• 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.
There is a reason why Userbenchmark has been banned at multiple places, including Intel's own community driven Reddit channel.

giphy.gif
 
humbug said:
Ryzen 4000 Desktop APU's spotted, these are Zen 2 but like the 3300X should be single CCX.

Top end SKU is the 4800GE, 8 cores 16 threads boost 4.35Ghz, iGPU 8 CU's at 2Ghz.

https://www.hardwaretimes.com/amd-r...ked-8-cpu-cores-and-igpu-on-par-with-the-ps4/
Click to expand...

3300X uses a normal CCD with 2 CCX of which 1 is cut off. So is not different than any Zen 2 CPU.

Now if the Ryzen 4000 APU is single octacore CCD and not 2 quadcore CCXs, that means is based on Zen 3 and not Zen 2.
 
humbug said:
Ryzen 4000 Desktop APU's spotted, these are Zen 2 but like the 3300X
Click to expand...
Mistake somewhere.
They are built like laptop 4x00H Ryzen Mobile parts, with integrated IO die (which helps latencies), but only 8MB total L3 cache. And if I remember correctly, that cache is still split 4MB per 4 core CCX

I really begin to hate AMD APU naming strategy. Confusing everyone.
Why call it next gen 4xxx when for purpose of architecture performance, motherboard compatibility its all 3xxx.
 
Panos said:
Now if that APU is single 8core CCD and boosts similar to 3700X, means it would be faster than any AMD Zen 2 cpu in gaming.
However the design doesn't display that. See the gap between the 2 quad core segments as it similar to the Zen 2 chiplet

wYLlUf9.jpg

Zen 2 chiplet

H1Ncsud.jpg
Click to expand...

Yeah CAT said that, he also said they don't have an IO, he is of course right, i was wrong, but be interesting to see if the lack of an IO die makes any difference, i don't think it does... i could be wrong again. :D
 
alec said:
Mistake somewhere.
They are built like laptop 4x00H Ryzen Mobile parts, with integrated IO die (which helps latencies), but only 8MB total L3 cache. And if I remember correctly, that cache is still split 4MB per 4 core CCX

I really begin to hate AMD APU naming strategy. Confusing everyone.
Why call it next gen 4xxx when for purpose of architecture performance, motherboard compatibility its all 3xxx.
Click to expand...

Its OEM's who want AMD to name these chips as if they are next generation, because that's what they want people to think, APU's ultimately are for OEM's.
 
humbug said:
Yeah CAT said that, he also said they don't have an IO, he is of course right, i was wrong, but be interesting to see if the lack of an IO die makes any difference, i don't think it does... i could be wrong again. :D
Click to expand...

It's been suggested previously that the APUs are monolithic because the GPU is highly dependent on the system RAM and everything being on the same die is better for this. It seems plausible and so in this instance an I/O die is worse but with the lack of anything to compare it to who knows?

There are probably a number of design considerations coming together making APUs what they are right now.

I'm sure all of us are interested in the possibilities chiplets bring in being able to mix/match CPU, GPU, memory and asics onto a single package. Lots to talk about there.
 
alec said:
compare latency between 4900HS and 3950X here https://www.anandtech.com/show/1570...k-business-with-the-ryzen-9-4900hs-a-review/2
3950X needs 81-89ns to read from another CCX on same die, 4900HS needs 61-69 ns. Integrated IO die saves 20ns.
Click to expand...

Yeah i get that :)

There are two things with Zen 2 CPU's:
One is memory latency caused by the IMC being external to the core dies, this is the latency measurable by AIData64.
The other is intercore latency, that's communication between the cores, the core dies are split into 2 4 core clusters, "CCX" which is fine inside an individual CCX but if the cores from CCX 1 need to communicate with cores in CCX 2 the data needs to jump across to the other side through an L3 Cache bridge.

No one really knew if it was Intercore Latency or Memory Latency that caused the discrepancy between IPC in productivity workloads and gaming, because compared to Coffeelake in productivity the IPC is about 13% higher in Zen 2, but in games its about the same, sometimes worse.

The 3300X still has an external IO die IMC, but only a single CCX, the Gaming IPC on the 3300X is much higher than any of the other 3000 Ryzen CPU's where the core difference is not at play, it beats a higher clocked 3950X, its faster in games at around 4.5Ghz than a Skylake / Coffeelake CPU running at over 5Ghz, its gaming IPC matches its productivity IPC.

From that we know its Intercore Latency that's the cause in that Gaming vs productivity IPC discrepancy, the memory latency on the 3300X is the same as it is on all other Zen 2 Desktop CPU's.

The memory latency may have a similar effect, or it may make no difference... we don't know yet, we will find out soon...
 
Last edited:
Status
Not open for further replies.
Back
Top Bottom