AMD - Zen chitchat

[CAT-THE-FIFTH]

I think I figured it out:

https://www.anandtech.com/show/12625...n-5-2600x-2600
https://www.anandtech.com/show/11859...tial-numbers/5

AT didn't use high end cooling on the Intel parts,but this rack mount CPU cooler probably to simulate a stock cooler instead of high end cooling:

http://www.silverstonetek.com/produc...rea=en&pid=652

It was also used in their Core i7 8700K review.They used the stock AMD coolers for the Ryzen 2000 series.

[Corky34]

Some details from a forum thread on Anandtech talking about the technical changes with the 2000 series, particular pertaining to the yet to be implemented precision boost overdrive.

The "Precision Boost Override" feature available on 400-series motherboards allows increasing the physical limiters mentioned earlier. On SKUs belonging to the 105W TDP infrastructure group, the default limiters are following: PPT 141.75W, TDC 95A, EDC 140A and tJMax of 85°C (absolute, excl. offset).

When "Precision Boost Override" mode is enabled (AGESA default), PPT becomes essentially unrestricted (1000W), TDC is set to 114A and EDC to 168A. These limits can be customized by the ODM so that the new limits will comply with the electrical characteristics of the motherboard design in question.

[GHz]

How has memory compatibility improved with Ryzen 2xxx?

I currently have a 1700 with 16GB 3200MHz Corsair CL16 (Hynix) which flat out refuses to boost past 2993MHz on a B350 Strix with latest bios. It's also a poor clocker and it segfaults.

Since the IMC is on the CPU, could I just plop in the 2700X in a B350 Strix (can it handle it?) and happily use the Corsair memory at 3200MHz? Or does the motherboard play a significant role?

[Firejack]

AMD answering questions about Ryzen 2 on Scans Youtube channel at 1pm (40 minutes from now)

https://twitter.com/AMD_UK/status/987285553644175361

[watercooled]

Another review showing results towards the higher side: https://www.techradar.com/reviews/amd-ryzen-7-2700x

The thing is, unusually slow results on a platform can be caused by a number of things. Higher results on the other hand, provided they're correct and not due to e.g. a timing issue, well it's hard to misinterpret them*, provided of course they've done a couple of runs and not ended up with any bizarre results due to benchmark errors and whatnot. For these tests however, the performance seems consistent across a range of applications so I'm struggling to see where an error could lie. If it's platform-dependent then perhaps some sort of multicore enhancement or whatever the board MFR calls it is ending up enabled, but that's not reflected in AAT's power draw figures which are within TDP (TR seem to be measuring wall power?)

As for the 8700k's power draw I suspect AAT are using some sort of AVX load which is causing high power draw, and TechRadar using something like an integer load - particularly on modern CPUs, what the 'load' is can have a significant impact on core power draw.

*Take download speed as an example. You've just bought a new phone and are trying out the LTE connection - you download a 100MB file in 10 seconds. Clearly, at that time the connection was transferring around 80Mb/s (on average) because you have the finished file stored on your phone at the end of it - the connection is clearly capable of doing so or that would have been impossible. Perhaps it varied during the download but that means it must have been capable of going above 80Mb/s as much as it went under as you measured an average. You just know the connection is capable of at least 80Mb/s. Other people getting 40Mb/s doesn't invalidate your result... because it happened.

Similarly if I transcode a video file on a computer and it's finished in 10 minutes (measured accurately), and the output is identical to the control, then that happened. Bugs generally don't make things faster!

I'm not dismissing the possibility of measurement discrepancies, but people criticising AAT for 'not setting it up properly' - well it would be nice if they explained how? Again, lower results could be down to HDD activity, buggy drivers, antivirus running in the background. These things all make performance go down, not up.

[CAT-THE-FIFTH]

Another review showing results towards the higher side: https://www.techradar.com/reviews/amd-ryzen-7-2700x

The thing is, unusually slow results on a platform can be caused by a number of things. Higher results on the other hand, provided they're correct and not due to e.g. a timing issue, well it's hard to misinterpret them*, provided of course they've done a couple of runs and not ended up with any bizarre results due to benchmark errors and whatnot. For these tests however, the performance seems consistent across a range of applications so I'm struggling to see where an error could lie. If it's platform-dependent then perhaps some sort of multicore enhancement or whatever the board MFR calls it is ending up enabled, but that's not reflected in AAT's power draw figures which are within TDP (TR seem to be measuring wall power?)

As for the 8700k's power draw I suspect AAT are using some sort of AVX load which is causing high power draw, and TechRadar using something like an integer load - particularly on modern CPUs, what the 'load' is can have a significant impact on core power draw.

*Take download speed as an example. You've just bought a new phone and are trying out the LTE connection - you download a 100MB file in 10 seconds. Clearly, at that time the connection was transferring around 80Mb/s (on average) because you have the finished file stored on your phone at the end of it - the connection is clearly capable of doing so or that would have been impossible. Perhaps it varied during the download but that means it must have been capable of going above 80Mb/s as much as it went under as you measured an average. You just know the connection is capable of at least 80Mb/s. Other people getting 40Mb/s doesn't invalidate your result... because it happened.

Similarly if I transcode a video file on a computer and it's finished in 10 minutes (measured accurately), and the output is identical to the control, then that happened. Bugs generally don't make things faster!

I'm not dismissing the possibility of measurement discrepancies, but people criticising AAT for 'not setting it up properly' - well it would be nice if they explained how? Again, lower results could be down to HDD activity, buggy drivers, antivirus running in the background. These things all make performance go down, not up.

I pointed it out in an earlier post,AT are not using some huge heatsink when comparing the CPUs. To equate it closer to Ryzen with a stock cooler they simply used a smallish heatsink for the Intel ones, instead of a £50 to £100 one.

For instance computerbase.de compared the Ryzen CPUs with stock coolers to Intel ones with a £50+ Noctua one.

[Xlucine]

Another review showing results towards the higher side: https://www.techradar.com/reviews/amd-ryzen-7-2700x

The thing is, unusually slow results on a platform can be caused by a number of things. Higher results on the other hand, provided they're correct and not due to e.g. a timing issue, well it's hard to misinterpret them*, provided of course they've done a couple of runs and not ended up with any bizarre results due to benchmark errors and whatnot. For these tests however, the performance seems consistent across a range of applications so I'm struggling to see where an error could lie. If it's platform-dependent then perhaps some sort of multicore enhancement or whatever the board MFR calls it is ending up enabled, but that's not reflected in AAT's power draw figures which are within TDP (TR seem to be measuring wall power?)

As for the 8700k's power draw I suspect AAT are using some sort of AVX load which is causing high power draw, and TechRadar using something like an integer load - particularly on modern CPUs, what the 'load' is can have a significant impact on core power draw.

*Take download speed as an example. You've just bought a new phone and are trying out the LTE connection - you download a 100MB file in 10 seconds. Clearly, at that time the connection was transferring around 80Mb/s (on average) because you have the finished file stored on your phone at the end of it - the connection is clearly capable of doing so or that would have been impossible. Perhaps it varied during the download but that means it must have been capable of going above 80Mb/s as much as it went under as you measured an average. You just know the connection is capable of at least 80Mb/s. Other people getting 40Mb/s doesn't invalidate your result... because it happened.

Similarly if I transcode a video file on a computer and it's finished in 10 minutes (measured accurately), and the output is identical to the control, then that happened. Bugs generally don't make things faster!

I'm not dismissing the possibility of measurement discrepancies, but people criticising AAT for 'not setting it up properly' - well it would be nice if they explained how? Again, lower results could be down to HDD activity, buggy drivers, antivirus running in the background. These things all make performance go down, not up.

If people were comparing the actual results between publications you'd have a point, but what got people excited for the AT review was a relative standings (i.e. 2700x fps Vs 8700k fps). Relative results can be skewed by a worse-than-typical result, which is likely what happened here.

As Cat pointed out, the intel chips in the AT review were under a 3U low profile cooler. It's probably a fine cooler for its size, but compared to the noctua one that the first gen ryzen chips were under the results aren't super surprising

[watercooled]

I was thinking that as I posted (the possibility of relatively low Intel results rather than high AMD results) but the post was rambling on enough already!

Now you both mention it I seem to recall reading about cooling differences before - if that's the case maybe we'll see more sites testing with different heatsinks if CPUs are starting to become more thermally-limited? A benchmark with an AIO cooler may not be that representative of a user's own performance if they intend to use the stock cooler which allows the CPU to throttle - something I can well imagine with the rubbish thermal conductivity of Intel's current models. Having to factor in the cost of an AIO for as-reviewed performance would likely skew the price/performance model a bit!

[CAT-THE-FIFTH]

Thats the thing the Intel CPUs are a best case scenario,and ideally reviewers should be using something cheaper like the stock cooler,Hyper 212,and also the expensive £50 to £100 ones for the Intel CPUs. When you have comparisons of a Ryzen CPU with a stock cooler an Intel CPUs with expensive ones,it gives Intel and advantage. Plenty of people probably are buying K series CPUs with a cheaper cooler,due to the higher boost frequencies but not overclocking them.

Remember,the tests of some of the OEM systems with CFL CPUs,where the CPUs were at rated TDP and with more basic cooling - they performed worse.

[Firejack]

der8auer has split open a Ryzen 5 2600. Main findings;

• IHS Soldered on as expected
• Die size is the same as Ryzen 1st gen as Anandtech pointed out
• Using liquid metal lowered temps by just 4c

[DanceswithUnix]

Thats the thing the Intel CPUs are a best case scenario,and ideally reviewers should be using something cheaper like the stock cooler,Hyper 212,and also the expensive £50 to £100 ones for the Intel CPUs. When you have comparisons of a Ryzen CPU with a stock cooler an Intel CPUs with expensive ones,it gives Intel and advantage. Plenty of people probably are buying K series CPUs with a cheaper cooler,due to the higher boost frequencies but not overclocking them.

Remember,the tests of some of the OEM systems with CFL CPUs,where the CPUs were at rated TDP and with more basic cooling - they performed worse.

Was interested to see kitguru using the same cooler on Intel and AMD. Do Intel ship with a cooler?

[DanceswithUnix]

Double post. Tablet finger trouble!

[scaryjim]

Was interested to see kitguru using the same cooler on Intel and AMD.

Hexus did the same - using a Noctua NH-D15S

... Do Intel ship with a cooler?

The Scan listing the for the retail boxed i7 8700k says no. I assume some of the non-K processors do, though...

[CAT-THE-FIFTH]

Computerbase.de also used a Noctua cooler and its massive,and compared to AMD CPUs with the stock cooling:

http://www.pcgameware.co.uk/images/N...lled-above.jpg

It costs around £70.

[DanceswithUnix]

Has anyone measured "heatsink scaling" yet? Thermals seem to be increasingly important as time goes on.

[Biscuit]

der8auer has split open a Ryzen 5 2600. Main findings;

• IHS Soldered on as expected
• Die size is the same as Ryzen 1st gen as Anandtech pointed out
• Using liquid metal lowered temps by just 4c

Interesting there is such a small difference with liquid metal, the existing compound must be pretty good stuff. Wonder if we will find that intel stop cutting those corners in future CPUs.

Edit... soldered... durgh