AMD: Next Gen Threadripper and Ryzen 9 3950X, Coming November

82 Comments

View All Comments

• Targon - Tuesday, September 24, 2019 - link

  You clearly don't understand the difference between what a chip can do, and what the TDP is. Going above base speeds will take more power, but how a chip is rated when it comes to TDP is very different between Intel and AMD.
  
  Intel ONLY reports the TDP for base speed. So if the i9-9900k says 95W TDP, that is at the base speed of 3.6GHz. The fact that the i9-9900ks has a TDP of 127W with a base speed of 4.0GHz indicates very clearly that just going from 3.6GHz to 4.0GHz boosts the TDP from 95W to 127W. So that magic 5.0GHz that people want isn't going to be at 95W, or 127W, it's going to be upwards of 200W.
  
  AMD rates TDP for typical usage, and while that is a bit vague, it means that a 105W TDP CPU will be enough to run above base speeds and stay within the official TDP rating.
  
  Again, Turbo/Boost speeds and the TDP rating are two different things. If power delivery to the CPU were limited to official numbers, you wouldn't see any Intel chips getting above 4.6-4.7GHz even with golden samples. Throw unlimited power at the chip, and you get higher clock speeds.

• Sushisamurai - Saturday, September 21, 2019 - link

  I think you should read the article that refers to how AMD's turbo's definition and how theirs works vs Intel, as its quite enlightening and I do agree with the observations made by Anandtech. I think AMD's done a great job and its nice to know that the CPU you're buying from them is pretty much at max performance, and could probably undervolt for better performance instead of overvolt and overclocking vs Intel's overclock for more performance because they're leaving more on the table for the user to exploit.

• Tabalan - Saturday, September 21, 2019 - link

  Hmm, how did we get from "low OC headroom is a drawback" (Intel) or "all chips should be able to OC" (Intel) to "0 OC headroom is good" (AMD)?
  
  Also, OC is not an exploit, it's just a feature of almost all (AMD) or selected (Intel) SKUs. That's why you buy Intel K series - to Oc them.
  
  Next thing - majority of AMD Ryzen 3000 CPUs don't hit Boost Clock ever. Watch De8auer video, hunders of people took part in survey and it turns out in some SKUs only few (<10) % of units hit Boost clock EVER.

• teldar - Saturday, September 21, 2019 - link

  I'm going to ask you to think about something. Who do you think is more likely to take part in a survey? Someone who is pissed off and has something to say? Or someone who is completely happy with what they have and have stopped thinking about it? It's a fact that voluntary surveys are skewed negative because of people are more like to complain if something isn't right than write about how good it is.
  I wouldn't take survey numbers like they're a scientific study.

• Martijn ter Haar - Sunday, September 22, 2019 - link

  There's that bias, but I also think Der Ba8er's survey has a problem, because of Windows Scheduler. AMD only guarantees the boost clock on one core, the one marked with the gold star in Ryzen Master. When I run the CB single core test on my Ryzen 3600 it does not get scheduled to that core, but to a seemingly random core. In fact, in spite of the name, I see loads peak on different threads during the run and I am pretty sure not all are on the same core (I use CB20 instead of CB15.) And there is no way to easily force Windows to use the gold star core (Maybe it can be done with Process Lasso?)
  
  If this is the case, it is no wonder that in Der Ba8er's survey the number of Ryzen 3000 CPU's that can't reach their advertised boost clock gets worse with a rising number of cores. The chance that the benchmark is run on the gold star core just gets lower.
  
  Of course, this is not the entire story, because AMD's ABBA Agesa revision did help. But I still think Windows not assigning the CB single core benchmark run to the right, 'gold star' core plays an important role in Der Ba8er's survey results.

• Korguz - Saturday, September 21, 2019 - link

  tabalan... has de8auer ever talked about how much power intel uses to reach the clocks and performance it gets ? like teldar said.. those surveys are bias towards the negative. and should be taken as biased

• FullmetalTitan - Saturday, September 21, 2019 - link

  So it's clear from your reply that you DIDN'T read the article, because the impossibility of measuring an instantaneous boost without tainting the result is covered pretty thoroughly.

• WaltC - Saturday, September 21, 2019 - link

  Next Thing is already here: with bioses built around AGESA ABBA, all Zen 2 CPUs are once again hitting their max advertised boosts MHz with ease--just as they did with with the early 1002 AGESAs--just like mine. How did we get to "overclocking is a guaranteed function of all cpus," ? when the *the fact* is just the opposite--overclocking is never guaranteed in any form by any CPU manufacturer. Oh--that includes Intel--which I guess I have to spell out. "OC headroom" is not something smart people care about--CPU performance, however, is very much something people care about. MHz, remember, is only half the performance picture--the other half--the more important part--is IPC execution. OC is not a "feature" guaranteed or warrantied" by any CPU manufacturer to any degree whatsoever. Sorry that you don't understand that OC is not a "feature"....*snicker*. You might be interested to know that 90% of the people who buy CPUs made by anyone--at least 90%-- do not bother OC'ing them at all.

• dgingeri - Saturday, September 21, 2019 - link

  I think the biggest thing in this change is that we shouldn't have to overclock. With Intel, we've gotten used to them lowballing us and nerfing the chips to force us to overclock in order to get what we paid for, and people even made that into a sport. Overclocking became a 'feature' because Intel has been artificially sandbagging and cheating us for years.
  
  With the Zen chips, it is more of "you get what you paid for out of the box" kind of mentality and that different tiers of products actually offer valid upgrades for the extra money instead of just lesser degrees of artificial nerfing from the manufacturer.

• Harry Voyager - Sunday, September 22, 2019 - link

  I'm neutral on this. On the one hand, AMD chips not meeting their boost and having low head room is a drawback, however the performance we are seeing in the benchmarks is with those same lower boosts and lower headroom.
  
  Yes, it's a downer that they don't seem to hit the peak speeds, but at the same time, if you have a multi threaded load, they are still out performing Intel by a serious margins and nearly on par in single threaded loads. That is ultimately what matters.