Proof that a heavily overclocked 6600K will bottleneck a GTX 1070 or above

[Tieox]

Just now, don_svetlio said:

But then that means the Pentium doesn't really belong to this discussion, now, doesn't it?

But still does not need to be slammed, sure it has no place in our builds, but let us be civil about it, or we just begin to look like elitist arseholes.

 

Not everyone on this forum has the cash flow to support a i5 6600k or i7 build. 

[don_svetlio]

Just now, Kierax said:

But still does not need to be slammed, sure it has no place in our builds, but let us be civil about it, or we just begin to look like elitist arseholes.

 

Not everyone on this forum has the cash flow to support a i5 6600k or i7 build. 

Well, if you look at my posting history, I am among the few who sing the praise of the 860K because it does what the G3258 can't - plays all games good enough while being cheap.

[Tieox]

1 minute ago, don_svetlio said:

Well, if you look at my posting history, I am among the few who sing the praise of the 860K because it does what the G3258 can't - plays all games good enough while being cheap.

 

Fair enough  

[Deadpool2onBlu-Ray]

1 hour ago, don_svetlio said:

Can we finally agree that at 1080p EVERY i5 IS a bottleneck to the 1070/1080? Yes? Good.

I am in the progress of upgrading my 4590. It was bottle necking my 1070 hard 

[ConiferousJelly]

So I literally have the 1070 Strix sitting in its box next to me while I wait for my 6600k to arrive, is this seriously going to be a huge issue for me gaming at 1080p? Or is this just to show that while the difference is 150fps -140fps it is still technically a bottleneck?

 

[Sintezza]

Nothing new realy.

And honnestly who cares?

 

A GTX1080 is not a 1080p card to start with.

But there allways will be a bottleneck at a certain point, no matter which cpu.

[zMeul]

16 hours ago, Matias_Chambers said:

Watch the video from 3:20. Especially in CPU intensive games the 6600K will bottleneck a GTX 1070 and anything above. Even when heavily overclocked. 

you blame texture streaming issues on the CPU ?!?! I'm facepalming right now

[ace_cheaply]

6 hours ago, ConiferousJelly said:

So I literally have the 1070 Strix sitting in its box next to me while I wait for my 6600k to arrive, is this seriously going to be a huge issue for me gaming at 1080p? Or is this just to show that while the difference is 150fps -140fps it is still technically a bottleneck?

 

You will get lower averages, lower mins, lower maxes, higher frame rates, and more stutter above 60hz.  In games that are cpu bound.  Yes, it's noticeable, how huge of an issue is up to you.  Will it stop you from playing?  No, absolutely not.  Will it break an immersive experience and frustrate you that your $1000 computer isn't performing as well as an $1100 one?  It would me, that choice is up to you.

[Nena Trinity]

6 hours ago, Sintezza said:

Nothing new realy.

And honnestly who cares?

 

A GTX1080 is not a 1080p card to start with.

But there allways will be a bottleneck at a certain point, no matter which cpu.

Everyone I know buys 980Tis and 1080s for 1080p... o3o

[TheGamingBarrel]

22 hours ago, Matias_Chambers said:

What? 90% is still bottlenecking and it does drop to 80% sometimes. 

90% Is also within margin of error, or just a segment where it needs less GPU.

[BuckGup]

So is this only an i5 problem? Also would adding more cores help or is it a speed of the cores/efficiency?

[Matias_Chambers]

Just now, SamStrecker said:

So is this only an i5 problem? Also would adding more cores help or is it a speed of the cores/efficiency?

Many games do use 8 threads. GTA V does for sure. I have heard people say that GTA V can use 16 threads but not sure if that is true or not. 

[BuckGup]

Just now, Matias_Chambers said:

Many games do use 8 threads. GTA V does for sure. I have heard people say that GTA V can use 16 threads but not sure if that is true or not. 

So next gen GPUs that everyone says will be bottlenecked by the GPUs can be avoided with say a 6 core or 8 core? I don't see that as a problem as we did this with dual cores and single for gaming. I know GTA V runs better on my 6800K then my 4670k

[ConiferousJelly]

4 hours ago, ace_cheaply said:

You will get lower averages, lower mins, lower maxes, higher frame rates, and more stutter above 60hz.  In games that are cpu bound.  Yes, it's noticeable, how huge of an issue is up to you.  Will it stop you from playing?  No, absolutely not.  Will it break an immersive experience and frustrate you that your $1000 computer isn't performing as well as an $1100 one?  It would me, that choice is up to you.

Above 60hz. So im good, my two monitors are only 60hz, and if I do upgrade to 1440p @ 144hz ill be good because with that a CPU upgrade will be viable.

[Guest]

22 hours ago, Cracklingice said:

1070 isn't intended for 1080p gaming.  You're taking a notoriusly CPU bound title and running it at the resolution most likely to become CPU bound on a GPU that is intended for 1440p / starter 4k / VR content.  Yes it's going to be CPU in the instance presented and likely quite noticeably; however, it's still very playable frame rates.

/thread 

 

stupid argument in the first place. 

[MageTank]

3 Pages of this, and I have to be the first one to bring it up? Oh well, time for another unpopular fact. 8GB of 2666mhz C14 ram, in single channel. Effective memory bandwidth is 21,300MB/s peak theoretical bandwidth. That is slower than 2133mhz in dual channel (17,000MB/s single channel, 34,000MB/s dual). We already proved countless times over and over again that memory bandwidth is impacted by the amount of CPU overhead. These videos present us with a GTX 1070 running in 1080p. The CPU overhead is apparent, and yet we treat this is an absolute fact when the video is showing a terrible memory setup? Come on people.

 

Before anyone says ram speed won't matter, please, do yourself and the world a favor and just test it yourself. Overclock (or underclock, if you are incapable of overclocking) or even remove a few sticks to use Single Channel. Pay very close attention to your minimum framerates, and see just how helpful memory speed (or lack thereof) is to gaming. While memory speed won't magically make this CPU not be a bottleneck, it WILL help when it does become the bottleneck. The fact that the i5 is using 100% across all 4 cores in certain situation tells you that hyper-threading is helpful in some titles. Does it mean the i5 is a bottleneck to a GTX 1070? That depends on the title/workload. Either way, I would like to see a better testing methodology than this. At least 16GB of decently clocked ram in dual channel for 1080p. We all know that at higher resolutions, the i5 would not be a bottleneck anyways. 

[SpaceGhostC2C]

51 minutes ago, MageTank said:

 The fact that the i5 is using 100% across all 4 cores in certain situation tells you that hyper-threading is helpful in some titles.

This is a misunderstanding of what hyperthreading is and what it's useful for. Four cores maxed out are four cores maxed out, no matter how you slice it into "logical cores".

 

Hyperthreading is a server technology that somehow made it to desktop computers, I guess to help in heavy multitasking. Server workloads typically consist in millions of tiny tasks (for example, receiving requests for a particular webpage from thousands of users). That's why so many Xeons have a lot of cores (compared to consumer CPUs of the same generation), even at the expense of significantly lower clocks. What matter is to handle as many requests per second as possible, and having more, weaker cores helps compared to having fewer, stronger cores (an Xeon with hyperthreading enabled  will have ~10% higher throughput than the same Xeon with hyperthreading disabled).

 

On heavy computational tasks, though, hyperthreading is useless at best and harmful at worst. That is, when you have to run fewer, but heavier tasks (as oppose as a lot of tiny tasks, as a server), stronger cores are better than more, but weaker cores (meaning twice as many cores running at half the speed). In the most paralelizable tasks, theoretically, twice the cores at half the speed should give the same result. In practice, there is always some paralelization overhead. I have tested myself in a cluster of 6x2xquad core Xeons and switching hyperhtreading was actually a little bit harmful for computational times: one round of 16 threads per node would take slightly more than 2 rounds if 8 threads per node, even in OpenMP (which doesn't add the communication overhead of MPI; for MPI it is much worse). All hypertrheading does in that context is pretending to be twice as many cores at half the speed.

I know that gaming usually doesn't entail much paralelization outside the GPU, so the multiple threads are typically running different tasks. Still, one core per thread i sall you need. If the game does have some degree of paralelization, though (the only case in which it would adapt to the number of available logical cores and expand to more  cores if available), hyprthreading will never do better than just equating the number of threads and cores, as it would fall in the same type of problem as the scientific computations I mentioned above.

 

If an i7 performs better than an i5, the explanation must lie elsewhere (but you can test this application by application by using an i7 with and without hyperthreading).

[MageTank]

14 minutes ago, SpaceGhostC2C said:

This is a misunderstanding of what hyperthreading is and what it's useful for. Four cores maxed out are four cores maxed out, no matter how you slice it into "logical cores".

 

Hyperthreading is a server technology that somehow made it to desktop computers, I guess to help in heavy multitasking. Server workloads typically consist in millions of tiny tasks (for example, receiving requests for a particular webpage from thousands of users). That's why so many Xeons have a lot of cores (compared to consumer CPUs of the same generation), even at the expense of significantly lower clocks. What matter is to handle as many requests per second as possible, and having more, weaker cores helps compared to having fewer, stronger cores (an Xeon with hyperthreading enabled  will have ~10% higher throughput than the same Xeon with hyperthreading disabled).

 

On heavy computational tasks, though, hyperthreading is useless at best and harmful at worst. That is, when you have to run fewer, but heavier tasks (as oppose as a lot of tiny tasks, as a server), stronger cores are better than more, but weaker cores (meaning twice as many cores running at half the speed). In the most paralelizable tasks, theoretically, twice the cores at half the speed should give the same result. In practice, there is always some paralelization overhead. I have tested myself in a cluster of 6x2xquad core Xeons and switching hyperhtreading was actually a little bit harmful for computational times: one round of 16 threads per node would take slightly more than 2 rounds if 8 threads per node, even in OpenMP (which doesn't add the communication overhead of MPI; for MPI it is much worse). All hypertrheading does in that context is pretending to be twice as many cores at half the speed.

I know that gaming usually doesn't entail much paralelization outside the GPU, so the multiple threads are typically running different tasks. Still, one core per thread i sall you need. If the game does have some degree of paralelization, though (the only case in which it would adapt to the number of available logical cores and expand to more  cores if available), hyprthreading will never do better than just equating the number of threads and cores, as it would fall in the same type of problem as the scientific computations I mentioned above.

 

If an i7 performs better than an i5, the explanation must lie elsewhere (but you can test this application by application by using an i7 with and without hyperthreading).

I stand corrected. You are correct. If all 4 cores are maxed out, no amount of additional threads will help, even if the task can handle additional threads. The only time additional threads help, is when the cores are underutilized and they are waiting for a current task to be completed (allowing the hyperthread core to handle that process simultaneously). I had completely neglected to think of that before saying it.

 

Still, my point on the memory still stands. When there is this much CPU overhead, faster ram will help. The fact that someone thought it was a good idea to test a potential bottleneck with a single, slow 8GB stick of ram is silly. 

[SpaceGhostC2C]

5 minutes ago, MageTank said:

Still, my point on the memory still stands. When there is this much CPU overhead, faster ram will help. The fact that someone thought it was a good idea to test a potential bottleneck with a single, slow 8GB stick of ram is silly. 

Certainly, underutilizing the IMC capabilities won't help the CPU doing its job.

 

Coming back to the i5 vs i7 clocked to the same speed comparison, something that remains different, and matters considerably for CPU performance, is the cache size, which is somewhat related to your point about memory speed.

Other than that, same architecture plus same speed means there's little else to expect different performances.

[MageTank]

Just now, SpaceGhostC2C said:

Certainly, underutilizing the IMC capabilities won't help the CPU doing its job.

 

Coming back to the i5 vs i7 clocked to the same speed comparison, something that remains different, and matters considerably for CPU performance, is the cache size, which is somewhat related to your point about memory speed.

Other than that, same architecture plus same speed means there's little else to expect different performances.

Man. Where were you when i was complaining about the LTT test that "emulated" lower CPU's by shutting off cores of an i7? I tried to explain the impact of cache back then, and people said it did not matter. Clearly they have never ran VM's, as cache becomes extremely important very quickly. I have personally tested the impact of different cache speeds, but could not get my cache to run slow enough to have a negative impact (200GB/s is about as slow as I could get it, when it is normally 1000GB/s by default). I will be testing 6mb vs 8mb later on, by forcing my iGPU to be enabled (uses 2mb of cache when active) to see what kind of impact cache amount has on gaming and other work related tasks.

[Tieox]

But which CPU can cook a rasher of bacon quicker  priorities people!