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.

 

[A///]

I do not know how these things can be in mobile systems.

Gamer laptops or those workstation laptops, the set it down and forget it kinds. They're aimed for units that'll be plugged up 24/7.

 

[Carfax83]

I bought that A Plague's Tale Requiem tale. If any 7950x owner has the game and wants to test it out in that problem area let me know if you want a comparison.

 

[Det0x]

There might be some truth to this. Raptor Lake's cache bandwidth is significantly better than Zen 4's, but Zen 4's cache latency is significantly better than Raptor Lake's. But since Raptor Lake has more OoO resources, it can presumably hide that latency better.

Case in point here is the 13700K Aida64. Keep in mind this benchmark scales with core count, so the 13700K and the 7700x are both 8 core CPUs and they are using the same RAM, but the Zen 4 has CR1 which should perform a bit better.

And now the 7700x:

As I said, Raptor Cove has much greater bandwidth going through the core than Zen 4, in addition to memory bandwidth.

Why stop at 7700x ?
7600x with its lower results fits even better with this cherrypicking agenda.. Here let me find the worst result i can find:


The important thing is not to show the 7950x i take it?

16core cpu vs 16cores cpu



below results is @AdamK47's results as i dont want to dig up my screenshot again.


Looking at the numbers above, I find it alittle amusing you are trying to convince us raptor Lake's cache bandwidth is significantly better than Zen4..

 

[coercitiv]

It's still impressive though that FG can deliver such high FPS when the graphics card is clearly limited by the CPU.

They're not "real" or "interactive" frames, as you probably already know. That means they're not suitable in a scenario where you're already hitting 30FPS without FG, because the resulting gameplay will have latency that is inferior to 30FPS.

I bought that A Plague's Tale Requiem tale. If any 7950x owner has the game and wants to test it out in that problem area let me know if you want a comparison.

Underclock your CPU and test in one of those CPU limited areas until you reach a 30 to 40 FPS average. Enable Nvidia Reflex if directly available in menus. Then enable DLSS while keeping FG disabled. Then enable FG. Compare the FPS improvement to the latency hit, if any perceived.

 

[Carfax83]

Why stop at 7700x ?
7600x with its lower results fits even better with this cherrypicking agenda.. Here let me find the worst result i can find:

The important thing is not to show the 7950x i take it?

Damn, people are so cynical these days, always assuming the worst! I just wanted to make it as apples to apples as possible.

Efficiency cores are obviously not the same as performance cores so a 8+8 Core i7 13700K is not the same as having 16 big cores like the 7950x. I wasn't even sure the efficiency cores were being activated during the cache and memory benchmark but I checked and verified that it was. Not to the same degree as the P cores though.

I also ran the Aida64 cache and memory benchmark with the E cores disabled, and it impacted the cache scores negatively. I mean, BIG TIME NEGATIVE!

I was actually surprised at that but I figured the E cores are more tightly integrated in the cache hierarchy for the Raptor Cove cores that disabling them affects the cache bandwidth for the entire CPU.

That said, the stock 13700K which is obviously being handicapped by the E cores in this benchmark still manage to beat AdamK47's Aida64 cache scores in the L1 write and copy and comes close in the L3 read. The E cores are obviously affecting the benchmark because the L2 and L3 write and copy are much lower than the read and I think it has to do with the hybrid architecture as the E cores are active during the benchmark as I previously stated.

Here's mine (underclocked) for comparison:

Looking at the numbers above, I find it alittle amusing you are trying to convince us raptor Lake's cache bandwidth is significantly better than Zen4..

But it is in certain ways. Raptor Cove can do 3x 256 bit loads per cycle while Zen 4 can only do 2. To achieve that, Raptor Lake requires enormous L1 cache bandwidth and you can see that because the write and especially copy scores for the L1 are substantially higher than the 7950x despite being handicapped by the efficiency cores.

Chips and Cheese goes into that a bit in his article about Zen 4's memory subsystem. He doesn't compare it to Raptor Lake but to Alder Lake which is close enough but not the same. If you want to read it, check it out here.

 

[igor_kavinski]

I wish Det0x worked for AMD so we could all have a massively tuned system like him at stock settings

 

[Carfax83]

They're not "real" or "interactive" frames, as you probably already know. That means they're not suitable in a scenario where you're already hitting 30FPS without FG, because the resulting gameplay will have latency that is inferior to 30FPS.

I messed around with it a bit and I honestly couldn't tell the difference between having FG on or off. But I had Reflex turned on though so maybe I will disable that next time.

Now as for that whole shindig about CapFrameX's benchmark for the game on Zen 4, after playing around with the game, I think the game may have some occlusion culling issues. I also noticed that despite being DX12, it still likes to keep one core pegged. The FPS drops in the Village area were the result of reduced GPU activity, but the CPU was obviously not bottlenecking the GPU so that's why I think it has to do with bad occlusion culling in this particular area as the area is very dense and the developers probably wanted to increase FPS by not having the graphics card render things the camera couldn't see and screwed it up.

I took some screenshots so people can see. I had to disable HDR otherwise the image looked completely messed up, but if anyone does end up buying this game, make sure HDR is enabled when you play it as the implementation is VERY GOOD. It

This is at 4K with DLSS quality and maxed ultra settings. As you can see, the GPU usage is at 70% for some reason while the CPU usage is nowhere close to being maxed out so it's obviously not being CPU bottlenecked.

This next scene has much higher GPU usage and thus much higher FPS. All I did was change the vantage point. That's why I think there might be some occlusion culling issues because vantage point seems to affect framerate and GPU usage in this area more than it should.

 

[igor_kavinski]

@Carfax83

I really enjoyed this post. Your commentary is GOOD, better than the commentary of a lot of youtube wannabe reviewers. I hope you can do this on the side on an actual games/PC hardware website.

Maybe get in touch with Ryan?

 

[Carfax83]

@Carfax83

I really enjoyed this post. Your commentary is GOOD, better than the commentary of a lot of youtube wannabe reviewers. I hope you can do this on the side on an actual games/PC hardware website.

Maybe get in touch with Ryan?

I have to admit that I do love talking about games from a technological perspective, but benchmarking is boring as hell to me so I could never do it.

And I know this isn't a game review thread, but man, this game is actually pretty damn good. I never played the first game and I didn't expect to like it at all because I hate stealth mechanics in games and I bought it purely out of technical curiosity because this is a true next gen title without any influence from the previous generation of consoles. So far I've been enjoying the story in particular but even the gameplay is good and not boring like I thought it would be.

From a tech perspective, it's very impressive with a particular emphasis on the fact that you have to look really close to spot any asset streaming or pop in. That's probably the main reason why it doesn't run at 60 FPS on the XSX and PS5, because the game is rendering quite a bit due to an extended draw distance at high LoDs. The HDR implementation is among the very best I've seen. Once you go HDR in this title, there's no going back provided your HDR TV/monitor is capable. All in all, it seems to be fairly well optimized with the exception of the area I spoke about earlier which I'm almost certain is occlusion culling errors. It's heavily GPU bound for the most part, but when the rats show up, the CPU usage goes up significantly. Still no real impact to framerate on my rig though and I still stay in triple digit territory.

 

[DAPUNISHER]

How is it bias just to post it? I considered buying the game myself so I watched some reviews, and many of them mentioned performance drops in the same area that CapFrameX did, namely the town and whenever the rats were on screen. Skill Up mentioned it in his review that his 3700x and 2080 Ti was in the 20s in terms of FPS at 4K in a non busy area, and he had to drop the rez down to 2K medium settings and DLSS when in the town and it still didn't get good FPS. So he ended up lowering it to 1080p with DLSS and medium settings and the FPS still wasn't satisfactory. That to me indicates a CPU bottleneck but he didn't specifically mention it. He then tested it with an RTX 4090 and a 3950x, but with FG turned on and then his FPS went through the roof! I wish he would have tested with FG turned off then I would imagine his FPS would have plummeted like a rock due to being CPU bottlenecked.

Something is seriously amiss with his setup. A stock 5700G in a miniPC with a RTX3050 on a external GPU dock PCIE 3.0, at the heavily CPU bottlenecked settings of 1080 low with DLSS Q gets over 60fps and plays smoothly.

There is no way a 7950X is not crushing FPS in this game. A 13900K being faster is readily believable though as well.

 

[eek2121]

Damn, people are so cynical these days, always assuming the worst! I just wanted to make it as apples to apples as possible.

Efficiency cores are obviously not the same as performance cores so a 8+8 Core i7 13700K is not the same as having 16 big cores like the 7950x. I wasn't even sure the efficiency cores were being activated during the cache and memory benchmark but I checked and verified that it was. Not to the same degree as the P cores though.

I also ran the Aida64 cache and memory benchmark with the E cores disabled, and it impacted the cache scores negatively. I mean, BIG TIME NEGATIVE!

I was actually surprised at that but I figured the E cores are more tightly integrated in the cache hierarchy for the Raptor Cove cores that disabling them affects the cache bandwidth for the entire CPU.

That said, the stock 13700K which is obviously being handicapped by the E cores in this benchmark still manage to beat AdamK47's Aida64 cache scores in the L1 write and copy and comes close in the L3 read. The E cores are obviously affecting the benchmark because the L2 and L3 write and copy are much lower than the read and I think it has to do with the hybrid architecture as the E cores are active during the benchmark as I previously stated.

Here's mine (underclocked) for comparison:

But it is in certain ways. Raptor Cove can do 3x 256 bit loads per cycle while Zen 4 can only do 2. To achieve that, Raptor Lake requires enormous L1 cache bandwidth and you can see that because the write and especially copy scores for the L1 are substantially higher than the 7950x despite being handicapped by the efficiency cores.

Chips and Cheese goes into that a bit in his article about Zen 4's memory subsystem. He doesn't compare it to Raptor Lake but to Alder Lake which is close enough but not the same. If you want to read it, check it out here.

You guys shouldn’t be using aida64 to benchmark cache and memory, there is another, much more useful benchmark to do this linked elsewhere. I will try and link it here later if nobody else does it, I have an appointment to get to.

 

[Carfax83]

There is no way a 7950X is not crushing FPS in this game. A 13900K being faster is readily believable though as well.

If you read my post above, I did some testing on the game and it appears it has some occlusion culling issues in the village area that can cause GPU usage to drop so I think that's what caused it.

 

[Carfax83]

You guys shouldn’t be using aida64 to benchmark cache and memory, there is another, much more useful benchmark to do this linked elsewhere. I will try and link it here later if nobody else does it, I have an appointment to get to.

I'm very interested in this other benchmark, do tell! I think memory benchmarks are the only benchmarks I actually like to run because it reminds me of my first Pentium 4 with RDRAM rig. Back in the day, P4 with RDRAM was a bandwidth monster, then DDR came on the scene and took over with similar bandwidth and much better latency.

 

[DAPUNISHER]

If you read my post above, I did some testing on the game and it appears it has some occlusion culling issues in the village area that can cause GPU usage to drop so I think that's what caused it.

That seems unlikely, though certainly not impossible. It is far more likely it is CPU limited due to poor optimization. Either issue should be resolvable via driver and/or game updates.

 

[igor_kavinski]

I think memory benchmarks are the only benchmarks I actually like to run because it reminds me of my first Pentium 4 with RDRAM rig. Back in the day, P4 with RDRAM was a bandwidth monster, then DDR came on the scene and took over with similar bandwidth and much better latency.

PS3 used XDR, a successor to RDRAM tech that solved the latency issues. If Rambus hadn't shot itself in the foot with their greed and listening to the stupidity of their legal department, XDR+++ could have been a great and exciting memory tech option for multicore monsters like 7950X / Genoa / Sapphire Rapids and 13900K.

 

[eek2121]

I'm very interested in this other benchmark, do tell! I think memory benchmarks are the only benchmarks I actually like to run because it reminds me of my first Pentium 4 with RDRAM rig. Back in the day, P4 with RDRAM was a bandwidth monster, then DDR came on the scene and took over with similar bandwidth and much better latency.

Here you go:

New Memory Bandwidth & Latency Competition

1) Download the benchmark here: MicrobenchmarkGui.zip 2) Extract files to a local directory 3) Run the program (MicrobenchmarkGui.exe): 4) Disregard SmartScreen filter and run it anyway 5) For bandwidth benchmark, be sure to max out your thread count. Limited to 64 threads, FYI. So for...

www.overclock.net

 

[AdamK47]

You guys shouldn’t be using aida64 to benchmark cache and memory, there is another, much more useful benchmark to do this linked elsewhere. I will try and link it here later if nobody else does it, I have an appointment to get to.

AIDA64 is only useful when comparing same system measurements after performing system tweaks. The benchmark is highly suspect. I firmly believe the developers make up a lot of the scores through meaningless algorithms. Not true system bandwidth being represented. For example, simply changing the FSB while having the memory speed stay exactly the same, produces impossible numbers. It's a garbage benchmark that really needs to be replaced by another company/individual that can do it better. Sadly, it's really the only easily accessible tool out there.

 

[AdamK47]

I'm very interested in this other benchmark, do tell! I think memory benchmarks are the only benchmarks I actually like to run because it reminds me of my first Pentium 4 with RDRAM rig. Back in the day, P4 with RDRAM was a bandwidth monster, then DDR came on the scene and took over with similar bandwidth and much better latency.

And much better pricing for DDR. Rambus was a sleezy company, so there was that.

 

[Carfax83]

That seems unlikely, though certainly not impossible. It is far more likely it is CPU limited due to poor optimization. Either issue should be resolvable via driver and/or game updates.

I initially thought it was due to inefficient multithreading, but the CPU was being utilized across more than one thread and wasn't close to being maxed on any single thread. And also it only occurs in this one particular area and nowhere else have I seen this behavior. That's why I think it's more likely to be an occlusion culling error.

 

[Carfax83]

I appreciate that software guys, here is my read bandwidth and latency test. One thing I noticed so far compared to Aida64 is that it doesn't test the efficiency cores. It only sticks to the performance cores which is good for Alder/Raptor Lake as testing the efficiency cores brings the scores down:

 

[DrMrLordX]

You guys shouldn’t be using aida64 to benchmark cache and memory, there is another, much more useful benchmark to do this linked elsewhere.

AIDA64's memory latency and bandwidth measurements are only relevant to one size of dataset. For spitting out quick numbers to compare to other similar systems, it's suitable. It shouldn't be taken as gospel truth for every situation, though.

 

[phillyman36]

• 2,967

Can someone tell me the benefit of the 8 extra E cores on the 13900k vs 13700k? Are all 16 E cores used for system resources and background tasks when gaming? What apps will actually use all the cores? I use i dont think Handbrake uses more than 6 to 8 cores so at what point are they help in non benchmarking use? Is there a list of software some can tell me the 8 extra cores are utilized? Real world usage.

 

[igor_kavinski]

Impact of E-cores on gaming performance:

RTX 4090 & 53 Games: Core i9-13900K E-Cores Enabled vs Disabled Review

In this week's TPU50 Megabench we're testing whether you can unlock additional gaming performance by disabling the E-Cores on a Raptor Lake Core i9-13900K processor. For our benchmarks we used the mighty GeForce RTX 4090, with 53 games at three resolutions.

www.techpowerup.com

1080p



1440p



2160p



Good to see that most of the games are not impacted by the presence of E-cores.

Worst case scenario is Prey at 11% performance loss. All other impacted games were less than 10% slower.

 

[nicalandia]

• 2,967

Can someone tell me the benefit of the 8 extra E cores on the 13900k vs 13700k? Are all 16 E cores used for system resources and background tasks when gaming?

The 13900K are for power users and people that use their computer to get things done faster. But... Also for bragging rights and pissing contests. Very few people use their 13900K and 7950X To their fullest. Most are just people with more money than sense.

Smart and Savvy buyers get a 13600K, OC their Chip to match the single core boost of the 13900K, invest the money saved and put it on a better GPU.

Someone with a 13600K/7700X CPU and a 3080/6800XT and a 1440P monitor are set for the next few years.. People with 13900K and a 4090, 1500 Watt PSU and a 4K monitor and don't use their system to their fullest have more money than sense.

 

[ondma]

Impact of E-cores on gaming performance:

RTX 4090 & 53 Games: Core i9-13900K E-Cores Enabled vs Disabled Review

In this week's TPU50 Megabench we're testing whether you can unlock additional gaming performance by disabling the E-Cores on a Raptor Lake Core i9-13900K processor. For our benchmarks we used the mighty GeForce RTX 4090, with 53 games at three resolutions.

www.techpowerup.com

1080p

View attachment 71281

1440p

View attachment 71282

2160p

View attachment 71283

Good to see that most of the games are not impacted by the presence of E-cores.

Worst case scenario is Prey at 11% performance loss. All other impacted games were less than 10% slower.

That is not exactly the whole story. Most games were the same or slightly faster with the e cores enabled. It is just that the four games that were faster with e cores disabled were faster by a quite large margin. The average difference over the entire suite of games is with within the margin of error for either condition.