Unofficial AMD Ryzen vs Intel Core Debate Thread!

[Ajay]

The place to post Facts and Friction on the state of the art AMD and Intel CPU platforms.

Please keep it civil!! These are CPU choices, not life and death choices

 

[Markfw]

So why do we need this when we have hundreds of "like" posts in all of the Ryzen threads already ? is this because a few of you never read those threads, and now want it all re-hashed here ?

 

[Ajay]

It's going to be an ongoing debate, as per usual, in CPU and Overclocking. I prefer having threads that have a higher signal to noise ratio for those interested in build a Ryzen system or those interested in Intel. I was hoping this could be a place for us to come together, or split apart

Also, I don't know of anyway to collate posts by whether they are 'liked' or not.

 

[ozzy702]

I like the idea of this thread, hopefully it will keep both the Intel and AMD threads free from clutter.

 

[Markfw]

Wow, so here one you all missed. 1600x beating 6800k @$175 less and on the heels of a 6850k@$358 more......

Yeah to me that's always been the value leader, and my choice from AMD. 6800k performance for $250..

I'm down.. Time to sell some X58 gear

And this is one of HUNDREDS of these Ryzens posts in the other threads.

Wait, in 10 seconds, here is one from the Anandtech review:

I can cherry pick all day, but the bottom line is, this is a serious contender, at far less in price.

And here is another one, beats everything but the $1000 and up Intel chips:

 

[Crono]

This won't end well. People tend to keep it civil enough in the "which CPU should I get?" type threads where there are specific use cases, but that's not likely in an explicit and broad head-to-head Intel vs AMD debate.

 

[cbn]

In other threads I have seen mention that Ryzen is restricted by its memory controller.

Can someone please post some links on how these processors are affected buy that?

 

[Ajay]

This won't end well. People tend to keep it civil enough in the "which CPU should I get?" type threads where there are specific use cases, but that's not likely in an explicit and broad head-to-head Intel vs AMD debate.

You may be right Crono - in which case, Mark will want my head on a spike

 

[itsmydamnation]

Theres not much to debate,

Intel have a core where they haven't progressed direction since haswell but have refined it massively ( can see this in mirco benchmarks of skylake vs haswell). What will really matter is what intel has been doing with its R&D over the last 3 or 4 years. So far we know about new AVX-512 and the cache system needed to feed the beast but have they been putting large resources into the next major mirco-architectural update? It also makes we wonder what is the point of knights-* if that top end 28 core skylake-X can hold 2.0ghz AVX-512 clock its already more power efficient then knights-landing and if given a 300watt TDP would probably still be behind in absolute perf but not far away.

But what all this means for consumer /HEDT i have no idea. What i will find more interesting is what happens after 14nm++, AMD used to run a continual process improvement program during the 90/65/45nm days and the end result was each new process node looked disappointing from an enthusiasts stand point. The expected jump in performance wasn't delivered upfront. IdontCare posted the slides on it in here somewhere a few years ago.


Zen is a completely new core and likely has a lot of low hanging fruit so next iterations should get solid IPC improvements, clocks we will have to wait and see. The big thing for AMD is what they do with SIMD width, do they keep a "lean" 128bit core, keep core bandwidth/size/power requirements lower and add more cores then rely on GPU compute. Do they increase SIMD width and is it to 256 or 512(unlikely i think) at the cost of size/bandwidth/power. Or do they do something "crazy" a bit IBM like and increase load and store width add optional SMT4 and keep the 4 128bit pipes.

Personally I prefer 1 and 3 over 2, there is no point being intel light, AMD have to deliver a fundamentally strong core but they will want to have a different set of strengths and given there market share they should prioritise your typical cloud/enterprise workload not very large vector widths. I think something like option 3 would be good for HEDT/games/consoles.


Thats my high level 2c +VAT/GST.

edit: found it, was it really in 2011 i read this.... rofl https://forums.anandtech.com/threads/can-amd-rescue-the-bulldozer.2210645/page-5#post-32680614

 

[Ajay]

In other threads, I have seen mention that Ryzen is restricted to some degree by its memory controller.

Can someone please post some links on these processors are affected buy that?

IMHO, given how well Ryzen scales with faster memory, I can only believe that AMD wanted to hit higher speeds for DDR4. There has been an implication, which I believe, that the Ryzen IMC, for what ever reason, is gimped in this regard. I've always hoped that the rumored B2 stepping would address this issue somewhat. But that stepping may be coming out after my purchase horizon has passed.

 

[Bouowmx]

Concerning distributed computing, the distributed computing platform is BOINC. BOINC exports a stat called "recent average credit" (RAC) basically denoting an entity's performance in a project. RAC can be for a individual computers, which will be relavent in comparing processors.

There are many BOINC projects; each can utilize different hardware (typically from the set CPU, AMD GPU, Intel GPU, and NVIDIA GPU). To compare CPUs, look in CPU-only projects by checking project's applications: NFS@home is CPU-only, whereas SETI@home is not.

From project's home page, navigate to Statistics, Top computers (or hosts), or directly to top_hosts.php (may be different in some projects): https://escatter11.fullerton.edu/nfs/top_hosts.php. Page will be sorted by RAC. Now compare computers with relavent processors by RAC (higher is better), assuming that each computer had the chance to build up to its maximum RAC, typically after 30 days.

However, there are some nuances. Some projects, like NFS@home, have different applications giving different number of credits. For greater detail, consult a computer's valid tasks: From top computers, click "Tasks" under an individual computer's ID, and then filter "Valid". Filter for a specific application. However, many project's applications are never consistent in run time, keeping all else equal.

Two CPU-only projects I know that have an application with consistent run time are theSkyNet POGS and TN-Grid. I personally run POGS on computer with Intel Xeon E3-1245 and 2×DDR3-1333 9-9-9-24-2T * (click here * to see tasks). Note that tasks giving same number of credits have very similar run times. Since POGS only has one application, I will compare computers simply by RAC. Searching POGS top computers, I see AMD Ryzen 7 1700 with ~36k RAC and Intel Core i7-6700 with ~25k RAC.

* POGS can take a while to load stats, as if the hard drive containing the data was off and is just spinning up to speed.

Relevant derived metrics of CPUs in distributed computer can be RAC-to-initial price, RAC-to-power, or RAC-to-ownership cost (initial price plus electricity cost).

 

[Rifter]

There really isnt much to debate. Its basically boils down to if you dont have unlimited funds get a ryzen. If you do have unlimited funds intels top tier CPU's still hold the higher performance crown.

I expect this to continue with SL-X but who knows intel could surprise me and drop prices, time will tell.

 

[formulav8]

is this because a few of you never read those threads,

Some people doesn't seem to want to admit AMD put a big black eye on Intel's face, and even want to finally admit the latest Intel Q/A price cuts coming soon from Intel is because of AMD forcing them. I never understand fandango's. Even when they get ripped like everyone else. But some seem to be Intel stooges.

 

[formulav8]

intel could surprise me and drop price

They are dropping prices from what I've seen. I think is was PCWorld as well as an Intel CEO Q/A. Of course the Intel haughtiness would not outright admit AMD is part of the cause.

 

[Markfw]

In other threads I have seen mention that Ryzen is restricted by its memory controller.

Can someone please post some links on how these processors are affected buy that?

This is a negative take on reality.

Ryzen speed is highly dependent on memory speed as the cores talk to each other at that speed. So when you up the memory speed, the system performs better. I use PC3200 CAS14, Ryzen also likes low latency.

Bottom line ? the faster you memory is, the better your system performs.

And the upcoming (I have heard) 8 channel server chips from AMD should just kick hiney if they are linked to good memory.

 

[raghu78]

Theres not much to debate,

Intel have a core where they haven't progressed direction since haswell but have refined it massively ( can see this in mirco benchmarks of skylake vs haswell). What will really matter is what intel has been doing with its R&D over the last 3 or 4 years. So far we know about new AVX-512 and the cache system needed to feed the beast but have they been putting large resources into the next major mirco-architectural update? It also makes we wonder what is the point of knights-* if that top end 28 core skylake-X can hold 2.0ghz AVX-512 clock its already more power efficient then knights-landing and if given a 300watt TDP would probably still be behind in absolute perf but not far away.

But what all this means for consumer /HEDT i have no idea. What i will find more interesting is what happens after 14nm++, AMD used to run a continual process improvement program during the 90/65/45nm days and the end result was each new process node looked disappointing from an enthusiasts stand point. The expected jump in performance wasn't delivered upfront. IdontCare posted the slides on it in here somewhere a few years ago.


Zen is a completely new core and likely has a lot of low hanging fruit so next iterations should get solid IPC improvements, clocks we will have to wait and see. The big thing for AMD is what they do with SIMD width, do they keep a "lean" 128bit core, keep core bandwidth/size/power requirements lower and add more cores then rely on GPU compute. Do they increase SIMD width and is it to 256 or 512(unlikely i think) at the cost of size/bandwidth/power. Or do they do something "crazy" a bit IBM like and increase load and store width add optional SMT4 and keep the 4 128bit pipes.

Personally I prefer 1 and 3 over 2, there is no point being intel light, AMD have to deliver a fundamentally strong core but they will want to have a different set of strengths and given there market share they should prioritise your typical cloud/enterprise workload not very large vector widths. I think something like option 3 would be good for HEDT/games/consoles.


Thats my high level 2c +VAT/GST.

edit: found it, was it really in 2011 i read this.... rofl https://forums.anandtech.com/threads/can-amd-rescue-the-bulldozer.2210645/page-5#post-32680614

excellent post. btw i agree that AMD should not chase Intel's design and keep Zen successors lean with 128 bit FP units. For vector workloads use the massively parallel GPU and thats where their server APUs with Zen and Vega fit in. AMD should target maximum integer compute/socket with traditional Zen servers and go for massive APUs with high bandwidth cache using HBM2 for HPC/vector workloads.

 

[frozentundra123456]

A glutton for punishment aren't you Ajay?Aren't there enough threads already taken over by smacktalking AMD fanboys extolling how poor abused AMD is going to stick it to the evil giant intel? Do you really need to give them another opportunity?

 

[Ajay]

There really isnt much to debate. Its basically boils down to if you dont have unlimited funds get a ryzen. If you do have unlimited funds intels top tier CPU's still hold the higher performance crown.

I expect this to continue with SL-X but who knows intel could surprise me and drop prices, time will tell.

Well, that why people like me are desperate for some info on Skylake-X pricing. It's either a win or a total deal breaker. Personally, I'd rather have 6 cores with higher ST performance than 8 cores with less, but if SKX pricing sucks - it may be a R7 1700 for me (in which case, I wish the damn things overclocked ~10% higher). I think the Ryzen platform will be pretty solid then, though I will have to pay more for the privileged of fast DDR4 3200 RAM. Oh, and a pox on smart phone companies for eating up DDR4 dies (j/k).

 

[Ajay]

A glutton for punishment aren't you Ajay?Aren't there enough threads already taken over by smacktalking AMD fanboys extolling how poor abused AMD is going to stick it to the evil giant intel? Do you really need to give them another opportunity?

Anything to keep the Skylake thread clean .

 

[frozentundra123456]

Anything to keep the Skylake thread clean .

Too late for that. The AMD lobby is already busy threadcrapping it.

 

[itsmydamnation]

Too late for that. The AMD lobby is already busy threadcrapping it.

there is a whole like 10 years of payback to be had

 

[esquared]

Closed, while the mods discuss this thread.


esquared
Anandtech Forum Director