News AMD Ryzen 3000 XT Refresh Reportedly Pushes Boost Clocks Up To 4.8 GHz

InvalidError

Titan
Moderator
Either AMD is micro-binning to milk a few extra bucks out of people or Zen 3 is slipping by some number of months. Between covid-19 messing everyone's schedules and next-gen console launches, AMD may need to borrow CPU 7nm wafers to meet Sony and Microsoft contract obligations..
 

vinay2070

Distinguished
Nov 27, 2011
255
58
18,870
Useless nanosecond boosts. A decent all core boost/manual overclocks to 4.5GHz OR atleast 1 or 2 core boost at 4.8Ghz for a few seconds would be helpful. PBO is so power hungry.
 

jpe1701

Honorable
I know my 3900x that I bought in February has 4 of its cores able to hit 4.61 ghz so unless I just got lucky it seems the node has improved since the boosting problems they had early on.
 
My 3700x ran hot on PBO. The board pushes more voltage to the CPU. 1.4+??? I undervolted to 1.25V and run the whole chip at 4.2GHz on the stock cooler. Much cooler. The average FPS might drop by 1% which I am not worried about.
Running hot is a product of cooling. Remember that if your CPU is pulling 117W and your cooler is rated for a 125W TDP, your temps will be high probably in the 85+C range.
 

Chris Fetters

Distinguished
BANNED
Dec 6, 2013
31
12
18,535
Useless nanosecond boosts. A decent all core boost/manual overclocks to 4.5GHz OR atleast 1 or 2 core boost at 4.8Ghz for a few seconds would be helpful. PBO is so power hungry.
You do realize PBO is an optional feature right? And it can actually drop your single-core clocks/performance if pushed too hard due to thermal reasons, and ESPECIALLY by setting everything to basically infinite. (It can't stay at max single-core boost as long due to the increased voltages). The secret to PBO is manual tuning of the parameters to maximize all-core boost w/o negatively impacting single-core. Definitely takes some work & time to find that sweet spot though. And many single-thread processes have super short spikes in load, so a short time at max boost IS definitely still useful vs not having that boost at all. Also, if you're running the latest AGESA version (1.0.0.5) it'll hit the max boost on a single core pretty regularly, especially with a good undervolt.

And the stock all-core boost for the R9 3900XT will be around 4.3 - 4.4GHz (up from 4.0 - 4.1 GHz), meaning a 4.5GHz overclock is totally within reason (not that I would do it though, ≈4.35GHz all-core + 4.8GHz max single-core boost > 4.5GHz all-core with no boost). That's not where the biggest performance gains from the refresh are going to come though. It's the +200MHz to the maximum Infinity Fabric clock that's the total game changer. Overclocking a Ryzen 3000 chip to like 5GHz on LN2 barely improves gaming performance after like 4.4 - 4.6GHz or so, due to major core-to-core (specifically CCX to CCX) communication bottlenecks. Running 4000MHz RAM with a 2000MHz F-Clock will essentially unlock a new gaming performance tier Ryzen's never been able to hit before.
 
  • Like
Reactions: gg83 and Makaveli

King_V

Illustrious
Ambassador
Call me a cynic, but I imagine there's much hand-wringing among the "Intel is better because its clock speeds are higher" crowd... and they will still talk about how AMD can't compete because they can't cross the 5GHz threshold, or somesuch.
 

JaSoN_cRuZe

Honorable
Mar 5, 2017
457
41
10,890
I think AMD will announce zen 3 by September/October but only the higher end and the lower end like the 8 and 6 core parts at the end of the year or beginning of 2021. As promised by Lisa su we will get zen 3 announcements this year but not the entire product stack.
 
May 26, 2020
4
0
10
If this is true (I don't care if the names a 3900XT or perhaps 3950(x) ).
My guess is that while zen2 use the org. 7nm the zen3 use a process that is a tad better, but in general the same (same design rules or at least close), so to validate the setup AMD moved zen2 to the newer process, and since it probably worked this is what we see.
And now it also make sense that the 450 chipset should be without zen3, because they will have these new "turbo" chips.
 

gg83

Distinguished
Jul 10, 2015
639
293
19,260
You do realize PBO is an optional feature right? And it can actually drop your single-core clocks/performance if pushed too hard due to thermal reasons, and ESPECIALLY by setting everything to basically infinite. (It can't stay at max single-core boost as long due to the increased voltages). The secret to PBO is manual tuning of the parameters to maximize all-core boost w/o negatively impacting single-core. Definitely takes some work & time to find that sweet spot though. And many single-thread processes have super short spikes in load, so a short time at max boost IS definitely still useful vs not having that boost at all. Also, if you're running the latest AGESA version (1.0.0.5) it'll hit the max boost on a single core pretty regularly, especially with a good undervolt.

And the stock all-core boost for the R9 3900XT will be around 4.3 - 4.4GHz (up from 4.0 - 4.1 GHz), meaning a 4.5GHz overclock is totally within reason (not that I would do it though, ≈4.35GHz all-core + 4.8GHz max single-core boost > 4.5GHz all-core with no boost). That's not where the biggest performance gains from the refresh are going to come though. It's the +200MHz to the maximum Infinity Fabric clock that's the total game changer. Overclocking a Ryzen 3000 chip to like 5GHz on LN2 barely improves gaming performance after like 4.4 - 4.6GHz or so, due to major core-to-core (specifically CCX to CCX) communication bottlenecks. Running 4000MHz RAM with a 2000MHz F-Clock will essentially unlock a new gaming performance tier Ryzen's never been able to hit before.
so are they improving the ccx to ccx as time goes by? or just learning how to manipulate it better?
 

vinay2070

Distinguished
Nov 27, 2011
255
58
18,870
You do realize PBO is an optional feature right?
You probably missed my next post?
The secret to PBO is manual tuning of the parameters to maximize all-core boost w/o negatively impacting single-core. Definitely takes some work & time to find that sweet spot though. And many single-thread processes have super short spikes in load, so a short time at max boost IS definitely still useful vs not having that boost at all.
There shouldnt be any manual tuning required, it should be pretty much automatic. Not everybody is a pro or has the time to sit and do that. If advertised by AMD, it should pretty much work out of the box. Aleast in my case, my 3700X still boosts the wrong core and does 4.4 rarely.
Also, if you're running the latest AGESA version (1.0.0.5) it'll hit the max boost on a single core pretty regularly, especially with a good undervolt.
I dont have the 1005 yet, but I wont bother to manually change all the settings for few FPS increase like I mentioned in my previous post. 4.2GHz all core is good enough over nanosecods/milliseconds 4.4.
And the stock all-core boost for the R9 3900XT will be around 4.3 - 4.4GHz (up from 4.0 - 4.1 GHz), meaning a 4.5GHz overclock is totally within reason (not that I would do it though, ≈4.35GHz all-core + 4.8GHz max single-core boost > 4.5GHz all-core with no boost). That's not where the biggest performance gains from the refresh are going to come though. It's the +200MHz to the maximum Infinity Fabric clock that's the total game changer. Overclocking a Ryzen 3000 chip to like 5GHz on LN2 barely improves gaming performance after like 4.4 - 4.6GHz or so, due to major core-to-core (specifically CCX to CCX) communication bottlenecks. Running 4000MHz RAM with a 2000MHz F-Clock will essentially unlock a new gaming performance tier Ryzen's never been able to hit before.
Lets hope so, but I wont bet on it. Zen 3 it is all the way.
 

spongiemaster

Admirable
Dec 12, 2019
2,273
1,277
7,560
Call me a cynic, but I imagine there's much hand-wringing among the "Intel is better because its clock speeds are higher" crowd... and they will still talk about how AMD can't compete because they can't cross the 5GHz threshold, or somesuch.

And they would be right. For the overwhelming majority of users that are on 8 core with SMT or even 6 core with SMT, higher clock speeds will be more beneficial in real world use than adding more cores. Granted, if we're talking 100-200 more Mhz, as we often are these days, the comparison is very little improvement all of the time, vs no improvement most of the time with more cores. Not exactly exciting choices.
 
  • Like
Reactions: vinay2070
And they would be right. For the overwhelming majority of users that are on 8 core with SMT or even 6 core with SMT, higher clock speeds will be more beneficial in real world use than adding more cores. Granted, if we're talking 100-200 more Mhz, as we often are these days, the comparison is very little improvement all of the time, vs no improvement most of the time with more cores. Not exactly exciting choices.
I think @King_V is referencing the old P4 days where a 2.0GHz P4 was slower than a 1.5GHz Athlon XP.
 
  • Like
Reactions: vinay2070

escksu

Reputable
BANNED
Aug 8, 2019
878
354
5,260
My 3700x ran hot on PBO. The board pushes more voltage to the CPU. 1.4+??? I undervolted to 1.25V and run the whole chip at 4.2GHz on the stock cooler. Much cooler. The average FPS might drop by 1% which I am not worried about.

Yes, PBO do use higher voltage than necessary... Its better to oc manually instead.
 
  • Like
Reactions: vinay2070

King_V

Illustrious
Ambassador
I meant it in terms of how there's certain people who come in, and no matter what improvements are made, say that AMD's improvements don't matter because Intel had clock speed superiority.

And then go through every possibly way to justify it.

Is clock speed beneficial in some tasks? Obviously. In everything? Obviously not.

Or, when the sacrifice is made in terms of ridiculous power consumption to get those speeds? Well, that gets ignored, because Intel or bust.
 

escksu

Reputable
BANNED
Aug 8, 2019
878
354
5,260
Call me a cynic, but I imagine there's much hand-wringing among the "Intel is better because its clock speeds are higher" crowd... and they will still talk about how AMD can't compete because they can't cross the 5GHz threshold, or somesuch.

Well, its all about marketing and both companies are equally guilty. During P3/P4 days, Intel keeps boasting about their clockspeed and AMD trying to show that IPC matters and not clockspeed.

During bulldozer days, the reverse happens. AMD had their 5GHz and "8 core" FX9590. Focus became clockspeed instead of IPC.

Now in Ryzen, focus became more cores. For the average joe, they don't know whats performance. They only see clockspeed and cores.
 
May 11, 2020
2
0
10
AMD is kind of messing with my plan. I built a new computer in early February based on a B450 motherboard and the so called Ryzen 1600 AF. The idea was that I would upgrade to the Ryzen 3600X when the prices dropped, but now I am not certain that will happen now that the 3600XT has been leaked. Will the Ryzen 3600XT be compatible with by B450 motherboard? Will the Ryzen 3600X just disappear?
 

King_V

Illustrious
Ambassador
Well, nothing to really worry about as long as the 1600AF is doing the job fine. Plus there's the 4000 series which should drive the prices down on the 3000 series as well.

I wouldn't sweat it until the 1600AF starts letting you down, performance-wise.