AMD delayed Ryzen 9 3950X launch due to unsatisfactory clock speeds?

OK In Canada I checked 10 stores and found 3 with 3900X Stock.

Canada =$676.00 In Stock Good Price
http://www.atic.ca/index.php?page=details&psku=229848

Canada =$715.00 In Stock Little high after exchange
https://www.extreme-pc.ca/showproduct.asp?productid=386945#

Canada =$905.99 In Stock Ripoff
https://www.cdw.ca/product/amd-ryzen-9-3900x-3.8-ghz-processor/5641456?enkwrd=3900x

 

Or could end up burning the whole lot

 

Thank feck I did not wait for this back in August when I decided to go for the 3900X instead and got it within a week. I was thinking it might be delayed and I would have to wait months for it.

 

But, from what I understand, as soon as you manually set an overclock like this, then the speed of cores is fixed... they will always be at 4.3GHz... they will not clock down lower to save power, and they will not clock up to 4.6HGz on single cores if they can for single threaded short boost workloads.

This is what I heard and why people said to not do any manual clock on these processors at all - just leave it to boost itself using the advanced boosting algorithms they have built into chip/bios.

Am I wrong about this?

 

Not being able to build stuff fast enough for demand is a lot better for a company than being able to build stuff fast enough because nobody wants to buy it!

 

@geogan
Depends which boost your talking about.
PBO is to boost past stock, if thermals/power etc allow it.
In reality, i have yet to see one site, able to "consistently" show an increase in performance, even when looking at avg numbers.
So far, all benchs/tests shown differences around 1%..

 

I'm talking about normal boost - from 4.3 up to 4.6 for single cores when needed.
As I said, my understanding is, when people manually OC their CPU to some setting such as 4.3 all core, this stops the CPU from ever going above this, even for a single core. So all single core tasks will be stuck at 4.3 and slower then.

 

i don't know but 3900x have finally landed in Switzerland yay

 

But so far it was happening like this? From Intel 2nd generation 32nm to 10nm and in AMD as well.With decreasing in every NM performance is increasing and Power is decreasing by certain percent till now.So i was believing that as NM decreasing they gets more Density and Area to put more stuff in it.So wasn't it like that?

 

I would write short script that would change affinity for tested software from logical core to logical core in 1 second intervals. That would enable user to see how good each core is.

At idle, I would agree. You can let all cores to tick empty cycles with most of cores not drawing much current.
But as you load many/all cores, power draw will go crazy high.

My 2700X could do 4.3GHz on all cores. But under full load 220W. Definitely not worth it. With Zen2, that point where power draw at hiven clock goes way above reasonable is higher, but still certain clock on all cores is not best of ideas even with good cooling and good board.

 

4.6GHz on all 12 cores? I'd say that would be hugely difficult - a lot of voltage / heat / continuous power used, and cooling would have to be amazing. Have you heard of anyone do this with 3900X?

And just to clarify - in this manual OC case... the CPU never downclocks when idle either? Yes? So IMO it is hugely impractical / inefficient way of running a day to day PC. Nice for short periods for special occasions only.

 

I'm not saying that node shrinks don't bring improvements - they do. But you can't compare Intel's 10nm node to TSMC's 7nm node just by looking at "x"nm node number.

There is fin pitch, width, height, gate length, gate pitch, densities at varying performance requirements, different cell tracks, etc. In some areas Intel's 10nm is significantly head of TSMCs 7nm and in others TSMC is ahead. You can't just look at "7nm" and say it should be better - it doesn't show all the characteristics of the node.

 

3900X can now be bought in the Netherlands with 2-3 days delivery time.

But It took a while !!

 

I'm a little confused about this. Typically, higher core count = lower clock speed, yet with Ryzen 3000 AMD seems to be trying to do the opposite. I also don't think it's necessary - those in the market for a 16-core CPU don't need to have the highest clock speed. I'd be perfectly happy if the 3950X can do 4.6 GHz on single core and 4.2 - 4.3 GHz all core.

 

Well, AMD wanted to give both MT and ST improvement. But I think it was not as good idea.

Maybe it was based on demand approximation and they failed. Logic could have been that demand for 12/16 core chips will be small and therefore they'll have enough of binned chips.

It is not like people look at 3800X with 4.5GHz max boost and say that 3900X with max boost 4.6GHz is worth extra money. Everyone who sees extra value in chip looks at extra cores.
3900X costs 50% more than 3700X and delivers extra 50% cores too. That's the real win there because traditionally more cores means exponential price increase due to lower yields.

I think someone at AMD made little mistake. But it will correct itself over time as TSMC improves on production.

 

I highly doubt that a problem w/ boost clock performance would be a reason for the delay. However, I'd rather AMD push back the release date instead of a POS unusable product being rushed to market.