Best money-can-buy gaming cpu?

[BrianMunching]

Hello,

 

Sorry this is probably asked 100s of times a day, but what right now is the best money-can-buy gaming cpu? 

Either Intel or AMD, no budget limit.

 

Thanks, :)

[will0hlep]

11 minutes ago, BrianMunching said:

Hello,

 

Sorry this is probably asked 100s of times a day, but what right now is the best money-can-buy gaming cpu? 

Either Intel or AMD, no budget limit.

 

Thanks,

14900K probably edges it in pure performance, but it is hard to cool and you'd be buying into a dead platform, so I'd actually say 7950X3D (although there are plenty of games were the 7800X3D will be just as good).

I realise this is not the straight answer you were after

[Bagzie]

3 minutes ago, will0hlep said:

14900K probably edges it, but it is hard to cool and you'd be buying into a dead platform, so I'd recommend the 7950X3D.

For strictly gaming performance the fastest CPU is the 7800X3D overall.

[Dedayog]

15 minutes ago, BrianMunching said:

Hello,

 

Sorry this is probably asked 100s of times a day, but what right now is the best money-can-buy gaming cpu? 

Either Intel or AMD, no budget limit.

 

Thanks,

7800x3d as it's been for months.  Google would have handled this easily and proven it as well.

[jaslion]

7800x3d

[YoungBlade]

Here's the most recent data from TechSpot/Hardware Unboxed when it comes to CPU gaming performance for the 14900K review. Across a range of games at 1080p with an RTX 4090, the 7800X3D is the fastest CPU, beating the 14900K by 7.5%

 

[Dedayog]

15 minutes ago, will0hlep said:

14900K probably edges it in pure performance, but it is hard to cool and you'd be buying into a dead platform, so I'd actually say 7950X3D (although there are plenty of games were the 7800X3D will be just as good).

I realise this is not the straight answer you were after

7950x3d if you manually work the CCD's properly.  @Agall wrote up a great thread or two on it.

 

So out of the box, the 7800x3d is the best (producing the most frames on average) gaming CPU.

[Agall]

52 minutes ago, Dedayog said:

7950x3d if you manually work the CCD's properly.  @Agall wrote up a great thread or two on it.

 

So out of the box, the 7800x3d is the best (producing the most frames on average) gaming CPU.

This weekend, I even disabled CCD1 to test some anomalies I've been getting in Warframe. It uses p2p hosting (I'm usually the host) and enemy density is heavily CPU bound, but I was getting substantial hitches in framerate while noticing light CCD1 utilization. I've been unable to replicate the issue since disabling CCD1, which was done after doing live testing with Processor Affinity that didn't fix it. Its also possible that it was an anomaly in 2 different missions and isn't the case anymore, but time will tell.

 

The current theory being that the most recent Dagath update did some backend DX12 multithreading optimizations that the 7950x3D isn't handling well, mind you, CCD0 isn't getting maxed out when it seems to load CCD1. Just seems like its forcing lower CCD0 usage and higher CCD1 usage which is causing latency having to cross the fabric.

 

Yeah, even the 7950x3D in a 8+0 is technically better since it can boost 200MHz higher, but you don't see that almost ever and it's not worth the extra ~$300. You don't even see that higher boost when down to 2c/4t on CCD0, so its a sharp thermal limitation of the chip, which is likely why the 7800x3D can only boost to 5.05GHz (since its likely to have a higher per core utilization than a 16c/32t CPU). 7900x3D's CCD0 even has a 5.25GHz max boost to support my narrative.

[will0hlep]

1 hour ago, Dedayog said:

7950x3d if you manually work the CCD's properly.  @Agall wrote up a great thread or two on it.

 

So out of the box, the 7800x3d is the best (producing the most frames on average) gaming CPU.

1 hour ago, YoungBlade said:

Here's the most recent data from TechSpot/Hardware Unboxed when it comes to CPU gaming performance for the 14900K review. Across a range of games at 1080p with an RTX 4090, the 7800X3D is the fastest CPU, beating the 14900K by 7.5%

2 hours ago, jaslion said:

7800x3d

2 hours ago, Dedayog said:

7800x3d as it's been for months.  Google would have handled this easily and proven it as well.

2 hours ago, Bagzie said:

For strictly gaming performance the fastest CPU is the 7800X3D overall.

Clearly I need to go and watch some newer benchmarks

[YoungBlade]

16 minutes ago, will0hlep said:

Clearly I need to go and watch some newer benchmarks

The 14900K still trades blows with the 7800X3D. In some games, like Starfield, it actually wins by a decent margin. It's just that, on average across a range of games, the 7800X3D comes out ahead overall.

 

But the numbers I put are with a 4090 at 1080p, which isn't very realistic for a 4090. At 1440p, the difference becomes rather marginal. And at 4K, there's functionally no difference on average between the 14900K and 7800X3D for gaming.

 

So if budget isn't a consideration, there's nothing wrong with going for the 14900K - it's not exactly slow.

[Dedayog]

1 hour ago, will0hlep said:

Clearly I need to go and watch some newer benchmarks

 

46 minutes ago, YoungBlade said:

The 14900K still trades blows with the 7800X3D. In some games, like Starfield, it actually wins by a decent margin. It's just that, on average across a range of games, the 7800X3D comes out ahead overall.

 

But the numbers I put are with a 4090 at 1080p, which isn't very realistic for a 4090. At 1440p, the difference becomes rather marginal. And at 4K, there's functionally no difference on average between the 14900K and 7800X3D for gaming.

 

So if budget isn't a consideration, there's nothing wrong with going for the 14900K - it's not exactly slow.

Basically we couldn't tell them  apart, similar to SSD's.  But if the OP is asking for a definitive BEST... then even a 1% variance produces a best.

 

That said, I would look more at performance, power usage and cost as Trinity.  

 

For me, that pushes the 14900K out the door regardless if it won on performance.

[pienutt]

a very well cooled overclocked Ryzen 7 7800X3D

[Agall]

@Dedayog @YoungBlade @will0hlep

 

Regarding the 7800x3D versus 14900k, the skeptic in me believes it simply has to do with MCM's architecture. When you boil it down, the CCD(s) have to negotiate across a bus on the substrate to communicate with the memory controllers. AKA, the less amount of instances where the CCD has to negotiate ALL the way to the I/O die then ALL the way to system RAM, the better. I'm no microprocessor engineer, but the logic with my understanding checks out.

 

 

 

As good as 3D v-cache is at improving performance, it seems like a bandaid to the issue of poor memory performance/timings/latency with MCM CPU architecture, where tasks that exceed the standard L2+L3 cache and have to go all the way to memory provide a sharp latency limitation without the extra 64MB of L3.

 

The simple fact that Intel keeps up while on 10nm to AMD with 5nm/6nm MCM points towards a limitation with MCM CPU architecture, especially 7 generations in. Almost as if they chose +64MB for a reason (outside of NAND limitations), because why not just put an HBM3 die on the substrate instead? (assuming the substrate is big enough to support the extra traces)

 

The reality is if Intel simply manufactured their CPUs on TSMC's 5nm with a monolithic die, it would be expensive, but probably annihilate AMD for generations.

[YoungBlade]

12 minutes ago, Agall said:

@Dedayog @YoungBlade @will0hlep

 

Regarding the 7800x3D versus 14900k, the skeptic in me believes it simply has to do with MCM's architecture. When you boil it down, the CCD(s) have to negotiate across a bus on the substrate to communicate with the memory controllers. AKA, the less amount of instances where the CCD has to negotiate ALL the way to the I/O die then ALL the way to system RAM, the better. I'm no microprocessor engineer, but the logic with my understanding checks out.

 

 

 

As good as 3D v-cache is at improving performance, it seems like a bandaid to the issue of poor memory performance/timings/latency with MCM CPU architecture, where tasks that exceed the standard L2+L3 cache and have to go all the way to memory provide a sharp latency limitation without the extra 64MB of L3. The simple fact that Intel keeps up while on 10nm to AMD with 5nm/6nm MCM points towards a limitation with MCM CPU architecture, especially 7 generations in. Almost as if they chose +64MB for a reason (outside of NAND limitations), because why not just put an HBM3 die on the substrate instead? (assuming the substrate is big enough to support the extra traces)

 

The reality is if Intel simply manufactured their CPUs on TSMC's 5nm with a monolithic die, it would be expensive, but probably annihilate AMD for generations.

I don't think fast memory alone can compete with lots of cache. This has been tested with Intel CPUs in the past, where if you disable cores and drop clocks to match lower core count parts, the higher tier part still performs better for gaming. The whole reason that the 14900K performs better than the 14700K for gaming is that the i9 has additional cache.

 

Obviously there's a limit to how effective that strategy is. But Intel absolutely needed to up their L3 cache values in the last couple of gens in order to compete with AMD for gaming. If the 14900K still only had the 16MB of cache for its P-cores that the 11900K had, rather than the 36MB it does have, it would be noticably worse for gaming.

 

So while V-Cache is a bit of a hack to deal with a limitation of MCM architectures, it does give real performance improvements in gaming because games generally benefit from more cache beyond mere memory latency improvements.

[Agall]

1 hour ago, YoungBlade said:

I don't think fast memory alone can compete with lots of cache. This has been tested with Intel CPUs in the past, where if you disable cores and drop clocks to match lower core count parts, the higher tier part still performs better for gaming. The whole reason that the 14900K performs better than the 14700K for gaming is that the i9 has additional cache.

I don't believe that, but I don't have a 14600k and 14900k to test that. The extra L3 cache might help a little in some situations, but I believe its simply a binning advantage and boost clock advantage. I haven't seen any testing of this, taking a 12th/13th/14th gen i9 and down-configuring the core and boost clocks to equate it to the i5 and seeing how much the extra cache matters (although I did submit it as a Labs idea on the forums). What's more substantial than a bit of extra L3 cache is the boost clocks between the 14700k and 14900k, being 400MHz. Increases in cache just not being directly comparable to how it is on Ryzen.

 

My whole point being that 3D v-cache, as good as it is, seems like a kludgey solution to a latency issue with MCM architecture, since compared to the monolithic dies of Intel, the memory controllers on Ryzen are more complex. Potentially why Ryzen is more sensitive on timings and can't go as low as Intel CPUs. The theory being that major operations in some applications fall within the 104MB of L2+L3 and therefore don't have to go all the way to system RAM, therefore reducing the latency of those operations.

[YoungBlade]

23 minutes ago, Agall said:

I don't believe that, but I don't have a 14600k and 14900k to test that. The extra L3 cache might help a little in some situations, but I believe its simply a binning advantage and boost clock advantage. I haven't seen any testing of this, taking a 12th/13th/14th gen i9 and down-configuring the core and boost clocks to equate it to the i5 and seeing how much the extra cache matters (although I did submit it as a Labs idea on the forums). What's more substantial than a bit of extra L3 cache is the boost clocks between the 14700k and 14900k, being 400MHz. Increases in cache just not being directly comparable to how it is on Ryzen.

The closest testing I've seen is this IPC testing from Hardware Unboxed where he disables the E cores and clocks 12th, 13th, and 14th gen at 5.0GHz. So this shows the difference between the i9 and i7 at identical clock speeds, where the only difference is the cache. The biggest difference is seen in Balder's Gate 3, which is a more cache sensitive game, where the 14900K is 2% faster on average and 5% faster for the 1% lows compared to the 14700K. If cache played no role, they should be identical. The 14900K has 9% more cache than the 14700K, so a 5% uplift in 1% lows is pretty significant considering that that's the only factor at play.

This was also tested more extensively with 10th gen CPUs, and it was shown that, with 6 cores enabled, the 10900K was up to 18% faster from cache alone (in R6S) and in many games, the cache was more important than the core count.

[Agall]

6 minutes ago, YoungBlade said:

The closest testing I've seen is this IPC testing from Hardware Unboxed where he disables the E cores and clocks 12th, 13th, and 14th gen at 5.0GHz. So this shows the difference between the i9 and i7 at identical clock speeds, where the only difference is the cache. The biggest difference is seen in Balder's Gate 3, which is a more cache sensitive game, where the 14900K is 2% faster on average and 5% faster for the 1% lows compared to the 14700K. If cache played no role, they should be identical. The 14900K has 9% more cache than the 14700K, so a 5% uplift in 1% lows is pretty significant considering that that's the only factor at play.

This was also tested more extensively with 10th gen CPUs, and it was shown that, with 6 cores enabled, the 10900K was up to 18% faster from cache alone (in R6S) and in many games, the cache was more important than the core count.

Its going to be application dependent since a given instruction that can operate in cache will. So, there's a balancing point on that, since if something need 1MB more cache to operate exclusively in cache, it'll have to use system RAM. 3D v-cache having far more than 5% of an increase, as an example.

 

Also have to consider the fact that Intel's binning scheme will inevitably allow the next tier to be better, since its simply a better bin of the same silicon. That'll allow it to boost higher more frequently, regardless of the cap. Where we'd likely need to see this is comparing the chips with turboboost disabled and at a fixed clock speed to truly test. Its obvious between the 7700x and 7800x3D since there's a 3x the L3 cache in comparison to 3MB between the 14700k and 14900k. I'd also suspect the L2 cache to be A LOT more impactful since its sum is nearly the same as the total L3 cache available to all cores.

[YoungBlade]

2 minutes ago, Agall said:

Also have to consider the fact that Intel's binning scheme will inevitably allow the next tier to be better, since its simply a better bin of the same silicon. That'll allow it to boost higher more frequently, regardless of the cap. Where we'd likely need to see this is comparing the chips with turboboost disabled and at a fixed clock speed to truly test.

Those tests are all with the clockspeed and ring bus speeds locked.

[Agall]

2 hours ago, YoungBlade said:

Those tests are all with the clockspeed and ring bus speeds locked.

I'm referring to newer generations, where L2 and L3 cache has increased significantly from 10th generation. Example being the 10900k having its L2 cache be 256KB/core and 20MB shared L3 versus 2MB/core and 36MB of shared on the 13900k. The architecture is dramatically different in that regard, where you're also losing less L3. Considering as well that the 13600k has more L3 cache than the 10900k. I would imagine +3MB of L3 cache mattering far less if any in some application compared to +12MB or +64MB when your case is already 26MB for a core.

 

The 10th generation test is nice and all, but I don't think its applicable to 13th/14th generation Intel because of how dramatic the increase in cache has been. Like comparing a 2700x to a 7700x, which has received incremental increases in cache over the generations.