• 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 4" thread (inc AM5/APU discussion) ***

i dont understand the push for PCIe standard even in server space. they still heavily reliant on HDD...it not like all of the sudden they are saturating their boards with a load of fast solid state drives.
 
i dont understand the push for PCIe standard even in server space. they still heavily reliant on HDD...it not like all of the sudden they are saturating their boards with a load of fast solid state drives.

Well, in theory the normal PCs can tremendously benefit from faster storage performance. Imagine RAM-class throughput from an SSD. That will make the OS lying on RAM-disk level of performance.
PlayStation 5 has already moved in that direction and will load textures directly from the storage - DirectStorage.
 
Big GPUs plus smart memory access, as well as direct storage, especially with 7.4GB/s SSDs... we may start needing more pcie. Probably not soon. But more so than in the past.
 
i dont understand the push for PCIe standard even in server space. they still heavily reliant on HDD...it not like all of the sudden they are saturating their boards with a load of fast solid state drives.

Isnt that only mainly due to cost and storage capacity?
 
Well, in theory the normal PCs can tremendously benefit from faster storage performance. Imagine RAM-class throughput from an SSD. That will make the OS lying on RAM-disk level of performance.
PlayStation 5 has already moved in that direction and will load textures directly from the storage - DirectStorage.
Ram is volatile memory it can operate faster due to its nature and construciton where solid state drives are not so they can no way be upto the same throughput as ram.

additionally there is architectural limitation on how you cant match RAM level throughput for SSD alone wiht security issues.

anyway fatter lanes that can be utilised by drives would be handy. but really they need to be pushing for reduced latency and better IOPS for small files. fatter lanes wont help those two factors and denser layer NAND will drive up latancy
 
If AMD manages to do another 15-20% IPC improvement AND scale with 5nm at the same time (e.g. more cores, or higher clocked cores), that would be so nice.
 
If AMD manages to do another 15-20% IPC improvement AND scale with 5nm at the same time (e.g. more cores, or higher clocked cores), that would be so nice.

I can see the IPC improvements, but increasing clock speeds when moving to 5nm will be tough. I would suspect similar (if not lower) clock speeds, but i am doubtful on it being higher.

Saying that though, this is AMD and they have done incredible work so far, so I wouldnt put it past them.
 
I can see the IPC improvements, but increasing clock speeds when moving to 5nm will be tough. I would suspect similar (if not lower) clock speeds, but i am doubtful on it being higher.

Saying that though, this is AMD and they have done incredible work so far, so I wouldnt put it past them.

I think there's still an extra 500MHz that AMD can push for within the existing uarch, of course meaningless for 7nm (the extra 10% of clock will cost 50-70% in thermals) but maybe 5nm will make that trade-off better to swallow. I care less about this though than IPC, if 5nm allows them to make bigger cores and therefore deliver better IPC, that's definitely preferred.

AMD had just been outstanding in the last 4 years and looks like they have no intention of slowing down. They could have postponed Zen 3 and blamed the virus but they pushed through and that's great. It's just so bad to be Intel right now :D But you reap what you sow and if you stagnate for 10 years, rest of the world will pass you by even if you had a huge head start.
 
I think there's still an extra 500MHz that AMD can push for within the existing uarch, of course meaningless for 7nm (the extra 10% of clock will cost 50-70% in thermals) but maybe 5nm will make that trade-off better to swallow. I care less about this though than IPC, if 5nm allows them to make bigger cores and therefore deliver better IPC, that's definitely preferred.

AMD had just been outstanding in the last 4 years and looks like they have no intention of slowing down. They could have postponed Zen 3 and blamed the virus but they pushed through and that's great. It's just so bad to be Intel right now :D But you reap what you sow and if you stagnate for 10 years, rest of the world will pass you by even if you had a huge head start.

For sure yeah. Im sure they could do an intel if they wanted and stick to 7nm for the next 5 years trying to squeeze out the absolute maximum, but whats the point? They are much more likely to get better results moving to 5nm (and even tually 3nm and 2nm etc etc) instead of sticking with the same process. Like you said, ghz really doesnt mean much anymore, and its more about the IPC. Rumours are that Zen4 is looking at another 20% improvement over Zen3 which is insane.

Im also glad that they are actually taking their time with this and not trying to rush out products every 12 months exactly, I think most people will be more than happy to wait an extra 3-6 months in each cycle to get a better product. Its also better for AMD as they have more time to sell their existing product as well, not to mention users getting the most out of what they currently have instead of the whole buyers remorse thing knowing within 12 months their product will feel old.
 
AMD must be very careful with what it plans for the next 2-3 years.
Zen 4 Raphael will most probably compete with Meteor Lake, while Zen 3 will be left to compete with Rocket Lake and Alder Lake.

AMD must be very careful not to prioritise heavy cost reductions with N5 process in the form of shrinking the chiplet and keeping the same core count. Making the individual cores slightly wider and polishing the slower parts of the design won't help neither for future-proofing of the line nor for thermal-density and the related problems with the temperatures.
Nobody is happy to see 80-90°C on Zen 3, if Zen 4 is a shrink, things will likely become even worse.

https://www.overclockers.co.uk/forums/threads/ryzen-5800x-87c-in-cinebench.18905709/

Given this - I fully expect the new Zen 4 chiplet, if one exists, at all, to be a 16-core and keep the IPC increases in check in favour of doubling the core count.

If AMD decides to add Navi iGPU and abolish the chiplet design for the consumer line, we will have very serious problems and probably Intel will be back to the top position.
 
AMD must be very careful with what it plans for the next 2-3 years.
Zen 4 Raphael will most probably compete with Meteor Lake, while Zen 3 will be left to compete with Rocket Lake and Alder Lake.

AMD must be very careful not to prioritise heavy cost reductions with N5 process in the form of shrinking the chiplet and keeping the same core count. Making the individual cores slightly wider and polishing the slower parts of the design won't help neither for future-proofing of the line nor for thermal-density and the related problems with the temperatures.
Nobody is happy to see 80-90°C on Zen 3, if Zen 4 is a shrink, things will likely become even worse.

https://www.overclockers.co.uk/forums/threads/ryzen-5800x-87c-in-cinebench.18905709/

Given this - I fully expect the new Zen 4 chiplet, if one exists, at all, to be a 16-core and keep the IPC increases in check in favour of doubling the core count.

If AMD decides to add Navi iGPU and abolish the chiplet design for the consumer line, we will have very serious problems and probably Intel will be back to the top position.

I always said you might not like it, But it is doable and it will sell. There is a threshold people will go to in order to get the performance they crave. And that seems to be about 88c in both GPU and CPU it has been done and Intel look to be going for 5.5ghz and a very hot chip so why not!

Doesnt worry me at all, I was always fond of the heat it blasted at my feet. ;)
 
I always said you might not like it, But it is doable and it will sell. There is a threshold people will go to in order to get the performance they crave. And that seems to be about 88c in both GPU and CPU it has been done and Intel look to be going for 5.5ghz and a very hot chip so why not!

Doesnt worry me at all, I was always fond of the heat it blasted at my feet. ;)

That's the trend we've been going towards anyway, now that thermal throttling has become very reliable manufacturers prefer to keep their chips at the highest levels in their thermal envelopes by taking overclocking into their own hands, that's why you can't overclock the chips by a few percentages more than their boost frequencies these days, and that's even if you're lucky, while in the good old days of C2Q or even i7 920, you could go above 50% with ease.
 
I'm hoping the 6000 (zen 4) series to be;

Ryzen 3s to be 6c/12t
Ryzen 5s to be 8c/16t
Ryzen 7s to be 12c/24t
Ryzen 9s to be 16c/32t

And no price hike like the current 5000 series

Thats what Im hoping for too, but not sure how likely it'll be. I was hoping it was going to be seen this generation
 
AMD Talks Zen 4 and RDNA 3, Promises to Offer Extremely Competitive Products

"Starting with Zen 4, AMD plans to migrate to the AM5 platform, bringing the new DDR5 and USB 4.0 protocols. The current aim of Zen 4 is to be extremely competitive among competing products and to bring many IPC improvements. Just like Zen 3 used many small advances in cache structures, branch prediction, and pipelines, Zen 4 is aiming to achieve a similar thing with its debut. The state of x86 architecture offers little room for improvement, however, when the advancement is done in many places it adds up quite well, as we could see with 19% IPC improvement of Zen 3 over the previous generation Zen 2 core. As the new core will use TSMC's advanced 5 nm process, there is a possibility to have even more cores found inside CCX/CCD complexes. We are expecting to see Zen 4 sometime close to the end of 2021."
 
AMD Talks Zen 4 and RDNA 3, Promises to Offer Extremely Competitive Products

"Starting with Zen 4, AMD plans to migrate to the AM5 platform, bringing the new DDR5 and USB 4.0 protocols. The current aim of Zen 4 is to be extremely competitive among competing products and to bring many IPC improvements. Just like Zen 3 used many small advances in cache structures, branch prediction, and pipelines, Zen 4 is aiming to achieve a similar thing with its debut. The state of x86 architecture offers little room for improvement, however, when the advancement is done in many places it adds up quite well, as we could see with 19% IPC improvement of Zen 3 over the previous generation Zen 2 core. As the new core will use TSMC's advanced 5 nm process, there is a possibility to have even more cores found inside CCX/CCD complexes. We are expecting to see Zen 4 sometime close to the end of 2021."

So about another 20% bump in IPC. That would put them about 25% past Rocket Lake again.
 
The don't mention anything if the CPUs would keep the chiplet design or will become APUs with integrated Navi CUs on the dies.

They say more cores in the CCX - maybe 6 cores in a CCX, 12 cores total, and bye-bye 16 cores top offering...
 
The don't mention anything if the CPUs would keep the chiplet design or will become APUs with integrated Navi CUs on the dies.

They say more cores in the CCX - maybe 6 cores in a CCX, 12 cores total, and bye-bye 16 cores top offering...

Zen 3 has 8 cores in a CCX, so maybe 10 cores? Actually Zen 3 doesn't have CCX's anymore, they are 8 core CCD's, the 5950X has 2 of those = 16 cores, so maybe 20 core top SKU's on mainstream?
 
Zen 3 has 8 cores in a CCX, so maybe 10 cores? Actually Zen 3 doesn't have CCX's anymore, they are 8 core CCD's, the 5950X has 2 of those = 16 cores, so maybe 20 core top SKU's on mainstream?

Yup, 10 cores is possible and plausible - that would leave space for the Navi iGPU.
I don't believe that they will go - 10-core chiplet + 10-core chiplet + Navi iGPU + cIOD.

10-core SKU will put them in a very bad position against the Alder Lake 16-core/24-thread offers.
 
Yup, 10 cores is possible and plausible - that would leave space for the Navi iGPU.
I don't believe that they will go - 10-core chiplet + 10-core chiplet + Navi iGPU + cIOD.

10-core SKU will put them in a very bad position against the Alder Lake 16-core/24-thread offers.

For one 4 of Alder Lakes core are Atom Cores, Intel are using tiny cores to get the core count up, those 4 Atom Cores are next to useless for MT workloads but it looks good on marketing slides, it looks like Intel are taking the fight to AMD in core counts. On paper.

If AMD are adding 2 cores to each CCD they will have 20 'Real Cores' and 40 threads, Zen 3 APU's are 8 core 16 thread + iGPU, APU's are different to mainstream Desktop, they are monolithic, they will have as many cores as AMD chose to give them it is not tied to how many cores are in the Zen 4 CCD, it could be 8 again, it could be 10, it could be 12.

Intel are not putting those 12 + 4 Alder Lake CPU's in Laptop's, Rocket Lake are a maximum of 4 in Laptop's.
 
For one 4 of Alder Lakes core are Atom Cores, Intel are using tiny cores to get the core count up, those 4 Atom Cores are next to useless for MT workloads but it looks good on marketing slides, it looks like Intel are taking the fight to AMD in core counts. On paper.

If AMD are adding 2 cores to each CCD they will have 20 'Real Cores' and 40 threads, Zen 3 APU's are 8 core 16 thread + iGPU, APU's are different to mainstream Desktop, they are monolithic, they will have as many cores as AMD chose to give them it is not tied to how many cores are in the Zen 4 CCD, it could be 8 again, it could be 10, it could be 12.

Intel are not putting those 12 + 4 Alder Lake CPU's in Laptop's, Rocket Lake are a maximum of 4 in Laptop's.

Alder Lake will include 8 big and 8 little cores. The little cores will be as fast as Skylake per clock. The big cores will have hyper-threading enabled.
8 (16) + 8 (8), or 24 logical processors.

CCD is a separate die - you want to see 8-core CCX + 8-core CCX or a 16-core CCX which effectively will be a 16-core CCD but not more than this.

In the best case, Raphael will be a monolithic die with:
- 16 CPU cores with 2-way multi-threading;
- 1024-CU Navi;
- DDR5-capable;
- PCIe 4-enabled;
- USB 4.0-capable.
 
Back
Top Bottom