News AMD Intros Zen 4 Ryzen 7000 CPUs and 600-Series Chipset: Up to 5.5 GHz, 15%+ Performance, RDNA 2 iGPUs, PCIe 5, DDR5

Bikki

Reputable
Jun 23, 2020
23
12
4,515
We actually looking at a mere 3% IPC increase here if we exclude the 13% frequency boost. With all the advantage they have on TSMC 5nm, AM5, DDR5 ... we expected much more. In fact, TSMC says chip produce with 5mn node has 15% performance increase at same frequency, if AMD did nothing and just port zen 3 to 5nm they may actually have 15% IPC increase. I am really confused.
 
May 23, 2022
7
4
10
We actually looking at a mere 3% IPC increase here if we exclude the 13% frequency boost. With all the advantage they have on TSMC 5nm, AM5, DDR5 ... we expected much more. In fact, TSMC says chip produce with 5mn node has 15% performance increase at same frequency, if AMD did nothing and just port zen 3 to 5nm they may actually have 15% IPC increase. I am really confused.

Or you could literally just read the AMD slides that explicitly state that it's a 15% IPC increase. You can type but you can't read? How odd
 
We actually looking at a mere 3% IPC increase here if we exclude the 13% frequency boost. With all the advantage they have on TSMC 5nm, AM5, DDR5 ... we expected much more. In fact, TSMC says chip produce with 5mn node has 15% performance increase at same frequency, if AMD did nothing and just port zen 3 to 5nm they may actually have 15% IPC increase. I am really confused.

That '~15% performance increase' indeed does not seem all that impressive when coupled with a greater than 10-15% clock speed boost...; I will simply hope that the 7800X can at least surpass the 5800X3D in gaming, which is, of course, the most important workload of all! :)
 

KyaraM

Admirable
Mar 11, 2022
1,494
661
6,790
Or you could literally just read the AMD slides that explicitly state that it's a 15% IPC increase. You can type but you can't read? How odd
From the article:
The quick snap is that AMD claims the Ryzen 7000 processors will have >15% more single-threaded performance than their Zen 3 predecessors (not IPC)
https://cdn.mos.cms.futurecdn.net/vUhEtvv249LAMnxJA6xGY5-970-80.jpg.webp
You mean this slide? It says literally nothing about IPC. Just a general statement about more single-thread performance, which is a function of clock speed, IPC, and cache. You can type, but you can't read? How odd.
 
May 23, 2022
7
4
10
That '~15% performance increase' indeed does not seem all that impressive when coupled with a greater than 10-15% clock speed boost...; I will simply hope that the 7800X can at least surpass the 5800X3D in gaming, which is, of course, the most important workload of all! :)
You guys can't be serious right? They are claiming to be 30% faster in cinebench vs 12900k. How are they going to do that with a 3% IPC improvement over Zen 3. 😂😂😂😂😂 Use your heads a little bit people
 
  • Like
Reactions: SunMaster
May 23, 2022
7
4
10
You guys can't be serious right? They are claiming to be 30% faster in cinebench vs 12900k. How are they going to do that with a 3% IPC improvement over Zen 3. 😂😂😂😂😂 Use your heads a little bit people
The 12900 has 8 P (full) cores and 8 e cores while the zen one will have 16 full cores. ~30% more cores= ~30% more score. (I know this is not exact, but that's the general idea)
 
  • Like
Reactions: shady28 and KyaraM

dk382

Commendable
Aug 31, 2021
50
48
1,560
You guys can't be serious right? They are claiming to be 30% faster in cinebench vs 12900k. How are they going to do that with a 3% IPC improvement over Zen 3. 😂😂😂😂😂 Use your heads a little bit people
They are not claiming to be 30% faster than the 12900K in Cinebench. They are claiming to be 30% faster in Blender during an all-core workload. You are mixing up the different claims AMD has made and seem confused as a result. To be clear, this is exactly what AMD has claimed if you actually look at the footnotes and understand the slides: 1) a 16-core Zen 4 CPU got a 15% higher Cinebench R23 single threaded score than the 5950X, 2) a 16-core Zen 4 CPU got a 30% faster render time in an all core Blender test versus the 12900K. This means a 15% total single-core performance uplift over the 5950X in that one test. Single-threaded gaming performance may be different. Other tests may be different. But it is what it is, just a 15% higher ST cinebench score than the previous gen, even after clocks were boosted by 12.25%. That means the IPC gain is pretty small for this test (~2.5%). We may be returning to the era where Intel was clearly better for single-threaded performance while AMD is clearly better for multi-threaded.
 

KyaraM

Admirable
Mar 11, 2022
1,494
661
6,790
So AMD is claiming to beat the 12900k in cinebench by 30% with a 3% IPC gain over Zen 3.... That's what you're telling me. 🤭😂😂
They claim that for Blender, not Cinebench. You know, Blender, multi-core application that scales very well with more cores in CPU rendering which was tested here? 16 full P-cores vs. 8P and 8E-cores? With the E-cores combined having the performance of a Skylake CPU from, what, 6 years ago at best?
Seriously, I think you really need to work on reading comprehension here, and the one to "use your head"...
 
  • Like
Reactions: martinch
May 23, 2022
7
4
10
The 12900 has 8 P (full) cores and 8 e cores while the zen one will have 16 full cores. ~30% more cores= ~30% more score. (I know this is not exact, but that's the general idea)
And it's the wrong idea. The 12900k is currently faster than the 5950x @ cinebench r23 MT by 5%~.
That's 16 cores vs Intels big little cores.

Tell me, how is AMD going to claim beating the 12900k by 30% if there is only a 15% improvement In overall speed compared to Zen 3?
 
May 23, 2022
7
4
10
They claim that for Blender, not Cinebench. You know, Blender, multi-core application that scales very well with more cores in CPU rendering which was tested here? 16 full P-cores vs. 8P and 8E-cores? With the E-cores combined having the performance of a Skylake CPU from, what, 6 years ago at best?
Seriously, I think you really need to work on reading comprehension here, and the one to "use your head"...
I mixed up cinebench with Blender. However my point still stands. 5950x, using 16 full cores , is about equal with the 12900k.
 
May 23, 2022
7
4
10
They are not claiming to be 30% faster than the 12900K in Cinebench. They are claiming to be 30% faster in Blender during an all-core workload. You are mixing up the different claims AMD has made and seem confused as a result. To be clear, this is exactly what AMD has claimed if you actually look at the footnotes and understand the slides: 1) a 16-core Zen 4 CPU got a 15% higher Cinebench R23 single threaded score than the 5950X, 2) a 16-core Zen 4 CPU got a 30% faster render time in an all core Blender test versus the 12900K. This means a 15% total single-core performance uplift over the 5950X in that one test. Single-threaded gaming performance may be different. Other tests may be different. But it is what it is, just a 15% higher ST cinebench score than the previous gen, even after clocks were boosted by 12.25%. That means the IPC gain is pretty small for this test (~2.5%). We may be returning to the era where Intel was clearly better for single-threaded performance while AMD is clearly better for multi-threaded.
I mixed up cinebench and blender but the point still stands.12900k = 5950x in blender. Wheres the 30% performance coming from? 3% IPC? LOL.

the cinebench test could have easily been done at the same clock speeds. They've made ipc comparisons before doing clock for clock tests
 

dk382

Commendable
Aug 31, 2021
50
48
1,560
I mixed up cinebench and blender but the point still stands.12900k = 5950x in blender. Wheres the 30% performance coming from? 3% IPC? LOL.

the cinebench test could have easily been done at the same clock speeds. They've made ipc comparisons before doing clock for clock tests
They detailed their testing procedures in the footnotes. They did not mention anything about intentionally underclocking the Zen 4 CPU. Lisa Su even said it was from a combination of IPC and higher clock speeds. As for how this gave Zen 4 a 30% lead over Alder Lake in blender, you should realize that different tests will test different parts of the CPU. Perhaps Blender performs better for AMD because of their larger cache. And actually, if you look at past CPU reviews, you'll see that the 5950X already performs better than Alder Lake at Blender, so this is not unexpected.
 
  • Like
Reactions: martinch and KyaraM

KyaraM

Admirable
Mar 11, 2022
1,494
661
6,790
And it's the wrong idea. The 12900k is currently faster than the 5950x @ cinebench r23 MT by 5%~.
That's 16 cores vs Intels big little cores.

Tell me, how is AMD going to claim beating the 12900k by 30% if there is only a 15% improvement In overall speed compared to Zen 3?
Because single-core =/= multi-core and single-core is what people are talking about here, jfc...

It doesn't matter how the 5950X performs here. Single-core performance does not 1:1 translate to multi-core performance. Besides, according to the official table, the 12900K does NOT beat the 5950X; it's the other way round:

The 12900K is better in Cinebench R23 than the 5950X, yes. However, the clock speed of the 12900K is also higher, which, as stated above, also plays a role for performance. You seem to miss that the new Ryzen 7000 got a significant clock rate uplift, which does help considerably with performance considering it means more of the cycles from the term IPC. That's why people state the IPC gain is small. A lot of the strength comes from the clock rate boost.

EDIT:
Actually, reading the article again, it does mention a possible reason for the 30% rendering difference:
Blender supports AVX-512, and although we aren't sure the Ryzen 7000 desktop chips will come with the feature enabled, we do know the Zen 4 microarchitecture supports the instructions. This could contribute to AMD's lead over Intel in this benchmark
If that really is true, extremely higher IPC even less likely the reason.
 
Last edited:
  • Like
Reactions: martinch

jp7189

Distinguished
Feb 21, 2012
321
182
18,860
We actually looking at a mere 3% IPC increase here if we exclude the 13% frequency boost. With all the advantage they have on TSMC 5nm, AM5, DDR5 ... we expected much more. In fact, TSMC says chip produce with 5mn node has 15% performance increase at same frequency, if AMD did nothing and just port zen 3 to 5nm they may actually have 15% IPC increase. I am really confused.
I agree with 3% IPC. I think we are indeed seeing nearly identical Zen 3 cores (with added instructions and cache) that benefit from higher clocks and higher memory bandwidth.
 

ottonis

Reputable
Jun 10, 2020
166
133
4,760
Autumn and holiday season this year will be plenty of fun as Ryzen 7000 and Raptor Lake will be released around the same time.
I hope that AMD played it very safely and was holding back a little bit during today's Computex keynote as an 15% SC "performance" (not IPC) increase seems a little bit short compared to what we used to see from AMD in the last years . Assuming that "performance increase" is the sum of IPC-increase and frequency increase, we did have gen-on-gen IPC-increases north of 15% in each generation Zen+--->Zen2 ---->Zen3. And then add the frequency increases to that.

Raptor Lake, on the other hand, will persumably primarily expand on the multi-core performance by increasing the number of cores.

For me, it seems that Ryzen 7000 may just be a transitional generation that leaves the microarchitecture mostly unchanged but rather focused on three things: 1. new process node, 2. New implementation of iGPU inside the I/O-chiplet and 3. Implementation of DDR5 and PCIe5 standards.
If that's correct, then AMD's focus was to get things running and give the motherboard vendors opportunity to quickly build up working and reliable products (just remember the RAM incompatibility issues during the first year after ZEN2 introduction). My guess is that Ryzen 8000 will introduce new microarchitecture features building on a stable and reliable I/O environment including RAM.

In any ways, I LOVE the competition between AMD and Intel. And let's hope that at least a third non-Apple player will get into the game as well.
 

KyaraM

Admirable
Mar 11, 2022
1,494
661
6,790
Autumn and holiday season this year will be plenty of fun as Ryzen 7000 and Raptor Lake will be released around the same time.
I hope that AMD played it very safely and was holding back a little bit during today's Computex keynote as an 15% SC "performance" (not IPC) increase seems a little bit short compared to what we used to see from AMD in the last years . Assuming that "performance increase" is the sum of IPC-increase and frequency increase, we did have gen-on-gen IPC-increases north of 15% in each generation Zen+--->Zen2 ---->Zen3. And then add the frequency increases to that.

Raptor Lake, on the other hand, will persumably primarily expand on the multi-core performance by increasing the number of cores.

For me, it seems that Ryzen 7000 may just be a transitional generation that leaves the microarchitecture mostly unchanged but rather focused on three things: 1. new process node, 2. New implementation of iGPU inside the I/O-chiplet and 3. Implementation of DDR5 and PCIe5 standards.
If that's correct, then AMD's focus was to get things running and give the motherboard vendors opportunity to quickly build up working and reliable products (just remember the RAM incompatibility issues during the first year after ZEN2 introduction). My guess is that Ryzen 8000 will introduce new microarchitecture features building on a stable and reliable I/O environment including RAM.

In any ways, I LOVE the competition between AMD and Intel. And let's hope that at least a third non-Apple player will get into the game as well.
My issue with "we had more the past couple generations so I expected more this time as well" is that growth isn't infinite. At some point, it will slow down inevitably, even with new techs and all. So to me, expecting big growths every single generation is setting yourself up for disappointment. It's also why, unrelated to the current discussion, I don't take leaks as gospel, either. At the end of the day, all that matters are the results, not what people claimed beforehand. So even here we have to wait and see before we can make the final verdict. Ryzen 7000 might end up better than presented, it might not. We have to see.

That said, I feel that both companies are back on track with their CPUs now and it will be very interesting to see where things develop. I honestly like the Intel architecture a lot and hope they keep their promise on higher efficiency they want to introduce with their 14th gen if memory serves right. Also interesting to see that Ryzen approaches Intel-levels of clock speeds and power draw again. Definitely good to have options and competition. Sadly, this is a very hard field to enter, though. Tachyum seems to have some interesting designs, but seeing how they all exist on paper only so far I don't expect anything from them anytime soon.
 

ottonis

Reputable
Jun 10, 2020
166
133
4,760
My issue with "we had more the past couple generations so I expected more this time as well" is that growth isn't infinite. At some point, it will slow down inevitably, even with new techs and all. So to me, expecting big growths every single generation is setting yourself up for disappointment. It's also why, unrelated to the current discussion, I don't take leaks as gospel, either. At the end of the day, all that matters are the results, not what people claimed beforehand. So even here we have to wait and see before we can make the final verdict. Ryzen 7000 might end up better than presented, it might not. We have to see.

That said, I feel that both companies are back on track with their CPUs now and it will be very interesting to see where things develop. I honestly like the Intel architecture a lot and hope they keep their promise on higher efficiency they want to introduce with their 14th gen if memory serves right. Also interesting to see that Ryzen approaches Intel-levels of clock speeds and power draw again. Definitely good to have options and competition. Sadly, this is a very hard field to enter, though. Tachyum seems to have some interesting designs, but seeing how they all exist on paper only so far I don't expect anything from them anytime soon.

I totally agree that microachitecture cannot improve indefinitely, especially not at that high of a pace. And i didn't criticize or judge that at all. Finite resources can only enable so much growth. As I already said, Raptor Lake will likely not see dramatic microarchitectural changes either but is mostly going to increase just the core counts (something AMD did with ZEN/ZEN+ generations).
 
  • Like
Reactions: KyaraM
Also interesting to see that Ryzen approaches Intel-levels of clock speeds and power draw again.
AMD used 142W for total package power before and now it is going up to 170W. The Intel chip with the closest clocks is the 12900KS that running no PL (standard for most Intel motherboards) pulls 265W. Worst case vs worst case the 12900KS is still pulling >50% more power than the AMD chip despite having 8P + 8E cores. Overall Intel's Core uArch is not very power efficient especially in higher core and frequencies.
 
  • Like
Reactions: King_V and ottonis
AMD used 142W for total package power before and now it is going up to 170W.
The way that it's worded in the article the 170W could be the equivalent of the old 105W and not the 142W.
Current boards are advertised as taking CPUs up to 105W.

In any case, it's way too early and we will have to wait and see.

AMD also revealed that the AM5 socket would support up to 170W CPUs,
 
  • Like
Reactions: KyaraM