 8700K but for me it is not significant enough to make a difference.

So you want to compare overclocked RyZen chip with a stock CoffeeLake... Techspot at least is comparing stock vs stock.

 

[Algrim]

Techspot has a data point for Blender Gooseberry Benchmark for the 8700K at 2949 seconds:
https://www.techspot.com/review/1497-intel-core-i7-8700k/page2.html

I have a hard time believing that score. That's worse than my 12 core 2010 Mac Pro.

EDIT: Here's the blend from my Mac just a few moments ago (46:52.52 = 2812.52 seconds):

 

[Meeho]

MCE and XFR are stock built-in enhancements.



XFR and MCE aren't exactly the same technology, but both are closely related. From that review that OrangeKrush likes to mention again and again:

What are you talking about?! XFR is like turbo boost, a stock CPU feature. MCE is a 3rd party motherboard induced overclocked overriding CPU stock functionality.

THAT you call stock but Ryzen with faster RAM should have an OC label?

 

[juanrga]

What are you talking about?! XFR is like turbo boost, a stock CPU feature. MCE is a 3rd party motherboard induced overclocked overriding CPU stock functionality.

THAT you call stock but Ryzen with faster RAM should have an OC label?

XFR is an automatic overclock feature similar to MCE. It is explained in three reviews given above.

RyZen with overclocked RAM would have an "OC" label because overclocking RAM also overclocks the interconnect of the chip. The specs of the chip have been manually changed and the performance varies.

 

[Meeho]

XFR is a stock CPU feature. MCE is not a stock CPU feature. It's not that hard.

 

[juanrga]

XFR is a stock CPU feature. MCE is not a stock CPU feature. It's not that hard.

Both XFR and MCE are automatic overclocks. It is not that hard.