5800x3d vs raptor lake in DCS

[skypickle]

So the 96 mb of cache in MSFS 2020 really allows the 5800x3d to post a higher fps (although the 1% lows=stutter) are lower than a 13900k. However, when it comes to cache the 13900K has 68 MB of L2+L3 (https://videocardz.com/newz/intel-13th-gen-core-raptor-lake-desktop-cpu-to-feature-up-to-68mb-of-l2-l3-cache) So is a 5800x3D still better than a 13900K for DCS? If cache is that critical for loading the large blocks of scenery quickly, then the lower cache raptor lake chips (13700K, 13600K) should be worse than a 13900k (at the same clocks)-but I haven't seen any data-even anecdotal (prob because no one would spend $300 more for a cpu and not use it to its max)

Is there a meta analysis someone might have seen and could share the data to answer this question?

[skypickle]

I saw this and it just looks at AAA titles(multicore optimized).

 

 

[TKu]

DCS is not MSFS: it has a completly different architecture.

Single core: I assume 13700k has in general a faster single core performance than the x3d. Maybe it us close in DCS Performance. Single core overclocking can push the limits for DCS on intel once again.

All-core useage: intel will outperform coz it is newer. (Beats threadrippper as well)

(My) Conclusion: I‘ve got 13700k/4090/VR Aero and I can tell you that the 4090 does the workload. So if cpu's performance is close, you would also have the fastest available GPU as well to be happy. There is no term of saving money or having the big advantage one over another. It must be fast enough to fire the GPU.

Today i would go for intel 13k. if there is no pressure, wait for the next gen x3d and compare it again in a few months.

Edited November 22, 2022 by TKu

[Mac D]

6 hours ago, skypickle said:

I saw this and it just looks at AAA titles(multicore optimized).

 

 

Just a bit of a warning : watch out with these kinda "benchmark" channels, a lot of them are fake. 

(And yes, I fell for it too)

Stick with known reviewers like Der8auer, Gamers Nexus , hardware unboxing etc 

Edited November 22, 2022 by Mac D

[skypickle]

@Mac DYes they are notorious. Framechasers comes to mind

 

@TKu Yes I know the architectures are different. And yes the 13700k has a big jump on the x3d in clock but the cache enables the chip to perform better because of the large segments that need to be loaded. The effect of cache in DCS is not clear. I have not seen a direct comparison of the x3d with raptor lake in dcs. I wish some of the people here would post their fps in a standard instant mission in the same aircraft whilst flying close to the ground in the same map. 

[TKu]

Let me know. I'll be available to share my 13gen/32gb 3600 cl16 ddr4/4090 results against 5800x3d.

But as said, I expect only a few % difference.

 

Edit:

Setup: High settings, heat blur off, instant action, marianas, ah-64 start on runway miz, in VR (Varjo Aero OpenVr 70%) and 2d screen 1920x1080. Reshade 5.5 level correction+ lumasharpen on. 

Added Screenshots from the engine using the given settings above. I know it is not very accurate, but I had only a few minutes so far.

Edited November 22, 2022 by TKu

[Thinder]

4 hours ago, skypickle said:

@TKu Yes I know the architectures are different. And yes the 13700k has a big jump on the x3d in clock but the cache enables the chip to perform better because of the large segments that need to be loaded.

That's the whole idea. But you won't take advantage of it with the wrong RAM kit.

And btw, this video is not the only one getting it wrong when it comes to RAM or CPU tests (like using RAM frequencies that doesn't result in a 1:1 ratio), you want real info and not commercially biased ones from youtube, ask manufacturer techies...

Here is what you can expect from bounding the 5800x3d with a proper RAM kit (B-die, 4 X 1 bank, 3200 or 3600 GHz), not a high street Cl16 non B-die (be it from G.Skill or Corsair, they aren't better than a Crucial Ballistix) which will never achieve those results simply because the RAM controller will not have what it needs for it and they haven't changed from the 5600X.

Test was conducted at 4K 2 X MSAA (3D Mark Pro) vs my old 5600X with B-die kit and optimum timings in BIOS, from this result, the only difference is 3200MHz vs 3600MHz.

The CPU kept all channels open and storing the data to the cache made the GPU work a lot easier, for the story, I used the same EVGA GEFORCE GTX 1080 Ti which was boosted like the 5600X (MSI Afterburner and AMD Ryzen Master) with the 5600X rig but not boosted for this last test.

In short, there is bottleneck CPU-RAM which if you remove it with a B.die kit, let you beneficiate from the full potential of your  Ryzen 3/3D, if not, expect a loss of up to 1.33% in Graphic score, 6.04% in Physics score and 1.35% in Combined score, that's the gains I got at 4K 2 X MSAA from a Cl16 high street RAM kit to a B.die 4 X 1 rank, 3200HGz kit with my 5600X. With this CPU using a high street RAM kit might well result on much more loss of performances.

My tip is to spend the dosh in a proper RAM and avoid high street RAM kits, even if they are branded G.skill or Corsair, especially with this CPU, 18.91% gain in Graphics score at 4K 2 X MSAA is more than triple the average gain of a water cooler for your GPU, and you'll retain your warranty.

B-Die Finder  Find Samsung B-Die DDR 4 memory kits on Amazon, Newegg and many more.

AMD Zen 3/3D controllers limits are: 3200GHZ (unless you use B-die with much larger timing range), 4 banks.

Optimum is: Same as above + 4 X 1 bank (interleaving). All infos supported by AMD and GSkill technical support.

Edited November 22, 2022 by Thinder

[skypickle]

 

@TKu thank you for posting your setup and your results.i see a really low frametime , congrats. In the second window, cpu at 30% and simulation at 24%. I guess that means you are gpu bound?

 

@Thinder I know about the argument regarding RAM. However, I am speaking about cache. If you are suggesting that different ram kits will influence framerate more significantly than cache in DCS, that is part of what I am trying to parse. By comparing utilization and fps (as TKu posted above in the same mission - Marianas runway start in an Apache), I am hoping to see cache impact. What are your results Thinder? Unfortunately I am not in your league-10700k clocked to 4.5, 1080Ti, 32 gb Bdie, z490 Hero. Since the 7700x3D is rumored to have higher clocks and more cache, I might wait for that. But yes unfortunately due to their architecture and mem controller, DDR5 wont go much above 6000. Still,the 13900k has 96 mb of L2 and L3 cache as well as significantly higher clocks so I dont understand why it doesnt spank the 5800x3d

[Thinder]

35 minutes ago, skypickle said:

@Thinder I know about the argument regarding RAM. However, I am speaking about cache. If you are suggesting that different ram kits will influence framerate more significantly than cache in DCS, that is part of what I am trying to parse. By comparing utilization and fps (as TKu posted above in the same mission - Marianas runway start in an Apache), I am hoping to see cache impact. What are your results Thinder? Unfortunately I am not in your league-10700k clocked to 4.5, 1080Ti, 32 gb Bdie, z490 Hero. Since the 7700x3D is rumored to have higher clocks and more cache, I might wait for that. But yes unfortunately due to their architecture and mem controller, DDR5 wont go much above 6000. Still,the 13900k has 96 mb of L2 and L3 cache as well as significantly higher clocks so I dont understand why it doesnt spank the 5800x3d

The Zen 3/3D caches have been doubled from Zen2 twice and if you ask AMD they will explain the reason to you: Latency.

Their architecture is designed to take advantage of lower latencies, not higher frequencies, and this is precisely the reason why they doubled the cache, so that the RAM could be accessed more often and this cache filled with the DATA the processor processes.

No problem about your own gear, Intel processors doesn't work the same way and beneficiate a lot more than Zen 3/3D from higher frequencies, you also can O.C RAM and gain a lot more with an Intel than an AMD. Intel: How to Overclock RAM.

DDR 5 is not making as much prrogress as AMD would have liked, the RAM manufacturers doesn't have the die to reduce latency at higher frequencies but on the contrary, they have to increase it to maintain their kits stability, on the other hand, AMD next generation Zen 4 will take more advantage of the actual available range of DDR5 witn an Infinity Frabric clocking 3000MHz, 1000MHz higher than the zen 3/3D.

They also will give you the choice between two profiles: Bandwidth and  low latency. This is farsighted since I'm sure Corsair and GSkill will be looking for the next generation of Die in order to reduce latency, the advantages are not to be demonstrated. So for those willing to wait for Zen 4 it's not so bad, but myself I'll wait for lower DDR 5 latency...

ps I'm waiting for full upgrade (GPU and VR headset) to test my gear in DCS, right now I'm validating with 3D Mark Pro.

Edited November 22, 2022 by Thinder

[TKu]

Stock with PL1/2 power limits enabled.

[Thinder]

31 minutes ago, TKu said:

Stock with PL1/2 power limits enabled.

i don't test with Time Spy but Fire Strike, with settings at 4K X MSAA, first reason is Direct X 12, second "DirectX 11 benchmark for high-performance gaming PCs and overclocked systems", most apropriate for my PC since I want to play 4K VR.

First I run test to compare back-to-back performances with the CPU/RAM/GPU I replace, then different BIOS and BIOS settings/timings in order to squeeze the last bit of performances of my PC.

As part of optimization, since I also upraded one of my SSDs to a 2TB, was to set up a 3.92 GB partition and allocated my paging files to it, as well as "degreasing" Windows 10 by uninstalling stuff I don't use and disabling other app, I'm still in the process of fine tuning and have been delivered a new Freezer 7XCO cooler, the other one is mechanically compromised.

And I forgot, this is only the stage 1 of my upgrade path, I'm still waiting for the RX 7900 XTX to be availale.

Edited November 22, 2022 by Thinder

[EightyDuce]

There seems to be some misconceptions about memory on this thread that's been repeated in other threads. 

 

Samsung bdie is not some magic sauce. Samsung B die and Micron Rev E at same speed same latency will  have same performance (within margin of error). The beauty of Bdie is its overclocking headroom. For example I wa able to take my 3200 CL12 bdie to 4000 (15-16-16-35). The use case for Bdie is that you can take inexpensive kit of Bdie, overclock it to 3600 then take the rest of that bdie headroom and focus it on latency. If you're not overclocking, B die is a waste of money but cause you can get the same results with other ram (if they can hit same timings/speed) as you can with Bdie and overclocking. Honestly, Bdie is arguable wasted on a ryzen system unless you do manual tuning and not just throwing on XMP. This will be affected by board topology and silicon lottery on the memory controller. 

You can achieve the same results with a native 3600 CL14 kit. 

X3D is a bit of a wildcard is that it's v-cash allows it to basically bypass memory bottleneck. So that ram isn't as cruicial as it is with a non 3D chip because vcash essentially trumps it. That being said, I wouldn't just run potato Ram just because you have an X3D. 

Edited November 22, 2022 by EightyDuce

[skypickle]

8 minutes ago, EightyDuce said:

 

X3D is a bit of a wildcard is that it's v-cash allows it to basically bypass memory bottleneck. So that ram isn't as cruicial as it is with a non 3D chip 

There are no repeated misconceptions. I am specifically asking about the effect of cache. Would those people who upgraded from 5800x to the x3d chip please post the frametimes and cpu% that dcs reports in its tool?

@EightyDucecould you state your hardware and these results? Thinder stated he is waiting for parts but he did not state his cpu

 

 

[Thinder]

1 hour ago, EightyDuce said:

There seems to be some misconceptions about memory on this thread that's been repeated in other threads. 

 

Samsung bdie is not some magic sauce. Samsung B die and Micron Rev E at same speed same latency will  have same performance (within margin of error). The beauty of Bdie is its overclocking headroom. For example I wa able to take my 3200 CL12 bdie to 4000 (15-16-16-35). The use case for Bdie is that you can take inexpensive kit of Bdie, overclock it to 3600 then take the rest of that bdie headroom and focus it on latency. If you're not overclocking, B die is a waste of money but cause you can get the same results with other ram (if they can hit same timings/speed) as you can with Bdie and overclocking. Honestly, Bdie is arguable wasted on a ryzen system unless you do manual tuning and not just throwing on XMP. 

You can achieve the same results with a native 3600 CL14 kit. 

 

 

Well that's actually not very accurate. You mention Samsung B-Die ability for overclocking but ommit to specify the reason why it CAN overclock and that's 1) stability, 2) timing ranges.

So saying "you can get the same results with other ram " is an huge overstatement, the truth behind it is "if they can hit same timings", which is a major factor, since another die won't give you either the stability nor the range of timing, if you know better and can direct players toward a non-B-die RAM kit that equals a Corsair or a Gskill, feel free to share.

For the time being, I know where to ask for opinions and advises.

Quote

X3D is a bit of a wildcard is that it's v-cash allows it to basically bypass memory bottleneck. So that ram isn't as cruicial as it is with a non 3D chip because vcash essentially trumps it. That being said, I wouldn't just run potato Ram just because you have an X3D. 

The cache doesn't allow to bypass any RAM bootleneck, the controller limits in relation to RAM timings and the desired Infinity Fabric 1:1 ratio is still there, the presence of the chache doesn't change those controllers limits, in short if you do not have the die providing you with what the CPU needs, range of timing timings and IF 1:1 ratio, you can forget about using the cache to its full potential.

Another point you got completely wrong; Bdie kits were useed by both Intel and AMD CPUs players when the RAM speeds weren't high enough for them, meaning Zen2, from Zen3, AMD players were able to use wider timing range and take advantage of lower latencies, not exactly your simplyfied version of reality.

Then if you ask AMD technicians what they think of it they will reply this:

So this clearly counterdict completely what you are saying.

 

Quote

There seems to be some misconceptions about memory on this thread that's been repeated in other threads. 

That would be you.

 

59 minutes ago, skypickle said:

There are no repeated misconceptions. I am specifically asking about the effect of cache. Would those people who upgraded from 5800x to the x3d chip please post the frametimes and cpu% that dcs reports in its tool?

@EightyDucecould you state your hardware and these results? Thinder stated he is waiting for parts but he did not state his cpu

 

 

My CPU is an AMD Ryzen 7 5800X 3D upgraded from a 5600X. RAM have equal capacity, 32GB, Cl4, 4 X 1 bank both 3600GHz from 3200GHz.

Here are the details of my two consecutive kits:

G.SKILL TridentZ RGB Series 32GB (4 x 8GB) DDR4 3200 (PC4 25600) Desktop Memory Model F4-3200C14Q-32GTZR
G.SKILL TridentZ RGB Series 32GB (4 x 8GB) 288-Pin PC RAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C14Q-32GTZRA

 

I didn't tested in DCS yet but you can trust the results I obtained with 3D Mark Pro at 4K MSAA X 2 compared to my previous set up. 5600X/32GB 3200 GHz Cl14 4 X 1 X rank.

You can be certain that those gains will translate to DCS as they did for my previous rig, even more with this CPU.

The cache will make access to data faster because the CPU won't have to try to access those data from the RAM as often, the goal as explained in the AMD video is to reduce latency, if your RAM allows for a further reduction regardless of the frequency, within the bandwidth of the CPU, you will gain in performances.

Edited November 22, 2022 by Thinder

[BitMaster]

B-Die slowly looses grip beyond 4000MHz and that's were Micron E-Die takes over with exceptional high frequency stability, tho at looser timings but the overall gain in MHz compensates for the looser timings so latency more or less stays the same while bandwidth increases. Not relevant for any modern AMD CPU as we all know, somewhere shortly beyond 3600-3800 most/all IMC start acting up.

more or less what Buildzoid says

Edited November 22, 2022 by BitMaster

[Thinder]

36 minutes ago, BitMaster said:

B-Die slowly looses grip beyond 4000MHz and that's were Micron E-Die takes over with exceptional high frequency stability, tho at looser timings but the overall gain in MHz compensates for the looser timings so latency more or less stays the same while bandwidth increases. Not relevant for any modern AMD CPU as we all know, somewhere shortly beyond 3600-3800 most/all IMC start acting up.

more or less what Buildzoid says

 

That migt well be but the issue here is that it is the CPU controllers which set the limits, and those are from the Infinity Fabric need to have a 1:1 ratio to be perfectly aligned, plus the Zen3/3D IF is limited to 2000MHz frequency which limits the RAM available at 4000MHz to (I believe) zero for optimum performances.

Loose timings are not what the Zen 3/3D needs, and the reason why until Zen4 and an IF running at 3000MHz or a die that can provide with stability and range of timing at 4000MHz, the B-die reigns when it comes to low latency/higher frequencies. note that Zen4 is designed for DDR5.

Right now RAM manufacturers are limited to higher latencies for higher frequencies, and even if a die is stable, at the cost of lower latencies and narrower timing range it is not the equivalent of a Cl14 3600MHz. especially with a 7 5800X 3D.

To have a similar situation to Zen3/3D/B-die with frequencies above 3600MHz, we (Zen 3/3D users) will have to wait if ever, RAM manufacturers breaks this low latency barrier.

As for a 4000MHz compensating for looser timings with a Zen3/3D, that's not gonna happen at 4K, the Zen architecture and the controller are not designed for this, if you force a timing (even tested) higher than the recommanded HGz on one of those, it will perhaps work at lower resolutions when load is minimum but here is the catch, it's under load that your CPU will throttle down and this is where the bottleneck is.

It's the controllers which manage the RAM and all other channels, limited to 4 X 1 banks, 3200MHz with high street RAM (Cl16 and other), 3600 MHz with a B-die kit for optimal performances, and when the controller struggle to manage this the loss is very perceptible.

Quote

beyond 3600-3800 most/all IMC start acting up

This way beyond what the Zem3/3D Infinity Fabric is designed to cope with. it is stable with a Cl14 at 3600MHz (what I am running now), but if you fit a RAM kit with higher latency no wonder it will start acting up. Players needs to provide their CPU with the right RAM. Sticking a JATO rocket in the butt of an old Chevy is not going to make a Formula One of it.

So guys, do your test at 4K if you really want relevant results, at 1080P the load on the CPU is not high enough to be fully indicative of the potential of your combo.

Edited November 22, 2022 by Thinder

[skypickle]

There are a lot of words happening in this thread but not addressing the question of cache performance with real data. Please stop talking about RAM. A 13900k has 2/3 the cache of a 5800x3d which is clocked almost 1/3 slower. What numbers are people getting with the x3d chip in dcs using the dcs performance monitor?