AMD unleashes Zen 3 and reckons it's faster for gaming than Core i9

[Rubarb]

the king is dead long live the king "hail amd"

And yes im using a i7 atm

[Iota]

Making AMD is trying to Jebait us regarding price,and suddenly will have some discounts,etc prior to sale.

We've seen that before from them, so obviously they have some margin to play with.

[CAT-THE-FIFTH]

We've seen that before from them, so obviously they have some margin to play with.

They have a ton of space to play with. I am very doubtful a 7NM Zen3 chiplet is massively bigger than a Zen2 one,and 7NM should be cheaper than last year. Also apparently AMD is re-using the same I/O die for Zen2.

Intel has been on Skylake for the last 5 years. Its taken 4 generations of Zen and a node shrink to finally beat the ringbus Skylake CPUs in gaming.

If you follow the mobile Intel CPUs,they are already two generations ahead of Skylake and have bypassed Zen2 IPC. Rocketlake S is 3 generations removed from Skylake,so the IPC jump is going to be decent as its going to be a 3 generation leap,and its on a very high clocking 14NM node. Intel 10NM seems to be able to hit around 4.8GHZ too.

The problem is now AMD has priced their 8 core at £430,and there is room for a more expensive XT version(the 5900XT and 5950XT already exist),what if Intel then manages to be faster in gaming than AMD with its own 8C CPU?? They will push the price up too.

Then AMD pushes out a slightly quicker 8C CPU,and then prices it a bit higher than Intel and so on.

AMD can do 12 and 16 cores,and thrash Intel,but if they are going to be priced so high it isn't going to matter for most DIY PC builders.

Is 6C going to be the new 4C for mainstream??

People have learnt nothing from what we saw Nvidia and AMD do with GPUs. We went from £450 being considered expensive for a top end GPU 10 years ago,to £1500~£2000 being the top end. It also meant the entry level and mainstream were pushed up.

[Iota]

They have a ton of space to play with. I am very doubtful a 7NM Zen3 chiplet is massively bigger than a Zen2 one,and 7NM should be cheaper than last year. Also apparently AMD is re-using the same I/O die for Zen2.

Intel has been on Skylake for the last 5 years. Its taken 4 generations of Zen and a node shrink to finally beat the ringbus Skylake CPUs in gaming.

If you follow the mobile Intel CPUs,they are already two generations ahead of Skylake and have bypassed Zen2 IPC. Rocketlake S is 3 generations removed from Skylake,so the IPC jump is going to be decent as its going to be a 3 generation leap,and its on a very high clocking 14NM node. Intel 10NM seems to be able to hit around 4.8GHZ too.

The problem is now AMD has priced their 8 core at £430,and there is room for a more expensive XT version(the 5900XT and 5950XT already exist),what if Intel then manages to be faster in gaming than AMD with its own 8C CPU?? They will push the price up too.

Then AMD pushes out a slightly quicker 8C CPU,and then prices it a bit higher than Intel and so on.

AMD can do 12 and 16 cores,and thrash Intel,but if they are going to be priced so high it isn't going to matter for most DIY PC builders.

Is 6C going to be the new 4C for mainstream??

People have learnt nothing from what we saw Nvidia and AMD do with GPUs. We went from £450 being considered expensive for a top end GPU 10 years ago,to £1500~£2000 being the top end. It also meant the entry level and mainstream were pushed up.

Pricing wise, the tech companies such as AMD / Intel / Nvidia will no doubt have been watching Apple / Samsung etc pushing generational price increases with mobile phones and watching with interest as consumers still upgrade. While some costs increase over time in a business by the nature of inflation etc, I'm not sure those increases translate to a similar strategy of mobile phone manufacturers. Nvidia found this out with the 2xxx series by all accounts by harming their own potential sales, considering PC usage scenarios, people can hold off upgrading PC hardware if they need to quite easily as software such as games is generally developed (eventually) for what people actually have etc. While I know they'll always try to increase pricing over time, I think they may need to slow down those increases unless they want to deter consumers from the usual upgrade cycle of PC hardware. A large increase at the lower end of the scale will deter people from upgrading, if they had increased the cost by $10 instead of $50? I doubt people would really be complaining, it's an incremental change much more in line with inflation, over time they would have the price increase they want, but without the cost of losing a lot of consumers along the way.

14nm+++++(+?) for Rocketlake I think causes Intel issues with the power envelope compared to Zen 3 on 7nm (refined), yields are probably comparable. From what I've seen of Zen 3, it looks like the main change is that the cache is shared between all the cores instead of individual caches, that should be easier and cheaper to make and really just requires tweaking of how the cores use the cache. It doesn't require extra space hence sticking with the same socket. Rocketlake requires a whole new platform to use, something Intel needs to address moving forward in their chip and platform design if they really want to compete with how AMD have developed Ryzen which allowed for faster iteration and likely lower R&D costs.

Mainstream 6C? Interesting, maybe more 4C/6C lower end - 8C/10C mainstream - 12C/16C+ higher end. I guess it really depends on usage scenarios, not many games utilise all of the cores presented to them by the nature of the game engines themselves, AotS is an exception not the norm. AMD have definitely shaken up the market with respect to cores counts, the software has yet to catch up to that.

[DanceswithUnix]

From what I've seen of Zen 3, it looks like the main change is that the cache is shared between all the cores instead of individual caches, that should be easier and cheaper to make and really just requires tweaking of how the cores use the cache.

The change there is that they went from 4 cores directly communicating with each other to 8 cores. For a traditional crossbar AIUI that is 16 times the number of transistors required. The 6 core crossbar in the old Phenom 1100T was apparently stupidly large and added a lot to the part's cost, but that was a long time ago (on 45nm? Can't remember) so I guess they figured they would push the boat out again.

Look at all the ARM chips etc and they use clusters of 4 cores, it seems a natural choice for scaling. Apart from the Intel ring buffer parts, which have other problems at high core counts. If the downside is just die area, then it could be worth it for gaming price points.

The cache is a similarly interesting choice; large caches are slower. If that wasn't the case, there wouldn't be a need for L2 cache basically doing the same job as L1, you would just make the L1 big. So they have made L3 higher latency, and I guess clawed the extra clocks back on average with the lack of cross CCX fabric hops? That could suck for NUMA friendly workloads that didn't really go cross CCX, but probably be better for things like games.

The detailed analysis of this could be quite interesting when reviews actually come out.

[Iota]

The change there is that they went from 4 cores directly communicating with each other to 8 cores. For a traditional crossbar AIUI that is 16 times the number of transistors required. The 6 core crossbar in the old Phenom 1100T was apparently stupidly large and added a lot to the part's cost, but that was a long time ago (on 45nm? Can't remember) so I guess they figured they would push the boat out again.

Look at all the ARM chips etc and they use clusters of 4 cores, it seems a natural choice for scaling. Apart from the Intel ring buffer parts, which have other problems at high core counts. If the downside is just die area, then it could be worth it for gaming price points.

The cache is a similarly interesting choice; large caches are slower. If that wasn't the case, there wouldn't be a need for L2 cache basically doing the same job as L1, you would just make the L1 big. So they have made L3 higher latency, and I guess clawed the extra clocks back on average with the lack of cross CCX fabric hops? That could suck for NUMA friendly workloads that didn't really go cross CCX, but probably be better for things like games.

The detailed analysis of this could be quite interesting when reviews actually come out.

That's interesting to know, I wonder how they'll update Threadripper, considering that is designed for different workloads?

[DanceswithUnix]

That's interesting to know, I wonder how they'll update Threadripper, considering that is designed for different workloads?

You could say the die is cast, as they will have to use the same chiplets

I'm sure they will have done the server and workstation workload simulations when they were designing this, and better energy efficiency should mean they can clock higher on a given workload which should iron out any NUMA kinks and make this still the better part for things like virtual machine farms (where lots of CCXs shouldn't be a problem as they are all doing different things).

[CAT-THE-FIFTH]

Pricing wise, the tech companies such as AMD / Intel / Nvidia will no doubt have been watching Apple / Samsung etc pushing generational price increases with mobile phones and watching with interest as consumers still upgrade. While some costs increase over time in a business by the nature of inflation etc, I'm not sure those increases translate to a similar strategy of mobile phone manufacturers. Nvidia found this out with the 2xxx series by all accounts by harming their own potential sales, considering PC usage scenarios, people can hold off upgrading PC hardware if they need to quite easily as software such as games is generally developed (eventually) for what people actually have etc. While I know they'll always try to increase pricing over time, I think they may need to slow down those increases unless they want to deter consumers from the usual upgrade cycle of PC hardware. A large increase at the lower end of the scale will deter people from upgrading, if they had increased the cost by $10 instead of $50? I doubt people would really be complaining, it's an incremental change much more in line with inflation, over time they would have the price increase they want, but without the cost of losing a lot of consumers along the way.

14nm+++++(+?) for Rocketlake I think causes Intel issues with the power envelope compared to Zen 3 on 7nm (refined), yields are probably comparable. From what I've seen of Zen 3, it looks like the main change is that the cache is shared between all the cores instead of individual caches, that should be easier and cheaper to make and really just requires tweaking of how the cores use the cache. It doesn't require extra space hence sticking with the same socket. Rocketlake requires a whole new platform to use, something Intel needs to address moving forward in their chip and platform design if they really want to compete with how AMD have developed Ryzen which allowed for faster iteration and likely lower R&D costs.

Mainstream 6C? Interesting, maybe more 4C/6C lower end - 8C/10C mainstream - 12C/16C+ higher end. I guess it really depends on usage scenarios, not many games utilise all of the cores presented to them by the nature of the game engines themselves, AotS is an exception not the norm. AMD have definitely shaken up the market with respect to cores counts, the software has yet to catch up to that.

I think the pricing is opportunistic,unless AMD is trying a Jebait pricing move,or suddenly releases much lower priced non-X versions. Regarding Rocketlake,it will be a three generation IPC/core performance jump from Skylake. Also remember its only 8C unlike Cometlake which is 10C. Intel is integrating 4 Atom class cores,so I suspect it might improve efficiency in certain scenarios. It also introduces PCI-E 4.0 which negates a major platform advantage AMD has.

The issue here,is Intel hasn't got anything over 8C with Rocketlake S to compete with AFAIK. But if AMD is genuinely trying to push 8C CPUs upto between £400~£500 again,it means by extension if Rocketlake S has good enough performance,Intel is still OK upto that pricerange which is literally the bulk of consumer DIY and OEM prebuilt desktop sales.

Why I say 6C mainstream,is because AMD is trying to push 6C pricing in its own stack closer to £300,and unless the Ryzen 5 5600 is $200,its more likely to be $250,which is well over £200. Zen2 based CPUs will eventually be gone,and if AMD does not have a sub £200 6C Zen3 part ,then I am uncertain whether Intel will really want to push its own 6C CPUs,with Rocketlake S much below £200 now. For timebeing we have Zen2 and Cometlake 6C parts under £200,but have you also noticed the AMD 6C parts are now closer to £200,and Intel 6C parts are now cheaper. That means again 4C might be dominant under £200,just like 5 years ago. Yet the first 6C CPU under £300 was the Intel Core i7 5820K in 2015.

[mikahawkins]

In my opinion the Ryzen 7 5800 XT beats the i9 as well. AMD have done a great job these past few years.

[atemporal]

sure, providing the price is reasonable.. It is worrying the potential price side of things. I wonder if it's the problem of PC being seen as more niche now. It's no longer a growth market when you're trying to get everyone to buy-in. Has the market evolved into enthusiast?

[DanceswithUnix]

Has the market evolved into enthusiast?

Pretty much.

Most people just have phones and tablets. Some people and most companies buy laptops.

We buy desktops at work, but only for us engineers. When I was at Cisco there was no desktop option.

[mers]

Yikes , reading through all this it seems like a battle ground. I don't upgrade as much as I did in the older days ( started way back in 286 days ). My needs are reasonably modest, game very rarely and certainly don't need the best or fastest , I have more important things to waste money on. I'm more at home with a desktop and only use a lappy when I have to. When I do upgrade it's now more or less because I have spare cash to do so and actually fancy a new toy. All this discussion about price is irrelevent to me , I don't care what a company charges , they are out to make profits to survive and fund further R & D. I set myself a budget so what falls within it is what I'll buy. A few extra frame rates or a few extra gains in speed does not justify a massive increase in cash outlay in my book unless you get hooked up in bragging rights. Top tier products to me are for peeps with more money than sense.

[Iota]

Why I say 6C mainstream,is because AMD is trying to push 6C pricing in its own stack closer to £300,and unless the Ryzen 5 5600 is $200,its more likely to be $250,which is well over £200. Zen2 based CPUs will eventually be gone,and if AMD does not have a sub £200 6C Zen3 part ,then I am uncertain whether Intel will really want to push its own 6C CPUs,with Rocketlake S much below £200 now. For timebeing we have Zen2 and Cometlake 6C parts under £200,but have you also noticed the AMD 6C parts are now closer to £200,and Intel 6C parts are now cheaper. That means again 4C might be dominant under £200,just like 5 years ago. Yet the first 6C CPU under £300 was the Intel Core i7 5820K in 2015.

Looking at the launch pricing for the 3600 / 3600X they were $199/$249 respectively (IIRC they were £199/£239 in the UK at launch pricing), the launch pricing for the 5600X we know is $299 (£289) the 5600 non x part is as yet an unknown. The 2600 / 2600X launch pricing was $199/$229.

It's interesting the non x part remained the same between the 2600 and the 3600, whether that remains the same for an unannounced 5600 remains to be seen, I have my doubts. Meanwhile the 3600X was 8.7% more expensive than the 2600X at launch pricing, the 5600X is 20% more expensive than the 3600X at launch pricing, which is a pretty large disparity. If the same follows for a 5600, it should be priced about $239 (a 20% increase).

Looking at just Amazon prices though, the 2600 was sold new between £110-£174, the 2600X between £115-£235, the 3600 between £153-£199, the 3600X between £186-£240. So while I understand the concern over the pricing increases, especially the RRP launch price listings, I think people will be far more savvy to wait for the prices to be adjusted by the market. What's the saying? Something is only worth as much as someone is willing to pay for it? I definitely think that will be true for the 6C Zen3 CPU. I also agree that it looks like AMD is trying to push the pricing upwards for the 6C parts, I think that 20% is too much (although it does tie in nicely with an almost 20% IPC increase ), however if that is the case, then it should make more sense to stick with the 3600X if owned unless you really needed that ~20% increase in performance. Price / performance wise it would make more sense.

[Jayja]

Looking super forward to upgrading to the 5800x. Been on the same system for 6 years now, and with cyberpunk on the way. Im hyped for a new build! Just need that 3080

[DanceswithUnix]

If the same follows for a 5600, it should be priced about $239 (a 20% increase).

I saw it reported as a flat $50 across the range so far, though I didn't fact check that. The market for a 16 core would wear that, but that is quite a chunk at the lower end.

[Hoonigan]

There's not a lot that makes me want to upgrade from my 3800X, not much at all, but if I was in the market for a CPU right now then this news would be hugely exciting.
Interesting to see the gains in CS:GO!