AMD Zen 3 (5000 Series), rumored 17% IPC gain.

Cenuij · 9 Nov 2020 at 12:45

alec said:
For segmentation reasons AMD has hardcoded boost clocks in AGESA. 5025 MHz for 5950X, 4925MHz for 5900X and 4825MHz for 5800X. These are not binning, silicon lottery limits. This is artificial ceiling.
If 8 core was allowed to boost as high as 16 core, it would make top SKUs look bad in games.

As far as I understand it, AMD really want people to get the advertised boost speeds. And since they have to bin the dies anyway surely the BIOS boost cap is to ensure stability and meet the advertised boost?

FoxEye · 9 Nov 2020 at 13:11

Cenuij said:
As far as I understand it, AMD really want people to get the advertised boost speeds. And since they have to bin the dies anyway surely the BIOS boost cap is to ensure stability and meet the advertised boost?

It's just as likely to be a purely business decision - to upsell people to the more expensive parts, that they might not necessarily *need*, but will want more if the boost cap is higher...

alec · 9 Nov 2020 at 13:50

Kamakazie! said:
Will there be an easy way to remove these or is it reliant on AMD updating the code at some point?
Seems a bit rubbish to do that then also charge proportionally more per core. I'd expect with decent cooling that that 5800X should sustain higher boost clocks than the 12 & 16 core parts which would see it perform better in games rather than very marginally worse.

I am putting my hopes on AutoOC. Given they made stock PB2 boost working better than stated, and that the silicon clearly has some headroom (all reviewed chips beat their stated boost by 100-150 MHz)

https://www.hardwaretimes.com/amd-ryzen-5000-precision-boost-overdrive-auto-oc-4-85ghz-on-the-5600x/
these guys used AutoOC to get another 200 MHz to 5600X boost, to 4.85

Destination · 9 Nov 2020 at 18:01

Probably worth mentioning.
Under new hardware, usershillbench are listing nvidia GFX cards, and the intel 10700, but not a word on ryzen 5000.
Not a solitary word!
They've also contrived to rig their tables to automatically sort by user rating, listing some intel chips at the top.

Amazingly disgusting, they are truly rating by Intel new standard of 'how is makes you feel'.....

4K8KW10 · 10 Nov 2020 at 05:50

Destination said:
Probably worth mentioning.
Under new hardware, usershillbench are listing nvidia GFX cards, and the intel 10700, but not a word on ryzen 5000.
Not a solitary word!
They've also contrived to rig their tables to automatically sort by user rating, listing some intel chips at the top.

Amazingly disgusting, they are truly rating by Intel new standard of 'how is makes you feel'.....

CPU-Z is also very bad. Right is pre-Ryzen launch in 2017 version, left is post-Ryzen launch version - latest version.

Could anyone explain how magically the performance difference jumped from 99% up to 170%? :confused:

Richdog · 10 Nov 2020 at 05:54

4K8KW10 said:
CPU-Z is also very bad. Right is pre-Ryzen launch in 2017 version, left is post-Ryzen launch version - latest version.

Could anyone explain how magically the performance difference jumped from 99% up to 170%?

Because they, gasp... changed the test calculation logic in the 4 years of application development from 2017 to now?

You seem to lack even the most basic of common sense.

4K8KW10 · 10 Nov 2020 at 05:56

Richdog said:
Because they, gasp... changed the test calculation logic in the 4 years of application development from 2017 to now?

You seem to lack even the most basic of common sense.

Wrong - the performance changed right after Ryzen was launched in April 2017. Until this day, they haven't fixed it.

GregI · 10 Nov 2020 at 07:13

I think something is going on with FCLK. The numbers I saw with FCLK 2000 and 4000C15 RAM does not make sense. This video has 4000C15 beating 3200C14 by like 1-2 fps which is ridiculous considering the advantage it has: https://www.youtube.com/watch?v=LPTbx4fk3rw

I didn't think much of it until I saw this from a popular leaker.

When AMD said they will be releasing an AGESA to enable 2000 they basically meant it will finally allow CPUs to actually act as if they are running at 2000 FCLK.

Patch D is going to be very interesting.

Rom · 10 Nov 2020 at 08:00

As a complete Ryzen new comer.
Is it normal that my fabric is set to 1000?

I tried the auto oc in Ryzen Master,it set my 5900x to a max of 5050mhz. But didn't change the fabric, is that always a manual thing?

Got mine Friday, but haven't been able to post. Got a temp ban for saying the town of a competitor.

chrcoluk · 10 Nov 2020 at 08:09

Zeed said:

watched it in mute, was it mentioned why the 5900X ran so hot?

eeek · 10 Nov 2020 at 08:11

Rom said:
As a complete Ryzen new comer.
Is it normal that my fabric is set to 1000?

I tried the auto oc in Ryzen Master,it set my 5900x to a max of 5050mhz. But didn't change the fabric, is that always a manual thing?

Got mine Friday, but haven't been able to post. Got a temp ban for saying the town of a competitor.

Fabric should be at 1800 unless you want to match it to the memory speed.

Joxeon · 10 Nov 2020 at 08:18

chrcoluk said:
watched it in mute, was it mentioned why the 5900X ran so hot?

He is probably running the 5900X overclocked since he doesn't show the clock speed while the 10900K in stock without even MCE on.

chrcoluk · 10 Nov 2020 at 08:35

Joxeon said:
He is probably running the 5900X overclocked since he doesn't show the clock speed while the 10900K in stock without even MCE on.

Yeah I noticed that as well, that is a good analysis. Is this the same dude that seems to post favourable bench's for ryzen's every launch?

RBImGuy · 10 Nov 2020 at 08:57

Cenuij said:
As far as I understand it, AMD really want people to get the advertised boost speeds. And since they have to bin the dies anyway surely the BIOS boost cap is to ensure stability and meet the advertised boost?

Logic don't make sense to conspiracy people

4K8KW10 · 10 Nov 2020 at 11:31

Cenuij said:
As far as I understand it, AMD really want people to get the advertised boost speeds. And since they have to bin the dies anyway surely the BIOS boost cap is to ensure stability and meet the advertised boost?

https://www.amd.com/en/products/cpu/amd-ryzen-9-5900x

Cenuij · 10 Nov 2020 at 11:59

alec said:
These are not binning, silicon lottery limits. This is artificial ceiling.
If 8 core was allowed to boost as high as 16 core, it would make top SKUs look bad in games.

There's also nothing to stop you disregarding this in BIOS - but as folk have said the 5800x has the same TDP in a single die that the 5900x and 5950x have across two dies, so as long as you understand that and can handle the thermals what's the big problem here?

Armageus · 10 Nov 2020 at 12:12

Usual Charlie Fanboyism aside, there is some decent information about where the IPC increases come from in this article:
https://semiaccurate.com/2020/11/07/a-long-look-at-amds-zen-3-core-and-chips/

Sounds like a lot of attention paid, working out what were the limiting factors in Zen/Zen2, and changes made to help relieve those limits

alec · 10 Nov 2020 at 12:52

Cenuij said:
what's the big problem here?

The rant was triggered by high level reviews bashing 5800X, which was quite shocking when other Zen 3 models got excellent marks.

Meanwhile 5800X is actually the best gaming CPU (once you tweak it to boost to 5GHz) and low-thread workload CPU.
5900X and 5950X are more expensive but don't add value for games.

Richdog · 10 Nov 2020 at 13:16

alec said:
The rant was triggered by high level reviews bashing 5800X, which was quite shocking when other Zen 3 models got excellent marks.

Meanwhile 5800X is actually the best gaming CPU (once you tweak it to boost to 5GHz) and low-thread workload CPU.
5900X and 5950X are more expensive but don't add value for games.

The 5800x doesn't add "value" over a 5600x!

I understand that you bought one and you like it, but...

alec · 10 Nov 2020 at 13:20

bad wording then. I claim it gives absolute fastest performance for those tasks, which can't be achieved by 5600X (even with overclock) and can only be reached by 12 and 16 core, but at a worse value
It is a splitting hairs kind of difference, but definitely not the horrible "DO NOT BUY" GN and HU made it look.