Why everyone say Zen 5 is bad ?

[phanbuey]

I feel like 8 cores have turned into the new quad cores with being stuck on them lol. With only multiple ccd or e cores being the option if you want more.

It really did.

and they killed HEDT which sent all of us core whores down to consumer.

 

[mkppo]

Yeah and its a shame its not,

Zen 4 despite all the rumors of underwhelming IPC uplift actually had a decent IPC uplift of like 12-13% across the board in almost all workloads at the same clock speed. And it had a nice clock speed bump so much better than Zen 3 overall.

Zen 5 has no clock speed bump over Zen 4 and almost the same clock speed. And it was suppose to be the one with the better IPC uplift. Except its not and it sucks and only like 5% at best better across almost all workloads except AVX512 and some other rare niche cases where it seems any real performance benefit.

Throwing the word IPC around isn't always ideal. IPC in what? Applications matter at the end of the day and you can either use SPEC to calculate IPC, or make up your own subset of tests that will actually impact you and make up an 'IPC' by averaging them out.

It's the latter category that have people's opinions all over the place. Zen 5 isn't '5% at best better across the board'. If you're talking single threaded, in SPEC it's 15% faster in INT and ~19% faster in FP without taking AVX512 tests into account. Even CB R23 ST shows a similar increase. Chips and cheese also got the same in their own subset of tests. But In MT, on average, Zen 5 is around 7-10% faster on average depending on PBO as there are cases of being memory bound.

But there will undoubtedly be applications that don't benefit from Zen 5's increased registers, front end and execution engines because they are instead back end/memory bound (eg. games) which even brings down the average % gain in apps. This is why some applications will show a decent increase (eg. Blender, ML tasks), some apps will see massive increases (V-Ray or anything that remotely leverages AVX256/512) while others will see little to no gains (games or lighter apps that aren't core bound and are instead memory bound). There are also some apps that get affected by higher cross CCD latencies but those are thankfully pretty rare.

Zen 4 was always going to be a bigger update 'across the board' because it had both a faster core (slight IPC from arch/cache, decent clock bump) and also faster memory with the new IO and DDR5. Consequently all apps got a decent speedup. Zen 5 doesn't have that luxury, because for all the improvements to the core they don't have a new back end/IO nor are there clock increases and unfortunately a lot of the lightly threaded apps and games are memory latency bound.

If AMD just introduced a new IO die and IF for Zen 5 and kept the rest the same, many of these consumer apps would get a larger speedup. Instead, they reworked the core significantly which instead offers very nice gains for apps that can leverage it, and i'm not talking AVX 512. It also sets the stage for future upgrades, because it's evident from the arch that there are nice low hanging fruit gains to be had through some improvements, whereas Zen 4 was nearer to being maxed out at least on the front end/execution side.

Also, since Zen 5 is even more bottlenecked by memory, X3D should theoretically offer even larger gains than Zen 4. But the average % gain will probably not reflect that because there are other bottlenecks in play when testing games. I do hear that they are going to have X3D temperature improvements and offer OC on these as well. If they can keep clocks higher than past generations, it can be a nice uplift overall.

 

[oxrufiioxo]

It really did.

and they killed HEDT which sent all of us core whores down to consumer.

Just so nobody's confused I'm not saying 8 cores isn't enough for 99% of consumer workloads I just want the damn option for more without the BS AMD/intel are shoveling us with dual CCD and E cores.

I fully realize that isn't going to happen and what I want either company couldnt give two shits about doesn't change the fact that the current high core counts aren't ideal and seem like bandaid solutions for those who want more cores.

I think people are confusing want/need... I don't need a 4090 to enjoy gaming I want a 4090 to enjoy gaming and while that obviously makes a much larger difference than a low latency 12 core cpu would I still want it lol.

 

[Wolverine2349]

It really did.

and they killed HEDT which sent all of us core whores down to consumer.

Yeah for sure.

Even when Intel had 0 competition from AMD back from 2014 to 2016, the HEDT options were way way way better.

In a way it would have been better had AMD struggled to get IPC much above Zen 2 and core count above 12 on mainstream consumer desktop platforms. Maybe Intel charges a lot more for the superior far better IPC Golden Cove, but I would take that if they made 10-12 P core option instead of max 8 cores and a bunch of e-cores big.little crap.

That way Intel would not have came up with stupid Big.Little and e-cores because they had to compete perfectly in infinite parallel apps which do not care about latency. Intel then likely would have made 10 P core Golden Cove and 12 P core Raptor Cove on same ring bus right off bat that would have tied or beat 12 core Zen 2 and even Zen 3 in parallelization apps and smoke them in gaming with superior cross core latency. They only made e-cores because their P cores were too big to fit in consumer sized chips unlike AMD's and needed to compete or exceed on multi threading for infinite threaded apps. But its a shame they gave no choice with an alternative of P core only product with more than 8 cores.

And AMD is too cheap or has design limitations to give choice?? Or maybe the CCX and CCD designs and TSMC and their methodology just struggle to get more than 8 cores on one unlike Intel?

Just so nobody's confused I'm not saying 8 cores isn't enough for 99% of consumer workloads I just want the damn option for more without the BS AMD/intel are shoveling us with dual CCD and E cores.

I fully realize that isn't going to happen and what I want either company couldn't give two shits about doesn't change the fact that the current high core counts aren't ideal and seem like bandaid solutions for those who want more cores.

I think people are confusing want/need... I don't need a 4090 to enjoy gaming I want a 4090 to enjoy gaming and while that obviously makes a much larger difference than a low latency 12 core cpu would I still want it lol.

Well maybe 12 P core Bartlett Lake. But we have if rumors are right at least a 10.5 month wait if it can make Q3 July 1, 2025 release date which is 1st day of Q3 2025. And oh will it be stable and reliable? Is it even true or coming?

 

[oxrufiioxo]

Yeah for sure.

Even when Intel had 0 competition back form 2014 to 2016, the HEDT options were way way way better.

In a way it would have been better had AMD struggled to get IPC much above Zen 2 and core ocunt above 12.

That way Intel would not have came up with stupid Big.Little and e-cores because they had to compete perfectly in infinite parallel apps which do not care about latency. Intle then likely would have made 10 P core Golden Cove and 12 P core Raptor Cove on same ring bus right off bat. They only made e-cores because their P cores were too big to fit in consumer sized chips unlike AMD's and needed to compete or exceed on multi threading for infinite threaded apps. But its a shame they gave no choice.

And AMD is too cheap or has design limitations to give choice?? Or maybe the CCX and CCD designs and TSMC and their methodology just struggle to get more than 8 cores on one unlike Intel?

While I agree to a certain extent desktop platforms still got the new architectures first I had both a 5820k and a 6700k and they traded blows with somtimes the 6700k being way faster. Once the 9900k and 10900k hedt was never faster in gaming at least.

Well maybe 12 P core Bartlett Lake. But we have if rumors are right at least a 10.5 month wait if it can make Q3 July 1, 2025 release date which is 1st day of Q3 2025. And oh will it be stable and reliable? Is it even true or coming?

Still too far out to know for sure you'd think they'd have it fixed although even if it's ok not investing in a platform that is eol for it

This should have been an alternative option even if more expensive to the 13900k at launch not 3 years later.

Don't get me wrong I'm happy for lga 1700 owners who want to upgrade to this assuming this isn't some weird orm only or embedded market cpu line.

 

[R0H1T]

Even when Intel had 0 competition from AMD back from 2014 to 2016, the HEDT options were way way way better.

Let's try that again ~

6950x cost at least 4-6x 4790k or 5775c & it was way way way way better

Did I get that right?

 

[Wolverine2349]

Let's try that again ~

6950x cost at least 4-6x 4790k or 5775c & it was way way way way better

Did I get that right?

Well cost was very high. But the option existed to get many more cores on a single ring up to 10. Now no such option exists even if money is of no object.

Sapphire Rapids Xeon Workstation Golden Cove sucks and has far worse IPC and latency than client Golden Cove. And its on a horrible mesh arch which cripples latency too. It just has AVX 512 and workstation heavy features and tons of PCIe lanes. So its not a good option even if you were willing to pay the high price for it.

And back then like 0 games had any benefit from more than 4 cores. Now 8 cores is kind of like the new 4 cores except some games are starting to benefit from more than 8 cores certainly more so than games benefitted from more than 4 back in 2014-2015 or even 2016.

 

[R0H1T]

The part about AMD killing any such option is fair but then you'd have to factor in the cost increase of the platform as well. AMD definitely do not want to sell you a $3k 96c zen5 or 128c zen5c chip, are you willing to shell out $6-12k for them & maybe a grand for a "good" board?

 

[oxrufiioxo]

Let's try that again ~

6950x cost at least 4-6x 4790k or 5775c & it was way way way way better

Did I get that right?

What I would say is hedt offered much better platforms connectivity wise and the 6 core options weren't bad price wise.

I had both and honestly liked the hedt platform better it's just once we got the 9th/10th generation it started to die out.

I think it was 3rd generation Threadrippers that finally killed it but then amd started acting like Intel on that platform charging an insane amount while offering pretty meh support.

 

[Wolverine2349]

While I agree to a certain extent desktop platforms still got the new architectures first I had both a 5820k and a 6700k and they traded blows with somtimes the 6700k being way faster. Once the 9900k and 10900k hedt was never faster in gaming at least.

That is true, though 6700K was Skylake IPC uplift of like 16% over Haswell and better clocks too.

But HEDT Haswell compared to consumer Haswell, neck and neck with the HEDT having many more cores.

I think we would have gladly taken a 6 month to 1 year delay for a Golden Cove 10-12 P cores on a ring bus HEDT option that uses regular XMP non ECC RAM. Instead we had more than 1 year for SPR and oh much worse latency gimped IPC and requires ECC RAM which is not XMP nor overclock friendly. And it costs even more than past HEDT too. I would be iwlling to pay up for it if it had what I wanted but it does not.

Still too far out to know for sure you'd think they'd have it fixed although even if it's ok not investing in a platform that is eol for it

This should have been an alternative option even if more expensive to the 13900k at launch not 3 years later.

Don't get me wrong I'm happy for lga 1700 owners who want to upgrade to this assuming this isn't some weird or only or embedded market cpu line.

I agree there. Way way too far out. They really need to release it now and should have been an option in 2022 or early 2023 at the latest.

But better late than never. Now it is a platform starting to EOL, But better late than never especially may apply much stronger given that Zen 5 performance uplift in gaming was almost nothing over vanilla Zen 4 which means Raptor Cove 12 P core even on older platform released in 2025 (3 years later) should age much better given Zen 5 underwhelming performance improvement situation. Arrow Lake we do not know but early reports are that its uplift is gonna be underwhelming too, but more so a new platform with reduced power draw and clock regression its likely to be nothing uplift as well, but oh stability on all chips instead.

So a 12 P core Raptor Lake assuming stability and degradation is fixed given the underwhelming uplifts of the above will age much better despite its tardy release date. And it gives us what we want.

Well there there is Zen 6 and Nova Lake. but those are a ways ways out even from Q3 2025.

 

[R0H1T]

The thing about Intel HEDT at the time, before Zen, was that it only offered better "connectivity" & PCIe lanes IMO. Intel, if they didn't on their collective arse, could've easily offered 8-10c on regular desktop just as they did 2 years later!

What I would say is hedt offered much better platforms connectivity wise and the 6 core options weren't bad price wise.

I had both and honestly liked the hedt platform better it's just once we got the 9th/10th generation it started to die out.

I think it was 3rd generation Threadrippers that finally killed it but then amd started acting like Intel on that platform charging an insane amount while offering pretty meh support.

So no I don't agree about teh pricing part one bit. AMD's doing almost the same, except they don't even have "HEDT" right now!

That is true, though 6700K was Skylake IPC uplift of like 16% over Haswell and better clocks too.

Skylake came after Broadwell, so 2 gens apart, 3 (years?) if you count Devil's Canyon.

 

[Wolverine2349]

What I would say is hedt offered much better platforms connectivity wise and the 6 core options weren't bad price wise.

I had both and honestly liked the hedt platform better it's just once we got the 9th/10th generation it started to die out.

I think it was 3rd generation Threadrippers that finally killed it but then amd started acting like Intel on that platform charging an insane amount while offering pretty meh support.

Yes I agree. I think 9th and 10th Gen would have aged well had Zen 3 not ever come to life. 10th Gen espeiclaly 10 core Comet Lake was the real deal. But Zen 3 IPC uplift was huge over Zen 2 and Skylake based Comet Lake.

Then Golden Cove 15% uplift and Zen 4 12% IPC uplift over Zen 3.

Though now it seems both Intel and AMD are hitting IPC uplift wall as seen by the Zen 5 results so maybe Golden or Raptor Cove 12 P core part on a ring will age well as it seems probable and likely that the next big IPC uplift will not be until 2027-2028 or later given how AMD and Intel both struggling to improve IPC much from todays levels at least not as fast as they once could.

 

[AnotherReader]

That is true, though 6700K was Skylake IPC uplift of like 16% over Haswell and better clocks too.

But HEDT Haswell compared to consumer Haswell, neck and neck with the HEDT having many more cores.

I think we would have gladly taken a 6 month to 1 year delay for a Golden Cove 10-12 P cores on a ring bus HEDT option that uses regular XMP non ECC RAM. Instead we had more than 1 year for SPR and oh much worse latency gimped IPC and requires ECC RAM which is not XMP nor overclock friendly. And it costs even more than past HEDT too. I would be iwlling to pay up for it if it had what I wanted but it does not.



I agree there. Way way too far out. They really need to release it now and should have been an option in 2022 or early 2023 at the latest.

But better late than never. Now it is a platform starting to EOL, But better late than never especially may apply much stronger given that Zen 5 performance uplift in gaming was almost nothing over vanilla Zen 4 which means Raptor Cove 12 P core even on older platform released in 2025 (3 years later) should age much better given Zen 5 underwhelming performance improvement situation. Arrow Lake we do not know but early reports are that its uplift is gonna be underwhelming too, but more so a new platform with reduced power draw and clock regression its likely to be nothing uplift as well, but oh stability on all chips instead.

So a 12 P core Raptor Lake assuming stability and degradation is fixed given the underwhelming uplifts of the above will age much better despite its tardy release date. And it gives us what we want.

Well there there is Zen 6 and Nova Lake. but those are a ways ways out even from Q3 2025.

Skylake's IPC increase over Haswell was barely 6%. Before Ice Lake, Haswell was the last Intel core with a big uplift in IPC. The 6700k also clocked lower than the 4790k.

Sandy Bridge to Ivy Bridge: Average ~5.8% Up
Ivy Bridge to Haswell: Average ~11.2% Up
Haswell to Broadwell: Average ~3.3% Up
Broadwell to Skylake (DDR3): Average ~2.4% Up
Broadwell to Skylake (DDR4): Average ~2.7% Up

 

[AusWolf]

That's wrong. Six cores only would not be enough for the game and Windows and all of its main and background processes to run without micro-stuttering, and drops of the 0.1% lows.

I call bullshit on that. What Windows processes are there that eat your CPU to the level where you can't game on it anymore? Nah, if I had to choose between a R5 7600 and a 3700X for gaming, I'd choose the 7600.

 

[dont whant to set it"']

Nerfed by ms so intel doesn't look as bad after the stock price hit the fan?(borderline on conspiracy).

 

[Wolverine2349]

I call bullshit on that. What Windows processes are there that eat your CPU to the level where you can't game on it anymore? Nah, if I had to choose between a R5 7600 and a 3700X for gaming, I'd choose the 7600.

Of course. Zen 2 IPC and latency sucked compared to what we have today.

But 7700X or 7600X. 7700X in a landslide.

Or 12 P core Bartlett Lake (as long as stability and degradation problems fixed) vs 7800X3D or 9800X3D. Give me the 12 P core Bartlett Lake as long as they are on a single ring. Same thing with a hypothetical 12 + 0 Arrow Lake die vs 9800X3D or even Zen 6.

Zen 2 and prior are so far behind in IPC and even latency more cores has no meaning or matter at all.

Though even modern core counterparts, we still do not need huge amounts of cores like thread ripper and Xeon provides for gaming. Just would be nice to get to the sweet spot 10-12. 8 is fine for now, but a little more headroom would be nice without a band aid solution dual CCD or e-cores.

 

[oxrufiioxo]

Nerfed by ms so intel doesn't look as bad after the stock price hit the fan?(borderline on conspiracy).

Doubtful, Amd is a partner on their consoles and on their Azure cloud platform.

It also needs to be benchmarked on Intel to verify if there is performance degradation.

I'm still not convinced this isn't just overhead from running a more secure OS and amd is just full of shite....

 

[Wolverine2349]

Nerfed by ms so intel doesn't look as bad after the stock price hit the fan?(borderline on conspiracy).

Well Intel would do nice and wise to make a 12 P core Arrow Lake die as alternative to the others ASAP and release in October and no scheduling issues at all with all 12 P cores on same ring within same tile and smooth as butter with Windows rather than trying to work with MS on thread director yet again. Well maybe Bartlett Lake will be the answer?? But a power heat dumping hog even if its stable.

Arrow Lake will dump much less heat inside the case with its 100W power reduction.

Or how about Intel give us a Skymont only die with all them on a ring instead of 4 Skymonts in a single cluster node we have Raptor Cove IPC with much lower power consumption and homogenous arch wit the new much better e-cores that can become new P cores maybe??

Cannot buy and disable P cores cause Skymonts are in a cluster not all on ring so latency would suck even if it had in theory Raptor Cove IPC. That was once my hope until I saw there are 4 in a cluster??

 

[sLowEnd]

Nerfed by ms so intel doesn't look as bad after the stock price hit the fan?(borderline on conspiracy).

No, it's just Windows' regular everyday suckage (i.e. Windows having more overhead than Linux on some things, and the performance penalty when not running on admin, scheduler not being great...etc.), not special planned suckage

 

[evernessince]

I call bullshit on that. What Windows processes are there that eat your CPU to the level where you can't game on it anymore? Nah, if I had to choose between a R5 7600 and a 3700X for gaming, I'd choose the 7600.

Agreed. Most windows processes run at 0.0% of total CPU usage with a few running at 0.1%. The Windows Search Indexer gets to 0.3% on an 8-core Zen4 CPU when active.

One can't make a blanket statement that 6 cores are not enough because 6 core Zen 4 is not the same as 6 core Zen 3 and earlier. The 6 core 7600X is faster than the 8-core 5700X in multi-threaded applications as the individual cores are much more powerful. I would absolutely take a 7600 over a 3700X as well, it's just a much faster all around CPU.

 

[ARF]

I call bullshit on that.

You must self censor as a beginning, then we will talk, if you are still here..

 

[oxrufiioxo]

Agreed. Most windows processes run at 0.0% of total CPU usage with a few running at 0.1%. The Windows Search Indexer gets to 0.3% on an 8-core Zen4 CPU when active.

One can't make a blanket statement that 6 cores are not enough because 6 core Zen 4 is not the same as 6 core Zen 3 and earlier. The 6 core 7600X is faster than the 8-core 5700X in multi-threaded applications as the individual cores are much more powerful. I would absolutely take a 7600 over a 3700X as well, it's just a much faster all around CPU.

Honestly if the 9600X beat or matched the 7700X in applications and offered similar gaming performance uplifts that the last 3 genenerations gave us nobody would be complaining but here we are not only stuck on the same core count per tier that we've had since 2019 and with negligible performance increases in gaming and most applications the double whammy of suckage.....

The 8 core 9700X should have become the entry level option at this point and even though core counts are higher we are now stuck with a very intel 2015 like generation uplift

Hopefully this generation has some of that AMD fine wine (by fine wine I mean fix their shit lol)

Also X3D could save the day for gaming but it won't fix the meager application performance uplift. Not holding my breath though.

Sounds like we are getting a 10% increase ish with Arrow Lake over Raptor Lake as well while meh would still be better than this.

 

[sLowEnd]

The 8 core 9700X should have become the entry level option at this point and even though core counts are higher we are now stuck with a very intel 2015 like generation uplift

Lisa basically admitted in an interview a little while ago that AMD is sandbagging on core counts

Q&A with AMD CEO Dr. Lisa Su

Computex 2024 Press Roundtable

morethanmoore.substack.com

A: I think our strategy is that we have to increase performance continually. For gaming in particular, the gaming software developers have not necessarily used all the cores from time to time. There is no physical reason we couldn’t go past 16 cores. The key is that we're going at a pace that the software guys can and do utilise it.

 

[SL2]

Bartlett Lake appears to be the only hope

Bartlett Lake is a better hope

Well maybe 12 P core Bartlett Lake

Give me the 12 P core Bartlett Lake

Well maybe Bartlett Lake will be the answer??

Help me, Bartlett Lakenobi. You're my only hope.

That's putting a lot of hope in a rumor. With Intel cutting cost in a lot of places, I'd be surprised to see them keeping this. This isn't the obvious time for them to have (almost) two new CPU types at the same time.. or do they want LGA1700 to become their AM4?

What would we do without hope.

 

[oxrufiioxo]

That's putting a lot of hope in a rumor. With Intel cutting cost in a lot of places, I'd be surprised to see them keeping this. This isn't the obvious time for them to have (almost) two new CPU types at the same time.. or do they want LGA1700 to become their AM4?

What would we do without hope.

Sshhh, let a man dream.

Joking aside I hope it's real and decent for lga 1700 owners.