Ryzen 9 3950X delayed due to "unsatisfactory clock speeds"

ir0nw0lf

Supreme [H]ardness
Joined
Feb 7, 2003
Messages
6,404
And boom. This is going to stir the pot pretty heavily...

https://www.neowin.net/news/amds-ry...ue-to-unsatisfactory-clock-speeds-says-report

A few days back, AMD officially delayed the 16-core 32- thread Ryzen 9 3950X to November which was previously expected to be released this month. The announcement did come as a surprise since only a couple of days earlier, an image of the 3950X had made its way to Reddit. AMD stated rather vaguely that the company wanted to meet the huge demand of the first 16-core mainstream CPU without going into more detail.

A new report by Digitimes, however, claims that the Ryzen 9 3950X is facing boosting issues in its current state and the chip is exhibiting "unsatisfactory clock speeds". This could be accurate since AMD's Ryzen processors are quite notorious for running at lower speeds than advertised. The clocking issue was especially pronounced in the new 3000 series Ryzen CPUs as the majority of these chips even failed to ramp up to their advertised boost speeds. Famous overclocking guru "der8auer" conducted a survey which showed that only 5.6% of all the participants' chips were hitting the advertised boosts.

The problem was later acknowledged by AMD and the company released a firmware update dubbed AGESA 1003ABBA to fix the issue. Tests showed that the update seems to have fixed the clocking issue on the 12-core Ryzen 9 3900X and other third-gen chips. However, it is possible that this BIOS fix may not be working on the 16-core Ryzen 9 3950X forcing AMD to defer its launch by a couple of months.

Source: Digitimes via SeekingAlpha
 
Plausible.

I think it has more to do with fab capacity - we still don’t see sufficient volume of 3900Xs

but it could be this
not evel close to enough volume, They havent fulfulled all their orders taken in July yet let alone many placed since then. The EPYC's and many of the Threadrippers are destined for OEM's and they are going to want guarantees on availability.
 
Makes me wonder if it's an issue with the chiplet process. The 3900 is two 6-core chiplets (Ryzen 5). It's also the only two-chiplet CPU out right now. The Ryzen 7's are 8 core but single chiplet (I believe). The 3950 would be 2 8-core chiplets.

It doesn't seem like there's an availability issue with Ryzen 7 availability. Certainly not with Ryzen 5. Maybe something to do with the IO chip rather than the actual CPU chiplets themselves.

I think Lakados is probably closer to the reality - production commitments on Epyc probably eating into Ryzen fab time, coupled with TSMC's being at capacity on 7nm.
 
Plausible.

I think it has more to do with fab capacity - we still don’t see sufficient volume of 3900Xs

but it could be this


These don't necessarily have to be different issues.

Insufficient fab capacity could be forcing them to use lower quality chips in their top bin, which in turn could be resulting in the disappointing clock speeds...
 
Last edited:
Make a 3950 vanilla with a lower boost speed. Charge $650... Profit. Make a 3900 vanilla with a lower boost speed charge $449.. profit. No worries about the CPUs hitting clocks that are too damn high.


High core counts are mostly useless to all but a very small minority of users with specialized workloads.

A typical user benefits much more from even very small clock speed (or IPC) increases than they do from adding more cores.

Of course, you are probably right. Someone will buy them. Probably in a misguided attempt to run a server on consumer hardware.
 
I don't know why everyone is so surprised. TSMC isn't known for Intel like clock speeds. Their process has never been able to do what Intel does. Not a fanboi statement...simply reality.
Reduced clockspeed on AMDs newest chips is a casualty of the process shrink more than anything.

Like gluing together two 4 cylinder engines out of front wheel drive Corollas and calling it a "V8", something was bound not to sit well.
 
Last edited:
whole article is food for thought. Clock issues? Sure, lack of supply? more likely.
 
What's the excuse with the previous chips then?

TSMC didn't do too well with their 14/12 nm chips as far as clockspeed goes.
Legally speaking? If I'm AMD/TSMC, I think there's some merit to the idea of mounting a vigorous "hey man, nobody's perfect, and besides, let's see you make a chip if you think you can do better" legal defense.
 
What's the excuse with the previous chips then?

TSMC didn't do too well with their 14/12 nm chips as far as clockspeed goes.

This only makes sense if the fabrication process is the only factor. There are many other factors which determine max clock speed.

Unless you want to say that the Global Foundry fabs back when AMD owned them had a better fabrication process than Intel during the P3 and K7 era. AMD's CPUs clocked higher than Intel's back then.
 
High core counts are mostly useless to all but a very small minority of users with specialized workloads.
A typical user benefits much more from even very small clock speed (or IPC) increases than they do from adding more cores.
Of course, you are probably right. Someone will buy them. Probably in a misguided attempt to run a server on consumer hardware.

For someone who wants/needs 16 cores, a few less 100 MHz will hardly be noticed. A lower price for an "E" version (to bring back the FX days) would be welcomed.
No need to mention "typical user" here. No typical user is buying a 16/32 CPU.
 
For someone who wants/needs 16 cores, a few less 100 MHz will hardly be noticed. A lower price for an "E" version (to bring back the FX days) would be welcomed.
No need to mention "typical user" here. No typical user is buying a 16/32 CPU.

I agree with this, this is why I'm planning to just shoot for a used 1950x in my upcoming workstation and just overclock it a bit. I'm sure the 3950x would be quite a bit faster for running compilers, but the price premium will be pretty huge. I went with a 3700x for my gaming rig and decided to just keep my work rig separate.

I'm curious how much dual channel memory will come into play compared to quad channel now that we're talking 16 cores. Ryzen 3xxx is clearly very memory speed hungry even with regard to games (maybe just lower latency helping there?), I kind of wish x570 brought quad channel support either way.
 
I'm curious how much dual channel memory will come into play compared to quad channel now that we're talking 16 cores. Ryzen 3xxx is clearly very memory speed hungry even with regard to games (maybe just lower latency helping there?), I kind of wish x570 brought quad channel support either way.
Agreed, that is my biggest concern as well.
Dual-channel memory is fine with 8 cores, but 16+, the amount of memory bandwidth gets smaller and smaller per core, and a quad-channel memory controller (and supported motherboard) would be extremely welcome.

Damn top-end of the mid-range. :p
 
I'm curious how much dual channel memory will come into play compared to quad channel now that we're talking 16 cores. Ryzen 3xxx is clearly very memory speed hungry even with regard to games (maybe just lower latency helping there?), I kind of wish x570 brought quad channel support either way.
Only you will know. RAM bandwidth is so application dependent. Not a one serves all use case scenario unfortunately.
 
Only you will know. RAM bandwidth is so application dependent. Not a one serves all use case scenario unfortunately.

Really not sure. Been ages since I've read anything on the matter for the GNU compilers. Last I knew the bottleneck was cache, not sure if that would still be the limiting factor.

gn-cpu-benchmark-gcc-compile-test.png


Usually nobody cares about this kind of thing but at least gamer's nexus did some benching. I believe the 2990WX drops down to 8 cores in game mode, falling a bit behind other Ryzen CPUs. So i'm going to wager more cores and more clocks would be the limiter for me and not memory speed. Still, quad channel sounds fun cuz it's twice as much teehee :p

(Maybe the Ryzen chips are faster because they have more l1 and l2 cache per core than competing intel parts.. in which case the 3950x should be a GCC monster)
 
Reduced clockspeed on AMDs newest chips is a casualty of the process shrink more than anything.

Like gluing together two 4 cylinder engines out of front wheel drive Corollas and calling it a "V8", something was bound not to sit well.
Car analogies are always ill advised. Technically, a V8 is two 4-cylinder engines joined together by a single crankshaft.

The original W16 design that was the basis for the engine that eventually powered the fastest production car in the world was four 4-cylinder motorcycle engines stuck together.

upload_2019-9-25_10-4-23.png
 
Car analogies are always ill advised. Technically, a V8 is two 4-cylinder engines joined together by a single crankshaft.

The original W16 design that was the basis for the engine that eventually powered the fastest production car in the world was four 4-cylinder motorcycle engines stuck together.

View attachment 189336

I'm glad someone caught that. I was gonna say that "slapping two engines together" ala W12 and W16 were some of the most amazing ways to increase engine performance. XD
 
"Sources" LOL! I followed the articles down and say nothing except basically, "unnamed sources."
 
I'm glad someone caught that. I was gonna say that "slapping two engines together" ala W12 and W16 were some of the most amazing ways to increase engine performance. XD

Epyc Rome anyone?
"gluing" or "slapping" stuff together has worked pretty well for CPUs so far and engines.
Sure there might be some trade offs but the pros FAR outweigh the cons (cheaper being the number 1)

Haven't seen any 10+ liter V8s or any V16s
 
Car analogies are always ill advised. Technically, a V8 is two 4-cylinder engines joined together by a single crankshaft.

The original W16 design that was the basis for the engine that eventually powered the fastest production car in the world was four 4-cylinder motorcycle engines stuck together.

View attachment 189336

Add 2 more and it'll be a radial!
 
Car analogies are always ill advised. Technically, a V8 is two 4-cylinder engines joined together by a single crankshaft.

The original W16 design that was the basis for the engine that eventually powered the fastest production car in the world was four 4-cylinder motorcycle engines stuck together.

View attachment 189336
"Hold my vodka." - Russia
 
Epyc Rome anyone?
"gluing" or "slapping" stuff together has worked pretty well for CPUs so far and engines.
Sure there might be some trade offs but the pros FAR outweigh the cons (cheaper being the number 1)

Haven't seen any 10+ liter V8s or any V16s

I work daily on two 27L V12s, two 52L V12s, a handful of 87L V20s -- and I've seen some bigger than that. Granted, it's industrial equipment, but it exists.
 
  • Like
Reactions: N4CR
like this
I work daily on two 27L V12s, two 52L V12s, a handful of 87L V20s -- and I've seen some bigger than that. Granted, it's industrial equipment, but it exists.

Yeah, I thought we were talking production cars.............
Don't stop there though, what about the Wärtsilä RT-flex96C? 14 cyl 1810 liters per cylinder
1 piston displaces more than the whole 87L V20 you mentioned.
 
Yeah, I thought we were talking production cars.............
Don't stop there though, what about the Wärtsilä RT-flex96C? 14 cyl 1810 liters per cylinder
1 piston displaces more than the whole 87L V20 you mentioned.

I have worked on some Wartsilas before - none that big. I think Mann has a bigger one that sits in a cargo ship, the block is like 3 stories tall.
 
this reads like doctored BS fake news... Some website thinks out loud that a non-problem associated with existing cpu's that's already been fixed is why an unreleased cpu is being delayed and another website takes that and tries blowing it up into a big news conspiracy and we're supposed to be just buy into it?

no.


None of that even makes remotely any sense. The claims that amd cpu's routinely dont reach their posted clocks is also garbage. The idea that the current 3000 series issue would delay an entire product release that they could just change the marketing to illustrate the 100Mhz or so lower clock rate to resolve is also idiotic.

The only thing that needs to be discussed about this is if digitimes is purposely being retarded with their posted theories, and how much of a troll seekingAlpha is for not putting any common sense thought into reading them.
 
Reduced clockspeed on AMDs newest chips is a casualty of the process shrink more than anything.

Like gluing together two 4 cylinder engines out of front wheel drive Corollas and calling it a "V8", something was bound not to sit well.


ummm here's epyc
Reduced clockspeed on AMDs newest chips is a casualty of the process shrink more than anything.

Like gluing together two 4 cylinder engines out of front wheel drive Corollas and calling it a "V8", something was bound not to sit well.
Tractor_pulling
some one tell that to these guys.

didnt work
but check out tractor pulls. lol
 
No typical user is buying a 16/32 CPU.

there will probably be a lot more than you think. 8 cores is gonna be the new mainstream, especially when the new consoles come out with 8 core chips in them and all the new games will be requiring an octo-core cpu.
 
Back
Top