9800X vs 9900K in gaming performance

[fogSource]

Assuming the clock speeds are set the same.  And a range of old and new, single and multi threaded games are tested.

 

Which would benchmark better in CPU intensive games?

 

Do you think there/will there be any advantages to the 9800X with it's MESH and the non inclusive cache in future games? 

 

Will older games with one or two threads perform better on ringbus cpus like the 9800K? 

 

 

 

 

[Enderman]

Just get a 9900K, it's the best for gaming that currently exists.

[corrado33]

4 minutes ago, fogSource said:

Assuming the clock speeds are set the same.  And a range of old and new, single and multi threaded are games tested.

 

 

 

I'm guessing the 9900k will usually win by a little in gaming performance overall, and have a significant advantage in older games with less threads?

 

Are there/will there be any advantages to the 9800X with it's MESH and the non inclusive cache in future games? 

 

And do older games with one or two threads perform better on ringbus cpus like the 9800K? 

 

 

 

 

The 9800X is on the 2066 platform, which is intel's "high performance" platform. You'll likely be able to OC the 9800X better, and the "extreme" chips tend to be higher binned than the non extreme chips. 

 

But honestly, you probably won't notice much of a difference in games. Both CPUs are good enough to drive any modern game (I mean....they are both top of the line).

 

Just depends if you want the extra features of the 2066 platform. (And the extra cost as well.)

[Mira Yurizaki]

I don't think this is a discussion of "should I get the 9800X or 9900K?" This is a discussion on "Which is better, mesh or ring?"

 

10 minutes ago, fogSource said:

Which would benchmark better in CPU intensive games?

More than likely the 9800X if the application really is CPU bound.

 

10 minutes ago, fogSource said:

Do you think there/will there be any advantages to the 9800X with it's MESH and the non inclusive cache in future games? 

If the game can use as many cores as the 9800X can provide, then the mesh architecture won't be a problem. It's only a problem if the application's threads landed on cores that are physically further apart and they share data constantly.

 

The choice of using an exclusive cache may or may not matter depending on the application. Exclusive caches have the advantage of not storing redundant data, thus increasing the chances of some piece of data being in cache. However, this may come at the cost of a higher miss rate latency.

 

10 minutes ago, fogSource said:

Will older games with one or two threads perform better on ringbus cpus like the 9800K?

It depends on where the threads land on. A mesh system theoretically can outperform a ring system if the threads land on neighboring cores while the ring one got stuck with cores as far apart from each other.

[fogSource]

Just now, M.Yurizaki said:

I don't think this is a discussion of "should I get the 9800X or 9900K?" This is a discussion on "Which is better, mesh or ring?"

 

More than likely the 9800X if the application really is CPU bound.

 

If the game can use as many cores as the 9800X can provide, then the mesh architecture won't be a problem. It's only a problem if the application's threads landed on cores that are physically further apart and they share data constantly.

 

The choice of using an exclusive cache may or may not matter depending on the application. Exclusive caches have the advantage of not storing redundant data, thus increasing the chances of some piece of data being in cache. However, this may come at the cost of a higher miss rate latency.

 

It depends on where the threads land on. A mesh system theoretically can outperform a ring system if the threads land on neighboring cores while the ring one got stuck with cores as far apart from each other.

Yes! Thank you!  Which is better mesh or ring?  In relation to games, older and newer. 

 

Fascinating, so threads will land randomly or does something control it?

 

I'd say my worst case scenario older game is Skyrim with tons of mods, scripts, ect.. Two really heavy threads. 

 

Does ringbus have any advantages over MESH if the threads in a 2-4 threaded game land far apart on say, an 8 or more core CPU?

 

 

 

[Mira Yurizaki]

23 minutes ago, fogSource said:

Which is better mesh or ring?  In relation to games, older and newer. 

I don't think either is. The problem mesh vs. ring intercommunication is trying to solve is scalability.

 

Quote

Fascinating, so threads will land randomly or does something control it?

For all intents and purposes, randomly. But for most OSes, it depends on which CPU core happens to be available first.

 

Quote

I'd say my worst case scenario older game is Skyrim with tons of mods, scripts, ect.. Two really heavy threads. 

 

Does ringbus have any advantages over MESH if the threads in a 2-4 threaded game land far apart on say, an 8 or more core CPU?

It depends on how the processor was laid out. Assuming a 2x4 arrangement, the worst case number of hops is going to be 4 on a ring bus. For mesh, if laid out in the same manner, it'll be the same. But if you arrange the mesh in a 3x3 array (with say one of the corners missing), then the worst case number of hops reduces to 3.

 

EDIT: In general, ring bus is fine for lower core counts. Mesh bush is needed for higher core counts. I guess after 8 is the magic number where you need to start using mesh.

[fogSource]

9 minutes ago, M.Yurizaki said:

I don't think either is. The problem mesh vs. ring intercommunication is trying to solve is scalability.

 

For all intents and purposes, randomly. But for most OSes, it depends on which CPU core happens to be available first.

 

It depends on how the processor was laid out. Assuming a 2x4 arrangement, the worst case number of hops is going to be 4 on a ring bus. For mesh, if laid out in the same manner, it'll be the same. But if you arrange the mesh in a 3x3 array (with say one of the corners missing), then the worst case number of hops reduces to 3.

Interesting, so the loss in performance in certain games and apps would be from the cache not the MESH? 

So ringbus hops at the same speed MESH hops at?  If that makes sense.

[Mira Yurizaki]

1 minute ago, fogSource said:

Interesting, so the loss in performance in certain games and apps would be from the cache not the MESH? 

So ringbus hops at the same speed MESH hops at?  If that makes sense.

More or less. If the cores have to share data then they should be physically closer to each other to avoid performance loss.

[fogSource]

That was extremely helpful, thank you.

[Lathlaer]

1 hour ago, M.Yurizaki said:

In general, ring bus is fine for lower core counts. Mesh bush is needed for higher core counts. I guess after 8 is the magic number where you need to start using mesh.

Generally it's game and resolution dependent but at the same core speed ringbus is superior vs. mesh. Intel themselves had confirmed it:

 

Quote

Intel themselves admits that Broadwell-E (last-generation, X99) CPUs can outperform Skylake-X (current-generation, X299) due to this change in thread-to-thread communication architecture. This increases core communication latency but offers improvements in other areas, especially for the super-high core counts seen in Skylake-SP.

 

And here is the direct quote:

 

Quote

we have noticed that there are a handful of applications where the Broadwell-E part is comparable or faster than the Skylake-X part. These inversions are a result of the “mesh” architecture on Skylake-X vs. the “ring” architecture of Broadwell-E.

Source: https://www.overclock3d.net/news/cpu_mainboard/intel_s_skylake-x_architecture_offers_less_gaming_performance_than_kaby_lake/1

 

I've also seen few benchmarks i7 6950x vs. i9 7900X in games and generally speaking Broadwell-E was either flat out better or at least the same in performance while achieving lower overclock (we are talking 4.4GHz vs. 4.7GHz).

 

Granted, there aren't higher core count CPU's with ringbus on HEDT so I have no idea how much that would affect if Intel made an 18-core ringbus CPU and compared it to the i9 7980XE.

 

All I can say this - if someone is going to use HEDT in gaming, the mesh absolutely has to be overclocked.

[Mira Yurizaki]

18 minutes ago, Lathlaer said:

.

That comment doesn't confirm that the ring is generally better at the same clock speed, only that in some cases it's better. And that sounds like a generic non-answer to describe performance dependencies
 

I've noticed in the article that yours is quoting, the two processors have a 2x5 arrangement. Mesh isn't going to be any better than ring in this regard in worst case scenarios (and likely it may be worse due to the system needing to figure out routing)

 

Another problem is that nobody's comparing a 18-core with a ring bus vs. an 18-core with a mesh bus. I would put down money that an 18-core ring bus will on average be worse than an 18-core mesh bus at the same clock speed.

[porina]

Look at gaming benchmarks when Skylake-X came out. Like for like against the consumer parts, they were generally slower. Part of it was the slower clock of the mesh bus, but even when overclocked it doesn't entirely close the gap. The only new question is if going to 8 cores rather than 6 changes this.

 

I forget the exact number but a single ring can cope with at least 10 cores. I have a 14 core Xeon which has some kind of ring configuration (may be more than one), so ring bus can be used at higher core counts also, but with additional considerations.

 

12 minutes ago, Lathlaer said:

All I can say this - if someone is going to use HEDT in gaming, the mesh absolutely has to be overclocked.

I had noticed a change in mesh clock with mobo bios updates. Early on with a 7800X on Asus TUF Mark 2 the mesh ran at 2000. Mine would overclock to 3000 happily. In GTAV it did give a small but measurable impact. Fast forward to recently, I was doing updates to the system including a new bios. I forgot to re-overclock the mesh but was interested to find it now defaulted to 2700. A nice bump. Maybe Asus or Intel were too cautious early on in the product cycle to turn the clocks up. As said, I knew mine would work at 3000 (no boot at 3100) so 2700 still leaves a fair bit of headroom. I knew others with higher models in the range and they didn't get much beyond 3000 with a manual OC.

 

Right now I'm moving bits around, with the intention of using my 7800X as my secondary game system. CPU is less important as I'm going 4k on this one. I'm running a benchmark against the 6930k it is replacing but it wont be at same clocks as I run them stock. IMO even if the 7800X isn't as good in gaming as say, 8700k, the difference isn't big enough to matter in practice and the platform will still give me the edge when doing things outside of gaming.

[Lathlaer]

5 minutes ago, M.Yurizaki said:

That comment doesn't confirm that the ring is generally better at the same clock speed, only that in some cases it's better. And that sounds like a generic non-answer to describe performance dependencies

Well yeah. But I have never seen a mesh vs. ringbus comparison with the same core count and core speed where mesh would be better in gaming (ie. "some cases").

 

6 minutes ago, M.Yurizaki said:

Another problem is that nobody's comparing a 18-core with a ring bus vs. an 18-core with a mesh bus. I would put down money that an 18-core ring bus will on average be worse than an 18-core mesh bus at the same clock speed.

Because there aren't any normal CPU's with that core count and ringbus, so you can't really compare. If you want an 18-core with ringbus then you probably have to get a Xeon and then you'd have to what, downclock the 7980XE to 2.5-3GHz in order to make the comparison fair? Not really practical

[Mira Yurizaki]

6 hours ago, Lathlaer said:

Well yeah. But I have never seen a mesh vs. ringbus comparison with the same core count and core speed where mesh would be better in gaming (ie. "some cases").

Then all that supports is in that processor layout that the ring bus is still viable.

 

6 hours ago, Lathlaer said:

Because there aren't any normal CPU's with that core count and ringbus, so you can't really compare. If you want an 18-core with ringbus then you probably have to get a Xeon and then you'd have to what, downclock the 7980XE to 2.5-3GHz in order to make the comparison fair? Not really practical

If you want to continue finding evidence to support that ring bus is superior in a general case, then you'd have to do that.

[Lathlaer]

3 minutes ago, M.Yurizaki said:

If you want to continue finding evidence to support that ring bus is superior in a general case, then you'd have to do that.

Did I give any indication that I wanted to do that? I have claimed very specific scenario (gaming) from the very beginning.

[Lathlaer]

@M.Yurizaki ok I see it now. This quote of mine:

 

7 hours ago, Lathlaer said:

Generally it's game and resolution dependent but at the same core speed ringbus is superior vs. mesh. Intel themselves had confirmed it:

sounds too general but I had meant only in gaming scenarios (as per the title of this thread).

 

I should've phrased that more clearly, my bad, but I agree that for general use that was meant for those CPU's (ie. NOT gaming) mesh is superior (as evidenced by 7900X smoking the 6950X in nongaming applications).

[Mira Yurizaki]

25 minutes ago, Lathlaer said:

@M.Yurizaki ok I see it now. This quote of mine:

 

sounds too general but I had meant only in gaming scenarios (as per the title of this thread).

 

I should've phrased that more clearly, my bad, but I agree that for general use that was meant for those CPU's (ie. NOT gaming) mesh is superior (as evidenced by 7900X smoking the 6950X in nongaming applications).

I've also been mostly talking about this topic from a generic point of view because I don't want to give the wrong impression that for some class of application, you will always (or generally) have this behavior. So that's also my bad.

 

I mean, who knows if games in the future will be designed around mesh architectures. AMD already put their foot in the door with Zen being more or less a fully connected mesh architecture.