Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[HurleyBird]

Yet in Cinebench R20 it's only 11%IPC over Zen2, so I wouldn't expect much gains in cinebench for Zen4 over Zen3 if AMD claims 8% IPC gain on average.

Application specific IPC deltas from prior generations are not predictive of future generations, obviously.

 

[tamz_msc]

Golden Cove shows strong performance in Cinebench because of its unified scheduler and larger L2. Larger L2 alone should give Zen 4 a sizeable boost in Cinebench, though not that it matters much for the end user in the end.

Analyzing Zen 2’s Cinebench R15 Lead

Cinebench R15 (CBR15) is a popular benchmark based on Cinema4D’s 3D rendering engine. It can utilize all available CPU threads, but here we’ll be analyzing it in single thread mode. In …

chipsandcheese.com

 

[biostud]

Personally I'm much more interested in gaming benchmarks than MT, and here the reason to upgrade to RPL over ADL doesn't seem to be very compelling. Sure a little speed bump, but probably nothing you're going to notice besides benchmarks.

 

[Hotrod2go]

Personally I'm much more interested in gaming benchmarks than MT, and here the reason to upgrade to RPL over ADL doesn't seem to be very compelling. Sure a little speed bump, but probably nothing you're going to notice besides benchmarks.

Depends on one's perspective & budget at the time when they are officially released for retail sale. Going from RL to RPL will be a noticeable improvement no doubt.

 

[biostud]

Depends on one's perspective & budget at the time when they are officially released for retail sale. Going from RL to RPL will be a noticeable improvement no doubt.

But that was already the case with ADL, and (depending on what you game and resolution) spending money on upgrading from RL to RPL for gaming purposes, would in most cases be better used on upgrading your video card.

 

[Hotrod2go]

But that was already the case with ADL, and (depending on what you game and resolution) spending money on upgrading from RL to RPL for gaming purposes, would in most cases be better used on upgrading your video card.

Not a good time to upgrade video cards until next gen from both camps comes out before year end in my view. Which funnily enough seems to coincide with next gen cpu as well.

 

[pakotlar]

Personally I'm much more interested in gaming benchmarks than MT, and here the reason to upgrade to RPL over ADL doesn't seem to be very compelling. Sure a little speed bump, but probably nothing you're going to notice besides benchmarks.

Yeah, neither non-X Zen 4 or RPL are for you.

 

[pakotlar]

Golden Cove shows strong performance in Cinebench because of its unified scheduler and larger L2. Larger L2 alone should give Zen 4 a sizeable boost in Cinebench, though not that it matters much for the end user in the end.

Analyzing Zen 2’s Cinebench R15 Lead

Cinebench R15 (CBR15) is a popular benchmark based on Cinema4D’s 3D rendering engine. It can utilize all available CPU threads, but here we’ll be analyzing it in single thread mode. In …

chipsandcheese.com

Thats about R15, now 2 Cinebench releases out of date.

 

[pakotlar]

Golden Cove shows strong performance in Cinebench because of its unified scheduler and larger L2. Larger L2 alone should give Zen 4 a sizeable boost in Cinebench, though not that it matters much for the end user in the end.

Analyzing Zen 2’s Cinebench R15 Lead

Cinebench R15 (CBR15) is a popular benchmark based on Cinema4D’s 3D rendering engine. It can utilize all available CPU threads, but here we’ll be analyzing it in single thread mode. In …

chipsandcheese.com

Thats about R15, now 2 Cinebench releases out of date. Do we have any indication on how R23 performs (the Cinebench Golden Cove is typically compared to Zen 3 on)?

Edit: Interesting article btw. One thing they mention is the large L3 improves IPC by avoiding the large latency spike by going to memory. This will, btw, also improve overall performance because Zen 3 incurs a huge power spike (large relative to L3 access) when going to main memory, so when accessing data from L3, Zen 3 can sustain higher clocks (all else held equal).

Alder Lake behavior is different. Its performance in L3 is abysmal bothin terms of latency and in terms of the power it takes to fetch a byte, compared to Zen 3. So for Alder Lake, and Raptor Lake, its very important to stay in L1 and L2. That’s why Raptor Lake’s doubled L2 is so important. We’re going to see better performance for latency sensitive applications that have many accesses that spill out of L2 currently on ADL, as well as lower power, and that benefit will be disproportionate to the benefit Zen 4 would get.

On this note, we should expect Raptor Lake and Zen 4 to converge in terms of power efficiency due to microarchitecture (not entirely of course), rather than diverge, because Zen 3 was already so darned efficient, and Alder Lake so poor (at the clocks needed to match Zen 3 performance). Regression to the mean. Zen 4 moving to TSMC 5nm will decrease the strength of convergence due to microarchitecture, unfortunately for Intel.

 

[biostud]

Yeah, neither non-X Zen 4 or RPL are for you.

I am planning to go for a 7800X or 7800X3D if it is launched at the same time or a couple of months later, combined with N32 or 4080.

 

[moinmoin]

Alder Lake behavior is different. Its performance in L3 is abysmal bothin terms of latency and in terms of the power it takes to fetch a byte, compared to Zen 3. So for Alder Lake, and Raptor Lake, its very important to stay in L1 and L2. That’s why Raptor Lake’s doubled L2 is so important. We’re going to see better performance for latency sensitive applications that have many accesses that spill out of L2 currently on ADL

Raptor Lake’s doubled L2 should offer a clear performance improvement independent of frequency. It's really curious that leaks so far show nothing of that so far. Any idea why that may be the case?

 

[nicalandia]

Raptor Lake’s doubled L2 should offer a clear performance improvement independent of frequency. It's really curious that leaks so far show nothing of that so far. Any idea why that may be the case?

Only the E cores are getting the L2 double. the P cores the L2 goes from 1.25 to 2 MiB. E cores don't have a saying on gaming/ST Performance

 

[moinmoin]

Only the E cores are getting the L2 double. the P cores the L2 goes from 1.25 to 2 MiB. E cores don't have a saying on gaming performance.

That's still a significant improvement ("only" 60% bigger). The only reasons I can think of are that either branch prediction is already working near perfectly, making the cache size increase not significant enough, or the existence of E cores prevents the scheduler from making the most of P cores.

 

[nicalandia]

That's still a significant improvement ("only" 60% bigger). The only reasons I can think of are that either branch prediction is already working near perfectly, making the cache size increase not significant enough

The 2 MiB per core was already taken into account when designing Golden Cove branch prediction system, they just segmented the core into client and server(1.25 MiB for client and 2 MiB for Server) and Golden Cove Branch Prediction system remains unchanged on Raptor Cove

 

[pakotlar]

Only the E cores are getting the L2 double. the P cores the L2 goes from 1.25 to 2 MiB. E cores don't have a saying on gaming/ST Performance

Yes, technically 1.25->2 isn’t 2x. It also doesn’t change the point; it’s a very large L2 increase, that will absolutely reduce the number of L2 misses and improve power efficiency, important for Golden/Raptor Cove because of their relatively poor L3 performance and power efficiency. Games that get a big uplift in Zen 3D may get a better than average boost in Raptor Lake. Or may not, who knows. Fingers crossed.

 

[nicalandia]

Yes, technically 1.25->2 isn’t 2x. It also doesn’t change the point; it’s a very large L2 increase, that will absolutely reduce the number of L2 misses and improve power efficiency

Ready my previous post. 2 MiB per core It has already been taken into account while designing Golden Cove/Raptor Cove.

 

[pakotlar]

Raptor Lake’s doubled L2 should offer a clear performance improvement independent of frequency. It's really curious that leaks so far show nothing of that so far. Any idea why that may be the case?

I don’t know, too many unknowns. If that ES3 part was really boosting to 5.7 it sure looks like the cache did not help Golden Cove achieve better than linear speedup with frequency. But keep in mind that performance doesn’t necessarily scale with clock speed, so even achieving linear increase may be evidence of cache benefit (for memory-bound applications). If 5.5ghz was more typical, we are seeing a 5% average boost due to stuff besides clock speed (L2, ring bus frequency, other?). It’s also possible that none of the tested applications are really all that memory bound on Golden Cove.

What do you think?

 

[IntelUser2000]

Raptor Lake’s doubled L2 should offer a clear performance improvement independent of frequency. It's really curious that leaks so far show nothing of that so far. Any idea why that may be the case?

Man the leakers were talking about a feature GC that was disabled. It's increasingly looking like it won't benefit general purpose code much. Whatever it is it's getting big gains on decompression for 7-zip where it's significantly behind now. So targetted optimizations to ratify weaknesses?

Keep in mind as cache sizes get larger increasing amount of applications will stop benefitting from it as it fits entirely in the cache.

One that will never reach that point is games.

Although for low single digit differences things such as internal firmware and prefetching mechanisms will affect the result. Just wait for reviews.

 

[mikk]

A bigger cache might benefit gaming workloads, it's expected that non gaming applications won't really benefit, maybe expect for 7zip/winrar and the likes. If there are small IPC gains like 1-2% they will be hard to spot on pre final hardware/software.

 

[HurleyBird]

A bigger cache might benefit gaming workloads

Same with a faster cache. Hopefully Intel finds the right balance.

 

[IntelUser2000]

Same with a faster cache. Hopefully Intel finds the right balance.

Fast as in latency or bandwidth? Latency is affected by CPU architecture, cell choice and of course target frequency. You can use 8T cells that are lot larger but faster and lower power. And within those high level parameters you can further tweak the balance. In terms of power there's a compromise between static power and dynamic power. CPU architecture itself can affect cache latencies.

For applications most are bound by latency not bandwidth. AVX code needs bandwidth which is why L/S units are beefed up since it directly correlates to L1 cache bandwidth. Latency being the more important feature since most don't need the bandwidth. Less so for capacity vs latency. In general code it seems even a cycle extra can be detrmental.

Wouldn't be surprised if RPL cache latencices didn't increase since it's already baked in with Willow Cove.

A bigger cache might benefit gaming workloads, it's expected that non gaming applications won't really benefit, maybe expect for 7zip/winrar and the likes. If there are small IPC gains like 1-2% they will be hard to spot on pre final hardware/software.

The magnitude of the gains suggest that for 7-zip, Raptor Cove changes are contributing to the performance not just memory level changes. It makes sense as engineers would logically target weaknesses first. Hence why eventually all perform roughly equal despite implementation differences since people always compare to existing and don't like ups and downs.

 

[nicalandia]

A bigger cache might benefit gaming workloads, it's expected that non gaming applications won't really benefit, maybe expect for 7zip/winrar and the likes. If there are small IPC gains like 1-2% they will be hard to spot on pre final hardware/software.

It's about 5% performance boost in gaming than 12900K, But it consumes 53% more power than the 12900K(in Red Dead Redemption 2)

Intel Core i9-13900K Raptor Lake CPU Gaming & Synthetic Performance Benchmarks Leaked, 5% Faster Than Core i9-12900K On Average

The first gaming and synthetic performance benchmarks of Intel's Core i9-13900K Raptor Lake 5.5 GHz CPU have been leaked.

wccftech.com

Both CPUs were tested on the same DDR5 6400 system



The 5800X3D is about 5% Higher FPS than a 12900KS with DDR5 6400 Mhz RAM or about 15% Higher than the 12900K with DDR4 RAM

 

[FangBLade]

So Raptor Lake will struggle against 5800x3d in many games that benefits that extra cache, i'm not impressed.

 

[nicalandia]

So Raptor Lake will struggle against 5800x3d in many games that benefits that extra cache, i'm not impressed.

It's Alder lake vs X3D again. It will lose to 5800X3D by a large margin where 12900KS lost before and it will win on games that Alder Lake won.


Based on this chart. It will beat 5800X3D within 1-5% on titles in red square, match it or barely lose to the ones in yellow square and most likely lose on the top green





The All Resolution Average is 4.83%

 

[Asterox]

A little more gaming oriented test.

Intel Core i9-13900K engineering sample tested in games shows improvement in minimum framerates - VideoCardz.com

Intel Raptor Lake faster in gaming ExtremePlayer shares more benchmark data on Intel’s upcoming Raptor Lake processor. Hardware reviewer recently acquired a sample of Intel 13th Gen Core CPU. This processor featuring 24-cores and 32-threads with unlocked power limits, which basically means it’s...

videocardz.com