First came a lake, then a dinosaur, then a meteor - Intel Raptor Lake is rumored to launch by the end of September

[Brooksie359]

On 2/21/2022 at 12:39 PM, IkeaGnome said:

Meteor, then Arrow. Seems it's the end of times Horizon Zero Dawn style.

Honestly I am wondering when they will run out of lakes. We have had lakes for so long that I am wondering if they will slap lake onto every architecture they make. 

[IkeaGnome]

Just now, Brooksie359 said:

Honestly I am wondering when they will run out of lakes. We have had lakes for so long that I am wondering if they will slap lake onto every architecture they make. 

If we are just using Alaska, roughly 3,197 years from now. Minnesota surely can add some time to that.

[porina]

1 hour ago, KaitouX said:

The issue with this strategy is Intel non-K SKUs, AMD needs to beat those soundly if they want to ask for premium prices, but that might be pretty hard to do.

AMD only need to be competitive against what will be the Intel CPUs of that generation, in perf and price. Intel non-K are cheaper but not as fast, and will go more against AMD APUs. So it still kinda works out.

 

1 hour ago, KaitouX said:

Please correct me if I'm wrong, but isn't high-end DDR4 faster than current DDR5 at most tasks? I thought that DDR5 only was better at some extremely niche tasks currently.

It depends on what you compare. Standard DDR5 will be faster than standard DDR4. But we now have a lot of XMP DDR4 far beyond standard. DDR5 is still somewhat young so faster XMP DDR5 isn't so plentiful yet.

 

For workloads I care about, which fall outside mainstream coverage, effective ram bandwidth is key. I'd expect DDR5 to do well there. Where DDR5 might fall a bit behind is latency sensitive uses, where high end DDR4 can still lead. 

 

1 hour ago, KaitouX said:

While I understand that there are people that will go DDR5 regardless, if you want the best performance in the high-end DDR4 might still be the better option.

As always it depends on your use case. I wanted an Alder Lake system to test for a while, but only just now managed to find (relatively) cheap DDR5 to enable that.

[KaitouX]

50 minutes ago, porina said:

AMD only need to be competitive against what will be the Intel CPUs of that generation, in perf and price. Intel non-K are cheaper but not as fast, and will go more against AMD APUs. So it still kinda works out.

 

It depends on what you compare. Standard DDR5 will be faster than standard DDR4. But we now have a lot of XMP DDR4 far beyond standard. DDR5 is still somewhat young so faster XMP DDR5 isn't so plentiful yet.

 

For workloads I care about, which fall outside mainstream coverage, effective ram bandwidth is key. I'd expect DDR5 to do well there. Where DDR5 might fall a bit behind is latency sensitive uses, where high end DDR4 can still lead. 

 

As always it depends on your use case. I wanted an Alder Lake system to test for a while, but only just now managed to find (relatively) cheap DDR5 to enable that.

That's what I mean by that, if my guess is accurate, and AMD doesn't change anything compared to their current lineup, Zen 4 parts would have similar performance to Intel non-K SKUs(~5% below the K ones), if Intel adds E cores to more non-K SKUs it would look particularly bad to AMD, as single core performance would be somewhat similar, but multicore would be a significant win to Intel, and if AMD keep current prices, Intel non-K SKUs would be significantly cheaper too with the DDR4 prices lowering further the cost of the Intel platform.

Basically I don't think Zen 4 without big changes to the lineup(core count and/or price) would make any sense below the 5950X successor, as it would be too expensive compared to their performance equivalents(including platform prices), or too slow compared to similarly priced options.

 

Yeah I was comparing the best DDR4 vs best DDR5 currently, and with the exception of some niche workloads that favor mainly bandwidth, DDR4 is either better or similar for half the price.

[Deadpool2onBlu-Ray]

16 hours ago, KaitouX said:

That's what I mean by that, if my guess is accurate, and AMD doesn't change anything compared to their current lineup, Zen 4 parts would have similar performance to Intel non-K SKUs(~5% below the K ones), if Intel adds E cores to more non-K SKUs it would look particularly bad to AMD, as single core performance would be somewhat similar, but multicore would be a significant win to Intel, and if AMD keep current prices, Intel non-K SKUs would be significantly cheaper too with the DDR4 prices lowering further the cost of the Intel platform.

Basically I don't think Zen 4 without big changes to the lineup(core count and/or price) would make any sense below the 5950X successor, as it would be too expensive compared to their performance equivalents(including platform prices), or too slow compared to similarly priced options.

 

Yeah I was comparing the best DDR4 vs best DDR5 currently, and with the exception of some niche workloads that favor mainly bandwidth, DDR4 is either better or similar for half the price.

Wait intel is upping core count AGAIN? If that is the case, I might skip Zen and just pop in a 13700k or whatever into my board

[KaitouX]

6 hours ago, Ryan829 said:

Wait intel is upping core count AGAIN? If that is the case, I might skip Zen and just pop in a 13700k or whatever into my board

Doubling the E-core count.

https://videocardz.com/newz/intel-showcases-13th-gen-core-raptor-lake-cpu-with-24-cores-and-32-threads

[LAwLz]

9 hours ago, KaitouX said:

Doubling the E-core count.

It's worth noting that Intel's E cores offers roughly 60% the performance of a P core, assuming no throttling. However, P cores has to throttle more than E cores when doing long multi-core workloads. So for a ~20 thread workloads 1 E core is worth somewhere in the neighbourhood of 0.7 P cores.

 

Very roughly speaking, the current i5 with 6+4 cores performs like an "8.8 P core chip".

This rumoured 6+8 core i5 will perform like a "11.6 P core chip".

 

The doubling of E cores in the i5 and i7 very roughly translates to about 2.8 P cores worth of extra performance, for very highly threaded workloads.

In the i9 the same math works out to be about 5.6 extra P cores worth of performance.

 

 

It would be a very substantial generational upgrade.

 

 

[Deadpool2onBlu-Ray]

5 hours ago, LAwLz said:

It's worth noting that Intel's E cores offers roughly 60% the performance of a P core, assuming no throttling. However, P cores has to throttle more than E cores when doing long multi-core workloads. So for a ~20 thread workloads 1 E core is worth somewhere in the neighbourhood of 0.7 P cores.

 

Very roughly speaking, the current i5 with 6+4 cores performs like an "8.8 P core chip".

This rumoured 6+8 core i5 will perform like a "11.6 P core chip".

 

The doubling of E cores in the i5 and i7 very roughly translates to about 2.8 P cores worth of extra performance, for very highly threaded workloads.

In the i9 the same math works out to be about 5.6 extra P cores worth of performance.

 

 

It would be a very substantial generational upgrade.

 

 

That would be indeed.