News AMD Ryzen 7000: Up to 16 Cores, AVX-512 Support at Launch

[Admin]

AMD's tech marketing boss does the rounds clearing things up in the wake of Computex confusion. He confirms launch max core configurations, new Threadrippers are on the way, and support for some AVX-512 features.

AMD Ryzen 7000: Up to 16 Cores, AVX-512 Support at Launch : Read more

 

[waltc3]

Robert also said that by the end of summer they'd be releasing a lot more info when they get the final Zen4 configurations nailed down. I had a feeling from the start that AMD was being coy, and keeping a lot of what they already know under wraps for a while. Shaping up to be a fun year, already.

 

[thisisaname]

Robert also said that by the end of summer they'd be releasing a lot more info when they get the final Zen4 configurations nailed down. I had a feeling from the start that AMD was being coy, and keeping a lot of what they already know under wraps for a while. Shaping up to be a fun year, already.

Yes they are being coy but it does not help someone decide weather to get an AM4 or hold out for an AM5 or if going Intel is going to be better.
With all their messing about I feel I may be better off going Intel rather than wait around for AMD to pull their finger out.

 

[SunMaster]

Yes they are being coy but it does not help someone decide weather to get an AM4 or hold out for an AM5 or if going Intel is going to be better.
With all their messing about I feel I may be better off going Intel rather than wait around for AMD to pull their finger out.

It's the start of "the leaking game". It'll be fun for sure - regardless of which camp you feel you're in.

 

[Deleted member 431422]

Higher TBP for ZEN 4 makes sens now. AVX-512 increased power draw for Intel processors.

 

[ottonis]

Yes they are being coy but it does not help someone decide weather to get an AM4 or hold out for an AM5 or if going Intel is going to be better.
With all their messing about I feel I may be better off going Intel rather than wait around for AMD to pull their finger out.

Currently, one cannot really go wrong with either AMD or Intel camp, since both provide more than plenty of performance and options to cope with virtually every task you throw at them.
Sure, people making money with certain software applications - e.g. 3D-rendering or video editing - will certainly want to have a deep look into how well their preferred software is optimized for each platform (multi core? AVX256/512? FP? cash sizes? latencies? Ram throughput?) and base their decisions on hard facts rather than on rumors or promises for the future.

People already running an ADL platform will most likely want to stay with Intel and upgrade with RaptorLake later on.
People already with AMD will have the opportunity to stay with a recent chipset/CPU, or upgrade from an older CPU, or even build a new Zen3 system from scratch and save some good money at that.
AMD cannot reveal every single benchmark of the Zen4 chips, as this would help Intel to adjust their price policies for their upcoming RaprorLake CPUs.

 

[mdd1963]

Higher TBP for ZEN 4 makes sens now. AVX-512 increased power draw for Intel processors.

To be fair, 12900K adeptly managed to consume 230-240 watts even without AVX512...

 

[jeremyj_83]

To be fair, 12900K adeptly managed to consume 230-240 watts even without AVX512...

Even with only have 8 performance cores as well. All 8 E cores only consume like 40W.

 

[Sleepy_Hollowed]

Wow, if they keep ECC memory support (to be implemented by vendors of course on boards), this might be big trouble in the smaller workstation market.

I'd get one of these in a heart beat for my occasional compiling needs.

 

[Makaveli]

Yes they are being coy but it does not help someone decide weather to get an AM4 or hold out for an AM5 or if going Intel is going to be better.
With all their messing about I feel I may be better off going Intel rather than wait around for AMD to pull their finger out.

That is a simple decision if you are doing a new build now it makes no sense to go AM4.

You hold off on AM5 or go intel.

 

[-Fran-]

Related:

View: https://www.youtube.com/watch?v=JnLbqB1FBu8


I think most people should keep an open mind about AMD just setting a baseline of what Zen4 will be and also try to read the fine print and compare Z690 to X670(E).

Regards.

 

[rluker5]

Hopefully I will hear of a reason to enable the AVX512 on my 12700k now. I'm just a normal user so I'm not seeing anything, but maybe the hype machine will find or coerce somebody to make something. I've been holding on to this old bios pretty much since release for that. That and all of my bios profiles would be lost, which would be a hassle to redo.

 

[wifiburger]

don't care ! they should of not talked about Zen4 this early in the stage or include some charts baking up what Robert is spewing around the web !

Making it worst is the lack of communication on how many cpu gen x670 will supports, chipset segmentation, OC segmentation & way too much talk about crap PCIE5 & NVME slots.

I give 0 f about their OC segmentation spanning 3 chipsets; worst than Intel at this point !

 

[jasonf2]

AMD releasing AVX512 is really kind of amusing considering that their stance has been something along the line of "you really don't need that in a consumer chip". Now that Intel has pulled back on it they are pulling an old Intel trick by being able to boast synthetic benchmark counts that really don't relate to real world performance based on AVX512 performance (for most available applications). The positive here is that after all this time AVX512 will finally go mainstream with all of its advantages.

 

[thGe17]

Wow, if they keep ECC memory support (to be implemented by vendors of course on boards), this might be big trouble in the smaller workstation market.

I'd get one of these in a heart beat for my occasional compiling needs.

If you really need ECC on Intel, you can use the W680 chipset for ADL/RPL. With that you have official/validated ECC-support with something like a 12900K.

AMD releasing AVX512 is really kind of amusing considering that their stance has been something along the line of "you really don't need that in a consumer chip". ...

Its always a different story what marketing wants you to believe and what the internal roadmap looks like. AVX512 is badly needed for Epyc and HPC workloads, therefore it was obvious that AMD would implement it at a certain point in time into their microarchitecture.
Intel already showed SPR (w/o HBM2) that outclassed Milan and Milan-X, most likely due to the usage of AVX512.

Hopefully I will hear of a reason to enable the AVX512 on my 12700k now. ...

Currently there is no real need for it on the consumer platform. With the 256 bit VNNI on ADL/RPL,you have the most likely used extension always available fo ML-workloads.
Besides, it remains to be seen if AMD enables or even has implemented full AVX512 subsets as they are currently available on Intel CPUs. Currently it is hard to guess why they avoid to confirm AVX512 support although they confirmed VNNI support.

 

[SunMaster]

AMD releasing AVX512 is really kind of amusing considering that their stance has been something along the line of "you really don't need that in a consumer chip". Now that Intel has pulled back on it they are pulling an old Intel trick by being able to boast synthetic benchmark counts that really don't relate to real world performance based on AVX512 performance (for most available applications). The positive here is that after all this time AVX512 will finally go mainstream with all of its advantages.

Do you have anything to back up your claim about AMDs stance wrt AVX512?
I'm talking about AMDs actual stance, not reviewers or others being vocal about AMDs take or opinion.

I'm very convinced the latest iteration of Intels got it fused off wasn't because Intel is saying this is server territory but there was no room (or power for) the E-cores to be equipped with avx512,

 

[jasonf2]

Do you have anything to back up your claim about AMDs stance wrt AVX512?
I'm talking about AMDs actual stance, not reviewers or others being vocal about AMDs take or opinion.

I'm very convinced the latest iteration of Intels got it fused off wasn't because Intel is saying this is server territory but there was no room (or power for) the E-cores to be equipped with avx512,

If you want a quote from their CEO the answer is no. However the glaring absence of the instruction set which was proposed in 2013 and in production on some (but not all) Intel core products since 2015 does establish that up until now AMD has not looked at it as important enough to implement. As far as Intel removing it from the current alder lake line I don't disagree at all. Intel is in a position that they are doing anything possible to crank out as much performance as possible and AVX-512 simply hasn't been adopted that well in the consumer lines. For it's use case AVX-512 works well but it's scope is somewhat limited. To that end any synthetic benchmark that utilizes it will see score increases on chips that implement it and in the early days Intel posted these benchmarks regularly to reviewers chagrin. However real world performance in most cases won't get anything from it unless your software specifically utilizes it. This is almost exclusively in the server domain. In my opinion the consumer chip market has been looking at game frame rates more than synthetic benchmarks as of late and this is one of those technologies that doesn't typically reflect in those numbers. My gut is telling me though that Intel is going to be forced to reimplement in future core products now that AMD is putting it in products. To those out there thinking there is going to be an Alder Lake microcode update that just turns it back on though I wouldn't hold my breath.

 

[SunMaster]

My gut is telling me though that Intel is going to be forced to reimplement in future core products now that AMD is putting it in products.

That I agree to, and I wonder if it's more of a strategical move than a performance move. There's no doubt that the usefulness of avx512 has been questionable. 'Linus Torvalds: "I Hope AVX512 Dies A Painful Death" ' - I believe he shares the views of many

As the upcoming server and desktop SKUs both use Zen4 cores, and AMD wants their servers to be able to do AVX512 then maybe the 7k-series get it as a "hidden bonus". The only real bonus is perhaps to strip Intel of bragging rights and is of no value to you and me.

 

[TerryLaze]

I'm very convinced the latest iteration of Intels got it fused off wasn't because Intel is saying this is server territory but there was no room (or power for) the E-cores to be equipped with avx512,

They didn't got it fused off, they got it disabled.
Putting AVX into e-cores would be the exact opposite of what e-cores are there for, they are supposed to be as small as possible and use as little energy as possible while giving the most amount of performance.
Anybody still on the old microcode can disable the e-cores and still have AVX.

My gut is telling me though that Intel is going to be forced to reimplement in future core products now that AMD is putting it in products.

Because now that a company that has like 10 or maybe 20% of the market share...what do you expect to happen?!

Also they didn't un-implement it in the first place, it's still in the CPU design.
My gut would tell me that intel will see how the task manager affair will play out, or more possible intel will fix it at some time so that AVX will happily be running alongside the e-cores and at that point they will re enable it with microcode.

 

[SunMaster]

They didn't got it fused off, they got it disabled.

Since you insist on nitpicking - the early batches have them disabled through microcode. For the rest, there's https://www.tomshardware.com/news/intel-nukes-alder-lake-avx-512-now-fuses-it-off-in-silicon

 

[TerryLaze]

Since you insist on nitpicking - the early batches have them disabled through microcode. For the rest, there's https://www.tomshardware.com/news/intel-nukes-alder-lake-avx-512-now-fuses-it-off-in-silicon

If I would be nitpicking it would point out that he says that it is planned...
Do we have numbers, like production runs where it started? Did it already happen? Will it happen at all?

Either way, it's still part of the design of the CPU even if it is fused off and if they so choose they can start producing them again without it being fused off without any issue.

 

[SunMaster]

If I would be nitpicking it would point out that he says that it is planned...
Do we have numbers, like production runs where it started? Did it already happen? Will it happen at all?

Dear god.

https://www.intel.com/content/www/us/en/support/articles/000089918/processors.html

Either way, it's still part of the design of the CPU even if it is fused off and if they so choose they can start producing them again without it being fused off without any issue.

That way you can pretend it never got removed. Good plan. Sounds extremely plausible.

 

[TerryLaze]

Dear god.

https://www.intel.com/content/www/us/en/support/articles/000089918/processors.html

Deer god in deed, "will be fused off on Alder Lake. "
As in, when?!
If it happens a few weeks before the new gen comes out then who cares?!

Raptor lake is supposed to come out in the second half of 2022, so within the next 6 months, if they haven't started fusing it off by now then how many will realistically be fused off, even if they start right now?

The real question is if it will be fused off on raptor from the beginning.