Soldato
Or, shock horror, do what Intel did with Alder Lake and just make better cores. Which is what they're doing, more shock horror.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD what you doing to fight off Alderlake?
- Thread starter subbytna
- Start date
You can't magic huge MT gains without more cores though - I think that's what he's referring to
Quick example, compare leaked benchmarks for next gen Threadripper vs several years old 39xx TR - the gains are less than 10%. The quickest way to improve MT is more cores
Can you count?
Yeah and the fact zen 4 will be competing against raptor lake where intel will increase MT further with more small cores so unless Zen 4 increases cores or adds SMT x4 I don't see how it will compete in MT at the lower end given that intel now already have a huge lead in that area.
Intel don’t have any lead.
Maybe you should go check out the reviews rather than spouting nonsense on here.
Maybe you should do more counting.
You sure?
Thanks for posting this. It's very different to the IPC results shown here:
https://www.guru3d.com/articles_pages/core_i5_12600k_processor_review,7.html
Which of these results do you consider to be more representative of single core CPU performance, per clock? I am something of a Cinebench newb...
R20 is newer, so are these results more relevent?
Last edited:
AMD what you doing to fight off Alderlake?
eating Nvidia's lunch as well as Intels
https://www.servethehome.com/amd-ceo-teases-the-oam-accelerator-era-on-twitter/
eating Nvidia's lunch as well as Intels
https://www.servethehome.com/amd-ceo-teases-the-oam-accelerator-era-on-twitter/
Caporegime
They are both accurate but as you say R20 is much newer and is using different extensions, like AVX 1 vs AVX 2 and ADL is better than Zen 3 at AVX 2 loads, don't know if that's true just using it as an example.
Caporegime
There is another aspect, Cinebench is a Cinema 4D benching utility, its the same sort of application as Blender, for whatever reason Intel does much better in Cinebench vs AMD than it does in Blender, Same workload different aplication.
There is a historical aspect to this, in the past Maxcomm, the makers of Cinebench (Cinema 4D) have added code that deliberately gimped performance on AMD's CPU's, at Intel's request, i have absolutely no idea if there is something fishy going on, again, but i doubt it given that AMD's performance in R20 and R23 is unchanged from before ADL, there could however be some collaboration between Maxcomm and Intel to make ALD very optimised for it, and i honestly don't think there is anything wrong with that
There is a historical aspect to this, in the past Maxcomm, the makers of Cinebench (Cinema 4D) have added code that deliberately gimped performance on AMD's CPU's, at Intel's request, i have absolutely no idea if there is something fishy going on, again, but i doubt it given that AMD's performance in R20 and R23 is unchanged from before ADL, there could however be some collaboration between Maxcomm and Intel to make ALD very optimised for it, and i honestly don't think there is anything wrong with that
Associate
The denial is strong in this thread.
AMD what you doing to fight off Alderlake?
eating Nvidia's lunch as well as Intels
https://www.servethehome.com/amd-ceo-teases-the-oam-accelerator-era-on-twitter/
Nvidia is fine, they're powering the Metaverse and will have 5 billion people grovelling at their feat
There is another aspect, Cinebench is a Cinema 4D benching utility, its the same sort of application as Blender, for whatever reason Intel does much better in Cinebench vs AMD than it does in Blender, Same workload different aplication.
There is a historical aspect to this, in the past Maxcomm, the makers of Cinebench (Cinema 4D) have added code that deliberately gimped performance on AMD's CPU's, at Intel's request, i have absolutely no idea if there is something fishy going on, again, but i doubt it given that AMD's performance in R20 and R23 is unchanged from before ADL, there could however be some collaboration between Maxcomm and Intel to make ALD very optimised for it, and i honestly don't think there is anything wrong with that
and this is why in my first post for this thread I had a laugh at g. You shouldn't measure IPC in a single application.
Both amd and intel measure their IPC using a suite of 30 different applications because that's how you should do it
This one don't even show what the score is so seem a bit odd and again to use R15 which no one uses.
Soldato
- Joined
- 31 May 2009
- Posts
- 21,257
One thing regarding sc versus mc perf, obviously when speaking of adl one only speaks of ipc of the big core, adding extra e cores to help multi is a misnomer with regards ipc. Any extra cores on amd will use the full ipc. Intel at this stage can't do so due to power overload, as seen from any 12900 stress test, they are truly on the absolute limit.
The 125watt results, or the 12700kf results seem to suggest they really lose efficiency in that final push to win benches.
The cb20 at 4 is interesting as it doesn't stress any of the actual workings of the adl processor as it will be used in any fashion.
I'd like to see further ipc investigation.
Soldato
Of course you can if the cores themselves are better, or intercore latency is improved, or clock speeds go up. Are you/he suggesting that the 15% performance boost V cache has already demonstrated on Ryzen 5000 is actually not true? Will the big IPC bump due with Zen 4 actually net 0% uplift?
Alder Lake didn't become competitive in MT workloads just by slapping more cores on, Alder Lake's underlying architectures have seen great improvement. And the same is coming for Ryzen 5000 with the V cache refresh and then with Zen 4 end of next year. And also Raptor Lake with its minor improvements to P cores and double the E cores.
But the entire claim of oh AMD will have to put a bazillion more cores on just to keep up" is so superficial it's bogus, almost smells of a fundamental lack of understanding. Look at the 12900K vs 5950X for example. Does the 12900K take the lead? Yes. Does it take the lead by more than 15%? No. Do AMD need to add "moar cores" to regain that MT lead? No, V cache refresh will handle that early next year.
And then what about Raptor Lake? 10% bump on P cores, doubled E cores? 24c/32t 13900K goes (let's say) 30% over the 5950X V cache. That's awesome. But what if AMD can get 30% more out of the Zen 4 arch alone? That's parity without changing core counts at all.
So no, it's not cut and dry and as simple as Joxeon would like it to be. Intel have pulled an absolute blinder here to overhaul their major deficits in a single generation and are now back fighting hard. Alder Lake is great, no doubt, but it does raise questions about future scalability. And yet AMD don't have to do all that much right now to counter Alder Lake, and may not have to do too much to contend with Raptor Lake either.
Would need to be a hefty bump considering the MT performance from zen 2>3 was around 15% despite an IPC increase of 23% so zen 3 >4 would need atleast 40% to keep pace with Intel on the Sku's further down the stack as right now the 5600X and 5800X are a long way behind their competing Intel Sku's in MT performance and remember raptorlake will further increase MT performance by doubling the small cores.
AMD what you doing to fight off Alderlake?
eating Nvidia's lunch as well as Intels
https://www.servethehome.com/amd-ceo-teases-the-oam-accelerator-era-on-twitter/
Ouch. I think what AMD might do to counter Alder lake is wreck them.
Caporegime
The extra Cache will only make a difference in Floating Point workloads, which is not everything but its quite a lot of things.
It will double the L3 Cache throughput from 1,000GB/s to 2,000GB/s, which by its self would not matter so much because right now its 64MB divided between two CCD's, instructions only execute on one CCD at a time, so in a 32MB Cache, which is why despite the 5900/5950X having 1,000GB/s throughput vs 600GB/s on a 5800X (32BM) there is no performance difference.
The 3D Cache actually triples the L3 available per CCD, its an extra 64MB per CCD, a total of 96MB on each CCD and that's why it will make a significant difference, the branch predictor for each execution has a huge amount more capacity, as does L2 to L3 victimisation.
If anything i think AMD could be sandbagging their performance claims a bit, perhaps not for gaming, or much, but some workloads outside of that could see a huge up tick.
Right now AMD don't need to do anything more to combat Intel, and Zen 4 will be a brand new core, again, and they will have a lot more power to play with as they lose the 142 Watt PGA limits.
It will double the L3 Cache throughput from 1,000GB/s to 2,000GB/s, which by its self would not matter so much because right now its 64MB divided between two CCD's, instructions only execute on one CCD at a time, so in a 32MB Cache, which is why despite the 5900/5950X having 1,000GB/s throughput vs 600GB/s on a 5800X (32BM) there is no performance difference.
The 3D Cache actually triples the L3 available per CCD, its an extra 64MB per CCD, a total of 96MB on each CCD and that's why it will make a significant difference, the branch predictor for each execution has a huge amount more capacity, as does L2 to L3 victimisation.
If anything i think AMD could be sandbagging their performance claims a bit, perhaps not for gaming, or much, but some workloads outside of that could see a huge up tick.
Right now AMD don't need to do anything more to combat Intel, and Zen 4 will be a brand new core, again, and they will have a lot more power to play with as they lose the 142 Watt PGA limits.