• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

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

Status
Not open for further replies.
Anyone else find it weird how this time last year there were tons of 3xxx series leaks and rumours and predictions going on, this year absolutely nothing really has surfaced at all.

This is because Ryzen 4000 will launch about 3 months later in the year than Ryzen 3000 did. Release is September/October.

So expect leaks any time from about June onwards.

The info about IF 2.0 being greatly enhanced and the IO being shrunk has me tremendously excited. Seems like it'll be a large leap over 3000.
 
This is because Ryzen 4000 will launch about 3 months later in the year than Ryzen 3000 did. Release is September/October.

So expect leaks any time from about June onwards.

The info about IF 2.0 being greatly enhanced and the IO being shrunk has me tremendously excited. Seems like it'll be a large leap over 3000.

that probably was the original date but whether thats the case now with whats going off around the world who knows. hope so.
 
Anyone else find it weird how this time last year there were tons of 3xxx series leaks and rumours and predictions going on, this year absolutely nothing really has surfaced at all.
That's because there's not much to say about Zen 3. It's evolutionary, rather than revolutionary. All of the important stuff has already been leaked and confirmed, namely the debunking of SMT 4 this generation, the redesign to an 8-core CCX (and at least 32MB unified L3 cache on the chiplet as a result), and about 15% IPC uplift. I don't even think the I/O die is getting shrunk this gen either, certainly not to 7nm. That's coming with Zen 4. Similarly, core counts are staying as-is this gen.
Seems like it'll be a large leap over 3000.
Nah. Evolutionary, not revolutionary. It'll still be great though, but this won't feel as big a leap as Zen 1 to Zen 2 was.
 
That's because there's not much to say about Zen 3. It's evolutionary, rather than revolutionary. All of the important stuff has already been leaked and confirmed, namely the debunking of SMT 4 this generation, the redesign to an 8-core CCX (and at least 32MB unified L3 cache on the chiplet as a result), and about 15% IPC uplift. I don't even think the I/O die is getting shrunk this gen either, certainly not to 7nm. That's coming with Zen 4. Similarly, core counts are staying as-is this gen.

Nah. Evolutionary, not revolutionary. It'll still be great though, but this won't feel as big a leap as Zen 1 to Zen 2 was.

There is no more CCX, the whole CCD is redesigned from ground up and as per AMD, the Zen 3 is more radical changed than Zen 2 over Zen+.
That you do not call "evolutionary".
 
OK, it's not called a Compute Complex any more then, big up the name change. It's still chiplets, it's still I/O die (and probably only 12nm if it even gets a shrink), it's still 7nm (albeit EUV), it's still PCIe 4. There is nothing revolutionary about Zen 3, this is simply an evolution of the chiplet concept. Redesigning a chiplet from scratch doesn't necessarily make it a revolution.

Unless the 8 core CCD is made from jam, uses photons instead of electrons and has a bazillion IPC. That'll be pretty revolutionary.
 
There is no more CCX, the whole CCD is redesigned from ground up and as per AMD, the Zen 3 is more radical changed than Zen 2 over Zen+.
That you do not call "evolutionary".

Em what at moment two 4 CCX makes a single CCD in the current chips. That isn't new. It's just going to 8 CCX, it is moving to 32mb L3 cache per CCX.
 
Em what at moment two 4 CCX makes a single CCD in the current chips. That isn't new. It's just going to 8 CCX, it is moving to 32mb L3 cache per CCX.

Each CCD has 2 CCX. There is an architectural difference between CCX & CCD so don't confuse them. With Zen 3 there won't be any CCX but just the CCD with it's 8 cores. No CCX.
 
Each CCD has 2 CCX. There is an architectural difference between CCX & CCD so don't confuse them. With Zen 3 there won't be any CCX but just the CCD with it's 8 cores. No CCX.
You're splitting hairs. We know what a CCX is (CPU or Core complex) and a CCD (Core Chiplet Die). Just because the actual core count of the computation element is moving to native 8 cores as opposed to a pair of 4 cores, doesn't automatically mean "CCX" is any less wrong. In fact, it's safe and accurate to say that a Zen 3 CCD will utilise a single 8 core CCX; it's still a core complex, it just has 8 cores now instead of 4.

But you're deflecting from the fact this is not "revolutionary", it is an evolution of the decentralised core architecture. Zen 4 is supposedly actually different, perhaps that's where we'll see another "revolution".

Zen 3 is to Zen 2 what Zen+ was to Zen 1; a refinement and upgrade addressing issues and bottlenecks of the established design.
 
Each CCD has 2 CCX. There is an architectural difference between CCX & CCD so don't confuse them. With Zen 3 there won't be any CCX but just the CCD with it's 8 cores. No CCX.

See it's still a CCX with 8 cores as a CCX is the cores plus cache which is what we have with the next release. It's just that it's gone from 4 to 8.

With that technically then you only need a single CCX for an 8 core and no CCD. You only need a CCD if you going for 12 core or more where you are still connecting 2no 6 or 8 core CCX together.

So yeah it's a CCX and there will still be CCD it is just that the ability to increase the number of cores per CCX and by extension CCD has improved and with that means more cores to a single unified cache amount. All is good but it is not what you've stated at all.

We will still need CCX and CCD as definitions because how they work.

So you take your 3900x say which has 4CCX with 16mb cache each and 2CCD and you would only need 2CCX with 32mb cache each and 1CCD to create the 4900x.

And that's the same for the 12 core 4900x that's coming and the 4950x.

This link explains CCX and CCD well.
https://www.google.com/amp/s/www.to...d-ccx-definition-cpu-core-explained,6338.html
 
Zen 2 has 16MB of L3 per CCX.

The 3900X and 3950X have 64MB L3 because they are 4 CCX, 'not 2', 3800X, 3700X, 3600X and 3600 are 32MB L3 because they are made up of 2 CCX. its 2 CCX per die. So 32BM per die. for example the 3600 takes 3 cores from each CCX in the die, each CCX has 16MB L3.
Think of it like 2X 7700K's stitched together in a single die, only with twice the L3 cache each, 16MB vs 8MB.

Zen 3 will also have 32MB per die but 8 cores in a single CCX. Think of this as a 9900K again with 2X the L3 cache, 32MB vs 16MB. A Ryzen 4800X would make up one full die (CCX) A Ryzen 4900X would make up two full dies (2x CCX)

counter to common belief Zen 2 is actually better with very low threaded games than it is with high threaded games, Zen 2 has hardware level AI that keeps the game threads inside a single CCX (where it can) so with CS:GO the performance is excellent as there is no CCX to CCX latency, a medium thered game like FC5 / New Dawn it isn't so good because the threads are moving between CCX's, often Zen 2 also does very well with higher threaded games that use a lot of AI and physics because those are threads that don't tend to move about, Like Assassin's Creed / Hitman.

Zen 3 will make Ryzen behave like a 9900K because its an 8 core CCX, but add on top of that the IPC difference Zen 2 vs Coffeelake, see CS:GO, and then the IPC gain on top of Zen 2 vs Zen 3.

HVnW9L5.png

6uQAqYp.png

S3HsspR.png
 
Zen 2 has 16MB of L3 per CCX.

The 3900X and 3950X have 64MB L3 because they are 4 CCX, 'not 2', 3800X, 3700X, 3600X and 3600 are 32MB L3 because they are made up of 2 CCX. its 2 CCX per die. So 32BM per die. for example the 3600 takes 3 cores from each CCX in the die, each CCX has 16MB L3.
Think of it like 2X 7700K's stitched together in a single die, only with twice the L3 cache each, 16MB vs 8MB.

Zen 3 will also have 32MB per die but 8 cores in a single CCX. Think of this as a 9900K again with 2X the L3 cache, 32MB vs 16MB. A Ryzen 4800X would make up one full die (CCX) A Ryzen 4900X would make up two full dies (2x CCX)

Yep that's all of what I said ;)
 
Status
Not open for further replies.
Back
Top Bottom