Tuesday, April 7th 2020

AMD Ryzen 4000 Rumored to Bring 15% IPC Uplift

AMD's Zen 3 architecture will power the next generation Ryzen 4000 desktop chips and the 3rd Gen EPYC lineup which are both expected to launch later this year. Adored TV has received some leaked information detailing the technical specifications of the Zen 3 architecture. The majority of the leaked information confirmed existing rumors such as the 8 core CCX, higher clocks and lower power draw.

However the leak suggests IPC improvements will be less than the expected 20% hinted at by AMD and may end up being closer to 10 - 15%. The leak also claims that L3 cache will remain at 32 MB however it will no longer be split due to the single CCX. While this may be disappointing for some, remember to take the claims with a grain of salt as with any rumor.
Leak
Source: Adored TV
Add your own comment

54 Comments on AMD Ryzen 4000 Rumored to Bring 15% IPC Uplift

#26
Kaotik
Dazzm8Which kind of brings us to the point of this whole topic, Zen2 was an incremental upgrade and not an entirely new architecture, it managed to bring 'around' 15% increased IPC gains.

Zen3 being an entirely new architecture, you can kind of see why people are expecting >15% IPC gains.

For people wondering, the quotes were from Forrest Norrod of AMD if you want to look them up in regards to Zen3 performance that people are reading in to.
No no no, that's why he specifically said that Zen 2's gains were higher than evolutionary upgrades usually get, so people wouldn't use it as reference like "if evolution got this new must get more"
Posted on Reply
#27
SL2
ARFNext year can be Q4 2021, so quite a lot of time till then.
Also, Threadripper can become 64-core
No, it won't happen. Too soon. More cores, maybe. 32, no.
Also, can become? TR3990X is already 64C.
Posted on Reply
#28
KarymidoN
MatsIt won't, because it would defeat the purpose with chiplets (partly).

There's no point in having 16 cores in one chiplet when the vast majority of CPU's sold are 8 cores or less, that's a huge waste of die area.
Imagine a 4600X being sold with TEN cores disabled.. :roll:
Remember, this is still 7 nm, only a tad better than the current one, so the chances of seeing AM4 CPU's with more than 16 cores are extremely slim.
or AMD could just slap intel in the face and increase the core count in all of its chips cause they got so good at making more cores/die. idk maybe not this GEN (ZEN3), but who knows ZEN4 on N5P or N5. few years ago that would be a laugh, but remember AMD is really pushing Intel on every market they can (Desktop, Mobile and server).

@edit
ZEN4 will probab have new Socket (AM4+ or AM5 maybe), with that they could increase the Number of Cores on all of AMD processors and really hit intel again on better Cores/Dollar.
Posted on Reply
#29
SL2
PunkenjoyIf you leave your game on the first die with I/O, Maybe, but as ssoon as there is something going to the other chiplets without direct memory access, the performance will suffer. Just look at the Threadripper second gen vs third gen.

OS, games and other programs do not like heteregenous core. They like when everything is the same and react the same way. Some handle this setup better (like linux run better on threadripper second gen), but they aren't still as fast as a processor without NUMA nodes.
What you're saying is right, but I didn't say how it would be achieved, or if it's even possible.
KarymidoNor AMD could just slap intel in the face and increase the core count in all of its chips cause they got so good at making more cores/die. idk maybe not this GEN (ZEN3), but who knows ZEN4 on N5P or N5. few years ago that would be a laugh, but remember AMD is really pushing Intel on every market they can (Desktop, Mobile and server).
Like I said, more than 16 cores doesn't make sense for mainstream so soon, AMD already have TR for that.
Just look at the sales, if 32C was the next big thing you'd expect 16 to be much more popular than it is right now. 16C is enough of a halo product for mainstream for now, and while it will be more than that in the near future, it won't be 32C.

www.reddit.com/user/ingebor/
Posted on Reply
#30
ARF
KarymidoNor AMD could just slap intel in the face and increase the core count in all of its chips cause they got so good at making more cores/die. idk maybe not this GEN (ZEN3), but who knows ZEN4 on N5P or N5. few years ago that would be a laugh, but remember AMD is really pushing Intel on every market they can (Desktop, Mobile and server).

@edit
ZEN4 will probab have new Socket (AM4+ or AM5 maybe), with that they could increase the Number of Cores on all of AMD processors and really hit intel again on better Cores/Dollar.
They don't have to bother with Intel anymore, to be honest. Intel is no competition right now.
They can and should up the game in the name of the progress - EPYC with 128 cores will be great, the consoles will get 16 logical processors, so mainstream getting 16-core and high-end mainstream getting 32 cores will be welcome for many.

After all, there are new use cases for this type of computing power - ray-tracing, improved physics, artificial intelligence, etc.
Everything feels better with more cores.

Long gone are the days of dual and quad-core processors.

Even today, most working laptops are insanely slow, dual-core with HDD ?! WTH?
Posted on Reply
#31
TheLostSwede
News Editor
Chloe PriceWould be an ok boost from 3rd gen, I decided no to upgrade yet from 2600, now let's just hope that B550 is arriving soon.
Sometime between now and July. I'm afraid I don't have any more detail information, but AMD will announce/launch something CPU related in early July and B550 will be out before then.
Posted on Reply
#32
trparky
Well if it's truly going to be an eight core CCX then that should solve a lot of the latency issues that have been plaguing Ryzen when you have to "talk" to another CCX over the Infinity Fabric.
Posted on Reply
#33
ARF
trparkyWell if it's truly going to be an eight core CCX then that should solve a lot of the latency issues that have been plaguing Ryzen when you have to "talk" to another CCX over the Infinity Fabric.
Well, I was examining what the issue with the so called Infinity Fabric is and it seems that its fundamental flaw is not its actual presence (it's fine to be there) but the fact that it's a 32-bit wide bus operating at the DDR4 frequency.
So, cores talking to each other with 30 GB/s is ultra slow.

Good to know that its speed can scale up to 512 GB/s according to AMD.

It's just an evolution of their old HyperTransport bus (10.4 GB/s and 12.8 GB/s).
Posted on Reply
#34
trparky
ARFWell, I was examining what the issue with the so called Infinity Fabric is and it seems that its fundamental flaw is not its actual presence (it's fine to be there) but the fact that it's a 32-bit wide bus operating at the DDR4 frequency.

So, cores talking to each other with 30 GB/s is ultra slow.

Good to know that its speed can scale up to 512 GB/s according to AMD.
By doubling the number of cores per CCX it should effectively reduce the times that it has use the Infinity Fabric. More cores per CCX means more chances that processes are going to end up on the same CCX thus eliminating the need to use the Infinity Fabric in most cases.
Posted on Reply
#35
tvamos
GeorgeManImagine someone who bought a Ryzen 1600 in Q2 2017, updated to Ryzen 3600 in Q4 2019, further updating to a used 4900X somewhere around 2022 (or even used 3950X in 2021 if the Zen 3 series prove not to be compatible). I think AM4 is the most successful platform ever, better than LGA775. He's gonna enjoy a highly performing computer for around 6 years without changing any other component.
That's me you described haha. Now I have to upgrade in 2021.
Posted on Reply
#36
birdie
Adored TV has a proven track record of delivering complete BS.

I don't know why you decided to post this rumor - more like a wild want-to-believe in guesstimate.
Posted on Reply
#37
ARF
birdieAdored TV has a proven track record of delivering complete BS.

I don't know why you decided to post this rumor - more like a wild want-to-believe in guesstimate.
He is the only person who posts very good analysis, for example about the chiplet approach superiority, and informing the viewers about the illegal Intel.

Charlie also, but his are behind a pay wall.
Posted on Reply
#38
Ruru
S.T.A.R.S.
TheLostSwedeSometime between now and July. I'm afraid I don't have any more detail information, but AMD will announce/launch something CPU related in early July and B550 will be out before then.
Yeah, there's already one pictured :) videocardz.com/newz/first-amd-b550-motherboard-pictured

Wouldn't need to change motherboard otherwise, but this El Cheapo B450 which I have now, has a very crappy VRM. Better just run this 2600 at stock speeds.
Posted on Reply
#39
Metroid
ARFZen 4 will likely be 16-core on one chiplet.
If AMD has no competition I wonder if that will be true.
Posted on Reply
#40
ARF
MetroidIf AMD has no competition I wonder if that will be true.
It's true and it's confirmed more or less. AMD has no competition neither with the process node, nor innovation level, nor performance, nor efficiency, but does have a problem with the large cash payments flows from the Intel HQ to the OEMs.
And continuous innovation and even more dramatic performance leadership may have an effect to offset those payments and illegal Intel practices.
Posted on Reply
#42
NeuralNexus
efikkanI wouldn't read too much into what a PR person calls a "new architecture", it's probably not the same classification as an engineer would use.
Zen 3 and Zen 4 will still be members of the Zen family, if it was a major architectural overhaul AMD would for sure come up with a different name to distinguish them. So we should expect small to medium architectural changes, which may of course still be solid improvements.
You don't pay attention to AMD much but both Mark Papermaster and Forrest Norrad stated that Zen 3 (4000 series) would be a NEW architecture, NOT A REFRESH AND NOT AN ITERATION on Zen 2.
Posted on Reply
#43
tfdsaf
Yeah, AdoredTV has as much credibility as Bill Clinton, but regardless of that, even 15% IPC improvement would be massive. I would actually be really impressed if Zen 3 is able to deliver 15% IPC improvement over Zen 2. That means that clock for clock Zen 3 would be 15% faster, so a 4600 at 4.2GHz would perform 15% faster, now imagine that 4600 running at 4.7GHz that is easily 20% faster and more.

I think IPC improvement combined with increased clock speeds, combined with more cores or at least same cores for cheaper price will lead to an amazing product stack for the Zen 3 series. If a R5 4600 comes with 8 cores at $200, able to boost up to 4.7GHz and has 15% higher ipc over the 3600, that would be an amazing product.

I'm just hoping for cheap B550 boards with PCI-e 4 support and DDR5 support.
Posted on Reply
#44
ARF
tfdsafI'm just hoping for cheap B550 boards with PCI-e 4 support and DDR5 support.
B550 boards with Zen 3?! Well, the options are two:
-X670 and B650, so that X570 becomes lower tier;
or
-X670, so that X570 becomes the cheaper option, kind of "B650" but not quite.

AM4 is DDR4.
Posted on Reply
#45
Kaotik
ARFZen 4 will likely be 16-core on one chiplet.
ARFIt's true and it's confirmed more or less. AMD has no competition neither with the process node, nor innovation level, nor performance, nor efficiency, but does have a problem with the large cash payments flows from the Intel HQ to the OEMs.
And continuous innovation and even more dramatic performance leadership may have an effect to offset those payments and illegal Intel practices.
No, it definitely is not "confirmed more or less". AMD has told absolutely nothing about Zen 4, except that it will be built on 5nm and server chips codename is Genoa. (also indirectly that it will be DDR5)
Posted on Reply
#46
ARF
KaotikNo, it definitely is not "confirmed more or less". AMD has told absolutely nothing about Zen 4, except that it will be built on 5nm and server chips codename is Genoa. (also indirectly that it will be DDR5)
AMD has said to expect more cores:

Quote from an article: ... On the consumer side, the Ryzen 9 3950X brings an almost-unbelievable boost to 16 cores on mainstream platforms, a tremendous improvement over the standard of four cores just a mere two years ago. As AMD moves forward to smaller processes, that means we could theoretically see another doubling in processor cores in the future. That makes a lot of sense for the data center, but begs the question of how many cores an average consumer can actually use. We asked Papermaster if it would make sense to move up to 32 cores for mainstream users:
"I don’t see in the mainstream space any imminent barrier, and here's why: It's just a catch-up time for software to leverage the multi-core approach," Papermaster said. "But we're over that hurdle, now more and more applications can take advantage of multi-core and multi-threading.[...]"
"In the near term, I don’t see a saturation point for cores. You have to be very thoughtful when you add cores because you don’t want to add it before the application can take advantage of it. As long as you keep that balance, I think we'll continue to see that trend."

www.tomshardware.com/news/amd-cto-mark-papermaster-more-cores-coming-in-the-era-of-a-slowed-moores-law
Posted on Reply
#47
Kaotik
ARFAMD has said to expect more cores:

Quote from an article: ... On the consumer side, the Ryzen 9 3950X brings an almost-unbelievable boost to 16 cores on mainstream platforms, a tremendous improvement over the standard of four cores just a mere two years ago. As AMD moves forward to smaller processes, that means we could theoretically see another doubling in processor cores in the future. That makes a lot of sense for the data center, but begs the question of how many cores an average consumer can actually use. We asked Papermaster if it would make sense to move up to 32 cores for mainstream users:
"I don’t see in the mainstream space any imminent barrier, and here's why: It's just a catch-up time for software to leverage the multi-core approach," Papermaster said. "But we're over that hurdle, now more and more applications can take advantage of multi-core and multi-threading.[...]"
"In the near term, I don’t see a saturation point for cores. You have to be very thoughtful when you add cores because you don’t want to add it before the application can take advantage of it. As long as you keep that balance, I think we'll continue to see that trend."

www.tomshardware.com/news/amd-cto-mark-papermaster-more-cores-coming-in-the-era-of-a-slowed-moores-law
That doesn't mean each chiplet would be 16 cores on Zen 4. He also added the last point to emphasize that you don't necessarily want to expect the core count going up this fast in near future, because while there isn't really saturation point, they don't want to add too many cores before applications are ready to take advantage of them.
Posted on Reply
#48
ARF
KaotikThat doesn't mean each chiplet would be 16 cores on Zen 4.
It has to be because you can't put more chiplets on the substrate for the EPYCs which won't stay limited up to 64-core, that's for certain.
Posted on Reply
#49
Kaotik
ARFIt has to be because you can't put more chiplets on the substrate for the EPYCs which won't stay limited up to 64-core, that's for certain.
Except that with Zen 4 EPYCs move to a new socket, just like desktop and HEDT will, so current restrictions don't need to apply. 5 nanometer chiplets will also be smaller per transistor.
Posted on Reply
#50
R0H1T
NeuralNexusYou don't pay attention to AMD much but both Mark Papermaster and Forrest Norrad stated that Zen 3 (4000 series) would be a NEW architecture, NOT A REFRESH AND NOT AN ITERATION on Zen 2.
That's as vague as it gets, fun fact every major uarch since Conroe are said to be "new" & yet all of them are its evolutions!
Posted on Reply
Add your own comment
Jan 11th, 2025 18:24 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts