Discussion AMD Ryzen MegaThread! FAQ and Resources

[gamerk316]

Well, as you raise overall clock speed I'd expect performance differences to start to shrink, because other components in the system (GPU, RAM, HDD) start to become more and more of a bottleneck.

 

[-Fran-]

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

 

[gamerk316]

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

 

[-Fran-]

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

And that is still something you can attribute to platform, but makes it a valid comparison for the reasons I already stated, right?

I mean, I don't disagree that could happen and is not desirable as a measurement, but if the platform as a whole can't make the CPU give out better numbers, that is still something an end user will be tied to anyway and not a "lab" isolated thing. Compressing the results at the top might actually be an interesting thing to see as well for both Ryzen and Intel. You could see how each platform handles an artificial bottleneck?

I'm murking the waters a bit here, but it's interesting to note how this could be a thing going forward: not only giving speed parity, but also max speed parity with everything equals?

Cheers!

 

[TMTOWTSAC]

Not even close. You are seeing XFR on action and the clocks aren't those of above. When clocks are locked RyZen is a good 10% behind Kabylake

Spoiler

Isn't XFR 100Mhz? That would mean the 1500X is 1 point behind CB at the same clocks. Hardly a 10% diff there. No matter what you say now, the impression I'm getting is Ryzen is closing the gap with these barrage of updates they're getting. That is not even wishful thinking anymore, there is plenty of proof in the numbers already to make that a fact.

Also, I wonder how the uArch scales with speed in Ryzen. I know the performance should grow linearly, but it's interesting to see that at higher speeds, Ryzen seems to be closing the gap. Am I wrong in seeing more than the evident information?

I wish reviewers would do 3Ghz, 3.5Ghz and 4Ghz in-line tests for it. Well, 4Ghz if Ryzen allows, haha.

Cheers! 😛

XFR varies from model to model. The 1500X lists a 200 MHz XFR, while the 1400 only has a 50 MHz XFR.

 

[salgado18]

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

That is a simple test for anyone with a 7700k or 7600k to test: lower the RAM to 2166 or less, and benchmark at 4.0, 4.5 and 5.0 GHz, and note the way the numbers will scale. I still believe that many CPU tests will not be affected much by RAM speed, since they are synthetic in nature and meant to isolate the CPU performance as much as possible.

 

[cdrkf]

This image caught my eye on a second read and is very interesting:

These are the 1:1 for the 1500X and the 7500 respectively:

3.5 / 3.7 GHz - Base/Turbo - 3.4 / 3.8 GHz
16 - PCIe 3.0 Lanes - 16
16 MB - L3 Cache - 6 MB
65 W - TDP - 65 W

At a first glance, that 1 point of difference in CB could mean the 100Mhz of turbo difference. Which in turn means... Ryzen is on par with Kaby Lake?

Not even close. You are seeing XFR on action and the clocks aren't those of above. When clocks are locked RyZen is a good 10% behind Kabylake

I think it's worth keeping in mind, aside from the last couple of months before Ryzen launch (when the hype got wildly out of control), AMD getting to within 10% IPC of Kabylake was pretty much unfathomable. I really think it's worth keeping in mind just what AMD have managed to do here, it's impressive, and moreover they've priced it very keenly.

Juan, I do get the impression from your posts that you are rather disappointed in Ryzen / the Zen core? That surprises me as from what I can tell it's surpassed most early expectations. I agree the hype in the last couple of months before launch got out of hand, still do you not agree it's a competitive core and given the pricing quite a compelling product? That's not to say it doesn't have it's quirks, just that overall I think given where AMD *were* prior to Ryzen they've done very well?

I mean the other thing I find positive is the performance is strong despite there being obvious areas where they could improve the core- which bodes very well for Zen 2 and Zen 3...

 

[salgado18]

Spoiler

This image caught my eye on a second read and is very interesting:

These are the 1:1 for the 1500X and the 7500 respectively:

3.5 / 3.7 GHz - Base/Turbo - 3.4 / 3.8 GHz
16 - PCIe 3.0 Lanes - 16
16 MB - L3 Cache - 6 MB
65 W - TDP - 65 W

At a first glance, that 1 point of difference in CB could mean the 100Mhz of turbo difference. Which in turn means... Ryzen is on par with Kaby Lake?

Not even close. You are seeing XFR on action and the clocks aren't those of above. When clocks are locked RyZen is a good 10% behind Kabylake

I think it's worth keeping in mind, aside from the last couple of months before Ryzen launch (when the hype got wildly out of control), AMD getting to within 10% IPC of Kabylake was pretty much unfathomable. I really think it's worth keeping in mind just what AMD have managed to do here, it's impressive, and moreover they've priced it very keenly.

Juan, I do get the impression from your posts that you are rather disappointed in Ryzen / the Zen core? That surprises me as from what I can tell it's surpassed most early expectations. I agree the hype in the last couple of months before launch got out of hand, still do you not agree it's a competitive core and given the pricing quite a compelling product? That's not to say it doesn't have it's quirks, just that overall I think given where AMD *were* prior to Ryzen they've done very well?

I mean the other thing I find positive is the performance is strong despite there being obvious areas where they could improve the core- which bodes very well for Zen 2 and Zen 3...

I think that Zen is the best CPU that AMD could have created, given their budget constraints:

- scalable, can get to huge core counts if needed, which adresses well the server market;
- power efficient, which is very good for low-power scenarios, like note/ultrabooks, portables, IoT, and why not servers;
- multi-thread friendly, is very strong with multiple independent thread workloads, which is good for servers and professionals;
- high IPC, got very close to current Intels, the gap may increase in the future with newer Intel chips, but is a lot better than any other arch that AMD had, and is very competitive.

They only have two big shortcomings:

- the modular design incurs a performance penalty for interdependent threads, which is not good for games for example;
- the process used cannot reach high clocks, which limits high-performance parts.

But, overall, it is a very successful CPU, in my opinion. Sure, they can optimize from now on, address the bads and increase the goods, but it already is a great chip and platform.

 

[gamerk316]

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

And that is still something you can attribute to platform, but makes it a valid comparison for the reasons I already stated, right?

I mean, I don't disagree that could happen and is not desirable as a measurement, but if the platform as a whole can't make the CPU give out better numbers, that is still something an end user will be tied to anyway and not a "lab" isolated thing. Compressing the results at the top might actually be an interesting thing to see as well for both Ryzen and Intel. You could see how each platform handles an artificial bottleneck?

I'm murking the waters a bit here, but it's interesting to note how this could be a thing going forward: not only giving speed parity, but also max speed parity with everything equals?

Cheers!

Personally, I suspect that due to the end of die shrinks fast approaching, AMD and Intel are going to hover around the same performance level for a while. Remember, we've only got two, maybe three shrinks left.

I think that Zen is the best CPU that AMD could have created, given their budget constraints:

- scalable, can get to huge core counts if needed, which adresses well the server market;
- power efficient, which is very good for low-power scenarios, like note/ultrabooks, portables, IoT, and why not servers;
- multi-thread friendly, is very strong with multiple independent thread workloads, which is good for servers and professionals;
- high IPC, got very close to current Intels, the gap may increase in the future with newer Intel chips, but is a lot better than any other arch that AMD had, and is very competitive.

They only have two big shortcomings:

- the modular design incurs a performance penalty for interdependent threads, which is not good for games for example;
- the process used cannot reach high clocks, which limits high-performance parts.

But, overall, it is a very successful CPU, in my opinion. Sure, they can optimize from now on, address the bads and increase the goods, but it already is a great chip and platform.

Zen does what it was required to do: Make AMD an option. People like us are still going to recommend the 7700k for those of us not budget constrained because, frankly, it's the better gaming CPU. But for budget gamers or people with unique (non gaming) use cases? You can recommend Zen with a straight face. You couldn't do that with FX.

The real question though: Will OEMs buy in? Because those are the ones who move product, not us.

 

[salgado18]

Zen does what it was required to do: Make AMD an option. People like us are still going to recommend the 7700k for those of us not budget constrained because, frankly, it's the better gaming CPU. But for budget gamers or people with unique (non gaming) use cases? You can recommend Zen with a straight face. You couldn't do that with FX.

The real question though: Will OEMs buy in? Because those are the ones who move product, not us.

Well, I won't recommend the 7700k. Ryzen is just as capable in gaming, with a lot more future-proofing and all-around performance. Spending more on a faster but limited and expensive CPU is a waste of money, even for unconstrained budgets (and then the 2011 platform is better anyway).

But OEMs might pick Ryzen up, if they follow the trend of mobile: there were single and dual cores, then quad cores, then dual-quad cores, but these were called octo-cores, and that was stronger for marketing. Saying the computer has 8 cores and is still cheaper sells. But we'll have to wait and see.

 

[8350rocks]

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

Zen also scales better with overclock because increasing the CPU clock increases cache performance as well, which reduces cache latency. Not to mention IF scaling with RAM speed also increases performance with RAM OC or AMP/XMP profiles.

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

And that is still something you can attribute to platform, but makes it a valid comparison for the reasons I already stated, right?

I mean, I don't disagree that could happen and is not desirable as a measurement, but if the platform as a whole can't make the CPU give out better numbers, that is still something an end user will be tied to anyway and not a "lab" isolated thing. Compressing the results at the top might actually be an interesting thing to see as well for both Ryzen and Intel. You could see how each platform handles an artificial bottleneck?

I'm murking the waters a bit here, but it's interesting to note how this could be a thing going forward: not only giving speed parity, but also max speed parity with everything equals?

Cheers!

The way the architecture scales with cache, memory, clockspeed, and fabric speed...Ryzen really does ramp up with higher speeds. I find it to be quite interesting because you are removing some of the bottlenecks as you increase clockspeed. Your cache performance increases, etc.

In pure CPU metrics, would that be the case as well? I would also think it's a fair thing to measure, even with those "drawbacks", since those are still platform deficiencies you're inherently tied to, right?

Cheers!

There's very few pure CPU metrics, though those should (theoretically) scale linearly with CPU performance.

But pretty much everything else will start to be measurably affected by RAM bandwidth once you start moving beyond the 4GHz range, compressing the result set to some degree.

The point is, it isn't so much that a CPU "scales better" at a higher clock, it's more that the result sets are being compressed by some other factor.

And that is still something you can attribute to platform, but makes it a valid comparison for the reasons I already stated, right?

I mean, I don't disagree that could happen and is not desirable as a measurement, but if the platform as a whole can't make the CPU give out better numbers, that is still something an end user will be tied to anyway and not a "lab" isolated thing. Compressing the results at the top might actually be an interesting thing to see as well for both Ryzen and Intel. You could see how each platform handles an artificial bottleneck?

I'm murking the waters a bit here, but it's interesting to note how this could be a thing going forward: not only giving speed parity, but also max speed parity with everything equals?

Cheers!

Personally, I suspect that due to the end of die shrinks fast approaching, AMD and Intel are going to hover around the same performance level for a while. Remember, we've only got two, maybe three shrinks left.

I think that Zen is the best CPU that AMD could have created, given their budget constraints:

- scalable, can get to huge core counts if needed, which adresses well the server market;
- power efficient, which is very good for low-power scenarios, like note/ultrabooks, portables, IoT, and why not servers;
- multi-thread friendly, is very strong with multiple independent thread workloads, which is good for servers and professionals;
- high IPC, got very close to current Intels, the gap may increase in the future with newer Intel chips, but is a lot better than any other arch that AMD had, and is very competitive.

They only have two big shortcomings:

- the modular design incurs a performance penalty for interdependent threads, which is not good for games for example;
- the process used cannot reach high clocks, which limits high-performance parts.

But, overall, it is a very successful CPU, in my opinion. Sure, they can optimize from now on, address the bads and increase the goods, but it already is a great chip and platform.

Zen does what it was required to do: Make AMD an option. People like us are still going to recommend the 7700k for those of us not budget constrained because, frankly, it's the better gaming CPU. But for budget gamers or people with unique (non gaming) use cases? You can recommend Zen with a straight face. You couldn't do that with FX.

The real question though: Will OEMs buy in? Because those are the ones who move product, not us.

Why would you buy a 7700K for ~$350, when you can buy a 12 thread R5-1600X that performs just as well in nearly everything? There are a few games where you can push higher average FPS with roughly similar minimums on a 7700K...but you can buy the 12 thread processor for ~$100 less. If you really want to spend the money, you can double your threads on a 1700 that most people can get to 4 GHz pretty easily at this point.

Not to mention that the video from Wendell the other day shows the subjective opinion of all 3 people to be that performance was better on Ryzen in a blind study across 4 systems looking for dropped frames, stuttering, and disruptive glitches. They took the diplomatic route on the conclusions, I think to avoid slandering one side or another, but the results showed the systems they said underperformed were the 7700K and the 5960X specifically.

 

[jdwii]

I enjoyed looking at the 1500X and comparing it to a 4770-4790K. In most tests its pretty close. Though Ryzen is using faster memory to get those scores its still impressive performance.

1500X+1070 would perform very well in gaming.

Still can't wait for R3 to launch


Personally i'd much rather have a R7 1700 over a 7700K but i would pick the fastest memory i could get with it. I'd basically pick all Amd parts over the current Intel lineup.

I respect wendell quite a lot but anecdotal evidence is not going to convince me it never has never well with anything. But I've been seeing a lot of benchmarks show ryzen ahead in min frame rates and we all know that is more important then average frame rates.

 

[jaymc]

This image caught my eye on a second read and is very interesting:

These are the 1:1 for the 1500X and the 7500 respectively:

3.5 / 3.7 GHz - Base/Turbo - 3.4 / 3.8 GHz
16 - PCIe 3.0 Lanes - 16
16 MB - L3 Cache - 6 MB
65 W - TDP - 65 W

At a first glance, that 1 point of difference in CB could mean the 100Mhz of turbo difference. Which in turn means... Ryzen is on par with Kaby Lake?

Not even close. You are seeing XFR on action and the clocks aren't those of above. When clocks are locked RyZen is a good 10% behind Kabylake

139/153 = 90.84 = 9.16% gap

The 1500x has more level 3 cache per core, is this whats giving it a better result in cinebench single core perf than the 1700's ?

So if it was clocked the same speed as the 1800x would cinebench then report a higher IPC in single thread perf than the 1800x does, with all it's threads an cores utilizing L3 Cache also ?

 

[8350rocks]

This image caught my eye on a second read and is very interesting:

These are the 1:1 for the 1500X and the 7500 respectively:

3.5 / 3.7 GHz - Base/Turbo - 3.4 / 3.8 GHz
16 - PCIe 3.0 Lanes - 16
16 MB - L3 Cache - 6 MB
65 W - TDP - 65 W

At a first glance, that 1 point of difference in CB could mean the 100Mhz of turbo difference. Which in turn means... Ryzen is on par with Kaby Lake?

Not even close. You are seeing XFR on action and the clocks aren't those of above. When clocks are locked RyZen is a good 10% behind Kabylake

139/153 = 90.84 = 9.16% gap

The 1500x has more level 3 cache per core, is this whats giving it a better result in cinebench single core perf than the 1700's ?

So if it was clocked the same speed as the 1800x would cinebench then report a higher IPC in single thread perf than the 1800x does, with all it's threads an cores utilizing L3 Cache also ?

Cache certainly helps, but AMD's L3 is a victim cache, meaning fewer cores running will push less into L3 cache. Technically, the way it is designed, the 6 core and 8 core parts should, theoretically, see more gains from all that L3.

 

[8350rocks]

I enjoyed looking at the 1500X and comparing it to a 4770-4790K. In most tests its pretty close. Though Ryzen is using faster memory to get those scores its still impressive performance.

1500X+1070 would perform very well in gaming.

Still can't wait for R3 to launch


Personally i'd much rather have a R7 1700 over a 7700K but i would pick the fastest memory i could get with it. I'd basically pick all Amd parts over the current Intel lineup.

I respect wendell quite a lot but anecdotal evidence is not going to convince me it never has never well with anything. But I've been seeing a lot of benchmarks show ryzen ahead in min frame rates and we all know that is more important then average frame rates.

What about the 250ms hang in frame times from the 7700K that his graph showed? The 5960X also experienced a hang that was 180ms.

Considering that Ryzen never broke 30ms gap, and at no point was considered stuttering, I would think the hard numbers from the frame time graph would corroborate the user experience. Especially true if we are considering it was duplicated by multiple individuals playing the same games across the same systems....

There is your definitive proof that Ryzen was smoother than Intel. Frame times do not lie, and a 180-250ms hang in frame times would be a horrendously bad stutter with a noticeable effect on performance.

 

[gamerk316]

What about the 250ms hang in frame times from the 7700K that his graph showed? The 5960X also experienced a hang that was 180ms.

Considering that Ryzen never broke 30ms gap, and at no point was considered stuttering, I would think the hard numbers from the frame time graph would corroborate the user experience. Especially true if we are considering it was duplicated by multiple individuals playing the same games across the same systems....

There is your definitive proof that Ryzen was smoother than Intel. Frame times do not lie, and a 180-250ms hang in frame times would be a horrendously bad stutter with a noticeable effect on performance.

I'm going to point out that normal tech review sites have found absolutely no evidence of those types of hangs, especially when you consider that a 250ms hang time represents 16 frames being consecutively dropped. Which would be quite noticeable, even without latency benchmarking.

So I'll come out and say it: the test is being rigged is some way to make Intel appear worse. Because if the 7700k (or other, slower, Intel CPUs such as my 2600k) had those types of issues in normal usage, someone would have noticed by now. Or are you going to accuse every single tech site in existence of being Intel shills for not reporting this issue?

 

[8350rocks]

Analysis of Wendell's video from yesterday: https://www.youtube.com/watch?v=11NfsMykyAk

 

[randomizer]

Analysis of Wendell's video from yesterday: https://www.youtube.com/watch?v=11NfsMykyAk

In short: It's as close to BS as it can be without being completely BS.

 

[Hellfire13]

Have to admit, the R5 linup looks more value for money than the R7 lineup. Specially the 1600x with its clockspeed. Interesting read... http://www.pcworld.com/article/3186811/computers/ryzen-5-review-vs-core-i5-ryzen-5-1600x-wins-for-best-mainstream-power-cpu.html?page=4

 

[randomizer]

The 1600 gives you a HSF and costs less at the same time. The 1600X seems to fit in the same awkward position as the 1700X does for the Ryzen 7 line. It's really only appealing to people who want to buy a HSF and not overclock.

 

[truegenius]

Zen also scales better with overclock because increasing the CPU clock increases cache performance as well, which reduces cache latency. Not to mention IF scaling with RAM speed also increases performance with RAM OC or AMP/XMP profiles.

but the problem is memory here, overclocking the cpu will make memory problem more obvious and give minimum benifit in games or tasks that depend on memory more. And l1 cache operate at cpu speed, l2 cache also operate at cpu speed and close to cpu speed in some, so overclocking both intel and amd will overclock those by same amount thus maintaining same ratio of those component's clock.
i would believe it when i see it ( preferably in gta5 ) , keep ram at 2133mhz and then overclock cpu only to see how much more fps we get, in my 1090t i get 30% performance increase from 1800mhz cl9 ram 2925mhz imc vs 1200mhz cl9 ram 2025mhz imc, on the other hand a cpu overclock from 3.6 to 4.05ghz gives only ~3% more fps ( at 1800mhz cl9 ram 2925mhz imc ) ( default setting 720p on 7950@1.3ghz )

Why would you buy a 7700K for ~$350, when you can buy a 12 thread R5-1600X that performs just as well in nearly everything?

if the want a gaming rig then its 7700k if they want more of transcoding/redering rig then it is 1600x, in games all zen performs more close to 2600k than to 7700k and 2600k overclock much higher % than ryzen and support great memory overclock too thus more to squeze from it.

What about the 250ms hang in frame times from the 7700K that his graph showed? The 5960X also experienced a hang that was 180ms.

pure bs, to get more click/views, post something normal then nobody cares so post something "omg you won't believe this" then then get more views ( just like a cancer to review media ) and fanboys will spread the link thus millions of views for bs.

 

[Jsimenhoff]

Hey all, last week AMD joined us for a very special Ryzen AMA. Dive into our lengthy digest to learn about the past, present, and future of Ryzen.

Want to read the AMA in its original format? You can check out the original thread here. While you're there make sure to enter our second and final giveaway for the ASRock X370 Taichi AM4 Motherboard.

 

[juanrga]

Not even close. You are seeing XFR on action and the clocks aren't those of above. When clocks are locked RyZen is a good 10% behind Kabylake

Spoiler

Isn't XFR 100Mhz? That would mean the 1500X is 1 point behind CB at the same clocks. Hardly a 10% diff there. No matter what you say now, the impression I'm getting is Ryzen is closing the gap with these barrage of updates they're getting. That is not even wishful thinking anymore, there is plenty of proof in the numbers already to make that a fact.

XFR is 200MHz for the 1500X and that is for standard cooling. It could get higher clocks on an open case with low tempts, because the XFR turbo depends on temperature.

RyZen is not closing the gap significantly. Here you have the impact of the new AGESA update that addressed the memory latency issue, how you can check yourself the overall performance has not varied a bit

Here you have a recent graph with last data for RyZen including the new models. First games and below compute

If you compare the R5 1400 with the i7 2600k. You can easily deduce

RyZen IPC+SMT ~ Haswell
RyZen IPC ~ Sandy

which is the same conclusion obtained at launch day with the reviews of the 1800X.

Juan, I do get the impression from your posts that you are rather disappointed in Ryzen / the Zen core?

At contrary, I think it is a good microarchitecture and it is very close to what I predicted. I am just tired of certain people that continues to hype it despite reviews already stated the facts. Remember OrangeKrush? He is now in another forum pretending to convince people that Zen has an IPC much higher than Kabylake. On CB15 Kabylake IPC is about 22% ahead of Sandy, and OrangeKrush claims Ryzen is 33% ahead of Sandy on the same bench. Comedy Gold.

 

[juanrga]

http://www.tomshardware.com/reviews/amd-ryzen-5-1600x-cpu-review,5014-2.html
Search: "Ryzen 5 1600X Memory Data Rate "

I have checked. You said "2333 seems like a sweet spot", but I think you mean 2666. The nonlinear evolution is just a consequence of what I wrote above. Increasing frequency of RAM only increase throughput on the units directly tied to RAM (RAM and CCX IF in the case of RyZen), but the rest of the system runs at same frequency and you have the nonlinear kind of curve I mentioned before because the part is not scaling up is generating a bottleneck.

There is not anything special here. Any other microarchitecture will follow the same trend. Overclocking RAM on Intel chips also increase performance nonlinearly. There is a point beyond which further increasing RAM clock does nothing because the CPU is bottlenecked.

 

[-Fran-]

XFR is 200MHz for the 1500X and that is for standard cooling. It could get higher clocks on an open case with low tempts, because the XFR turbo depends on temperature.

RyZen is not closing the gap significantly. Here you have the impact of the new AGESA update that addressed the memory latency issue, how you can check yourself the overall performance has not varied a bit

Here you have a recent graph with last data for RyZen including the new models. First games and below compute

If you compare the R5 1400 with the i7 2600k. You can easily deduce

RyZen IPC+SMT ~ Haswell
RyZen IPC ~ Sandy

which is the same conclusion obtained at launch day with the reviews of the 1800X.

Yeah... Well, No.

A single data point doesn't help compare "day1" to "current day". I still see in today's data that Ryzen is within 7% of Kaby Lake clock per clock when day1 was 12%-ish. Whatever the reasons, I don't really care, but like you say "the data is there". The most beaten to death horse, CB, seems to agree with me there.

Which, at the same time, makes me even more curious to explain *why* we see bigger differences elsewhere. Your default answer is "because latency". Well, again, no. That is not a convincing answer. Have more data that could indicate why? Something that you haven't thought of yet?

I have read that different RAM and cooling should be the biggest differentiation variables, but outside of Toms and Anands, I haven't read anything else around those topics.

Cheers!