AMD/Intel Processors

[Budgetgamer04]

I have been following the battle between AMD and Intel for about a year, and I've noticed that AMD processors tend to be more for workstation PCs while Intel processors tend to lean more towards gaming. I would like to know what makes something like the Ryzen 7 2700x worse in gaming than the Core I9 9900k, as I don't have enough knowledge about the two  to see much of a difference that would affect performance

Edited March 13, 2019 by Budgetgamer04
Grammar

[Mister Woof]

As of this moment:

 

1. Single core performance is better on Intel

2. Frequency is higher on intel

3. Ringbus is better for gaming than infinity fabric 

 

This only matters really when you are not GPU bound. If your GPU is maxed out and you have a faster CPU, it doesn't matter.

[KarathKasun]

3. Ringbus

Speculation.

 

1. and 2.

Both the same thing.  Per MHZ performance is just about equal at this point.  So the only meaningful difference is frequency.

 

Supposedly AMD will be at frequency parity when the new 7nm Ryzen CPUs launch.

[Jurrunio]

1. Lower single thread performance on Ryzen. 2700X only gets 180cb in Cinebench R15 single thread for example (4.35GHz), while 9900k at 5-5.1GHz regularly do 200-225cb.

 

2. Higher latency inside the CPU and memory latency on Ryzen. Some AIDA64 cache and memory benchmark screenshots should tell you that with numbers.

 

3. Lower multicore performance on Ryzen. This is a minor thing and doesnt apply on 8600k/9600k versus Ryzen 5 2600X, but 9900k does 2000-2100cb in Cinebench R15 multi thread easy while Ryzen 7 2700X tops at around 1900cb

[Mister Woof]

7 minutes ago, KarathKasun said:

3. Ringbus

Speculation.

 

1. and 2.

Both the same thing.  Per MHZ performance is just about equal at this point.  So the only meaningful difference is frequency.

 

Supposedly AMD will be at frequency parity when the new 7nm Ryzen CPUs launch.

Ryzen 1 is about on par with Haswell in IPC IMO, with Ryzen + slightly behind skylake.

 

Frequency is separate because it applies to not just single core tasks, but multicore tasks too. Some games care about multithreaded perf. and some just dont care.

 

The intercore latency as described above is what has been observed to make a different in gaming.

[TahoeDust]

5 minutes ago, KarathKasun said:

3. Ringbus

Speculation.

Core to core latency being better with ringbus and better for gaming is a fact, not speculation. 

[KarathKasun]

6 hours ago, Jurrunio said:

1. Lower single thread performance on Ryzen. 2700X only gets 180cb in Cinebench R15 single thread for example (4.35GHz), while 9900k at 5-5.1GHz regularly do 200-225cb.

 

2. Higher latency inside the CPU and memory latency on Ryzen. Some AIDA64 cache and memory benchmark screenshots should tell you that with numbers.

 

3. Lower multicore performance on Ryzen. This is a minor thing and doesnt apply on 8600k/9600k versus Ryzen 5 2600X, but 9900k does 2000-2100cb in Cinebench R15 multi thread easy while Ryzen 7 2700X tops at around 1900cb

With those benches, Ryzen is actually scaling to more cores better than intel.

 

In MT 9900k should be getting ~25% higher scores compared to Ryzen as per the ~25% difference in ST.  However, the gap closes when all threads are used.

 

If this is still the case with 7nm Ryzen, Intel is in for a very painful 2H of 2019.

 

6 hours ago, TahoeDust said:

Core to core latency being better with ringbus and better for gaming is a fact, not speculation. 

Ringbus has worse latency as core count rises.

 

The signal has to be re-transmitted, resulting in significantly worse latency.  Intel has a benefit due solely to the fact that they have it clocked sky high compared to IF.

[Mister Woof]

1 minute ago, KarathKasun said:

Ringbus has worse latency as core count rises.

 

And infinity fabric is slower even with just 2 CCXs

 

9900k even with 8 cores isn't experiencinf latency increases if we look at benchmarks

[KarathKasun]

2 minutes ago, Plutosaurus said:

 

And infinity fabric is slower all the time

 

9900k even with 8 cores isn't experiencinf latency increases if we look at benchmarks

Except it does in very specific cases.  Also, realize that ring clock is 3ghz+ compared to ~1600 for IF.

 

Latency is just a function of how the interface is clocked, and IF is significantly faster per clock.  Expect it to get faster with optimization and process refinement where the ring bus is already close to the limits of what the silicon can handle.

[GoldenLag]

4 minutes ago, KarathKasun said:

Ringbus has worse latency as core count rises.

doesnt really matter untill you go above 10 cores. 

 

at that point its not better than say meshbus

[KarathKasun]

2 minutes ago, GoldenLag said:

doesnt really matter untill you go above 10 cores. 

 

at that point its not better than say meshbus

Remember, each memory controler is a stop on the ring as well.  There is a reason Intel is moving away from the ringbus.

 

IF = mesh if ya didnt know.

[GoldenLag]

3 minutes ago, KarathKasun said:

Except it does in very specific cases.  Also, realize that ring clock is 3ghz+ compared to ~1600 for IF.

 

Latency is just a function of how the interface is clocked, and IF is significantly faster per clock.  Expect it to get faster with optimization and process refinement where the ring bus is already close to the limits of what the silicon can handle.

IF clockspeed is locket to the speed of the memmorybuss, which is why faster memmory helps. Zen 2 is said to possibly have a unlocked IF clockspeed

[TahoeDust]

2 minutes ago, GoldenLag said:

doesnt really matter untill you go above 10 cores. 

 

at that point its not better than say meshbus

My 9900k latency smokes my 7820x.  The 7820x had more memory bandwidth.  Clock for clock, gaming/gpu benchmarks favor the 9900k hands down.

[Mister Woof]

1 minute ago, KarathKasun said:

Except it does in very specific cases.  Also, realize that ring clock is 3ghz+ compared to ~1600 for IF.

 

Latency is just a function of how the interface is clocked, and IF is significantly faster per clock.  Expect it to get faster with optimization and process refinement where the ring bus is already close to the limits of what the silicon can handle.

 

Specifically, we are talking gaming, and all the benchmarks I've seen show the 9900k doing as good or better than every other Coffeelake at equal frequency.

 

Ryzen has had two years to sort it out, and as of now it isn't as fast at gaming. Maybe further refinements or Zen 2 will change that, but I did say "at this moment"

 

 

[KarathKasun]

5 hours ago, GoldenLag said:

IF clockspeed is locket to the speed of the memmorybuss, which is why faster memmory helps. Zen 2 is said to possibly have a unlocked IF clockspeed

Yup, if the mesh clock is decoupled from RAM and can be pushed up to 2ghz+ reliably the primary inter-core latency difference between the two architectures will mostly go away.

 

5 hours ago, Plutosaurus said:

Specifically, we are talking gaming, and all the benchmarks I've seen show the 9900k doing as good or better than every other Coffeelake at equal frequency.

 

Ryzen has had two years to sort it out, and as of now it isn't as fast at gaming. Maybe further refinements or Zen 2 will change that, but I did say "at this moment"

These current differences mean nothing with Zen2.  IF is getting a MAJOR overhaul.  Primarily due to the fact that the memory/PCIe complex is divorced from the CPU itself.

[GoldenLag]

Just now, KarathKasun said:

IF = mesh if ya didnt know.

im reffering to the intel mesh bus that is actually known as meshbus. its not the same as IF. 

 

IF isnt perfect, and latency is "meh", but its vastly better than any other die interconnect. 

 

1 minute ago, KarathKasun said:

Remember, each memory controler is a stop on the ring.  There is a reason Intel is moving away from the ringbus.

they arent really moving away from ringbus. they are still using it. Intel just uses 2 different architectures between consumer and HEDT. 

 

ringbuss is going to continue. at 10 cores its still pretty fantastic, and 12 core isnt great its still pretty good. and clever sqedualling and cache managing helps against any slowdowns. 

 

intel is node limited and modularity limited.

[Mister Woof]

1 minute ago, KarathKasun said:

These current differences mean nothing with Zen2.  IF is getting a MAJOR overhaul.

And?

 

I said "As of this moment"

 

We don't know how well Zen 2 will perform.

[TahoeDust]

5 minutes ago, Plutosaurus said:

We don't know how well Zen 2 will perform.

Let's hope it is not as big of a disappointment as Radion VII....

[KarathKasun]

1 minute ago, GoldenLag said:

im reffering to the intel mesh bus that is actually known as meshbus. its not the same as IF. 

 

IF isnt perfect, and latency is "meh", but its vastly better than any other die interconnect. 

 

they arent really moving away from ringbus. they are still using it. Intel just uses 2 different architectures between consumer and HEDT. 

 

ringbuss is going to continue. at 10 cores its still pretty fantastic, and 12 core isnt great its still pretty good. and clever sqedualling and cache managing helps against any slowdowns. 

 

intel is node limited and modularity limited.

Thought they had something other than topology"bus" names for them. XD

 

The ring will be going away if they are forced above 10 cores in the mainstream.

[Mister Woof]

Just now, TahoeDust said:

Let's hope it is not as big of a disappointment as Ryzen VII....

I've got big hopes.

 

I, too, want a 5ghz 8/16 for $200.

[KarathKasun]

Just now, TahoeDust said:

Let's hope it is not as big of a disappointment as Ryzen VII....

Technically different companies, so there is promise.

[GoldenLag]

2 minutes ago, TahoeDust said:

My 9900k latency smokes my 7820x.  The 7820x had more memory bandwidth.  Clock for clock, gaming/gpu benchmarks favor the 9900k hands down.

hence at lower corecounts its not prefferable. 

2 minutes ago, KarathKasun said:

Yup, if the mesh clock is decoupled from RAM and can be pushed up to 2ghz+ reliably the primary inter-core latency difference between the two architectures will mostly go away.

um, i dont think we will ever see Zen 2 beating a 6 core ringbus in latency, regardless of how much you can clock the IF. 

2 minutes ago, KarathKasun said:

These current differences mean nothing with Zen2.  IF is getting a MAJOR overhaul.

WE DONT KNOW ANYTHING FOR CERTAIN. if you hype up a product you will be dissapointed, we dont know what Zen 2 is actually capable of. 

1 minute ago, TahoeDust said:

Ryzen VII....

yes, because that is a real product

 

 

 

 

OP: intel has the edge due to clockspeed, which is the main contributing factor. even though they have the best product doesnt mean they are the go to product for every budget. we are straying a bit of topic so i thought id mention it

[TahoeDust]

Just now, Plutosaurus said:

I, too, want a 5ghz 8/16 for $200.

Lol....everyone does!  But if people think AMD is going to sell a chip that truely trade blows with the 9900k for a third the price, well...

[Mister Woof]

1 minute ago, TahoeDust said:

Lol....everyone does!  But if people think AMD is going to sell a chip that truely trade blows with the 9900k for a third the price, well...

Zen 2 just needs to deliver on the 6/12 and 8/16 models with good frequency and low prices. They don't have to even beat the 8700k or 9900k, but be at parity.

 

12 and 16 core chips are overkill for 99% of users. 

[TahoeDust]

1 minute ago, GoldenLag said:

yes, because that is a real product

Sorry, I think everyone knew what I meant.  But damn, you got me good.