[News/Rumor] AMD might not release a hexacore Ryzen

[StyM]

http://www.guru3d.com/news-story/amd...ore-ryzen.html

Alright, this is a bit of a speculative posting, however the thesis behind it sounds plausible. According to new rumors AMD will not release hexacore (6 core) processors or even dual-core processors. AMD will release an 4 and 8-core model with and without hyper-threading.

The info comes from Thai based zolkorn. The idea here is that AMD only can release 4- and 8-core models both with and without hyper-threading. The explanations is as follows, the Ryzen processors are build out of partitions that each have 4 physical cores in one module. Traditionally they had a dual-core setup so an 8-core version had four of these which could be activated and disabled per module.

Now it might very well be that AMD can somehow disable laser-cut two cores in a 4 core module, no idea there. But it also could very well be that there will be no 2 and 6 core models as to date there has not been any mention such models from the AMD side.

One thing is certain, we'll see four and eight core models with SKU differences in HT, clock frequencies and TDPs. At least all mainstream and higher specced parts will get unlocked multipliers. At the end of this month more info should become available with a launch in March.

[AliG]

Guru3d + the fact they admit it's pure speculation = 0% confidence.

Making a cut-down 6 core allows AMD to compete with the cheaper Extreme processors without actually developing anything new. It makes too much business sense to keep it.

[JPQY]

Would like to see more a 16cores or higher for desktop..
In 2011 we get told to see 80cores..and where are we now in 2017..it goes too slow for me..or they just holding back..

JP.

[StAndrew]

Guru3d + the fact they admit it's pure speculation = 0% confidence.

Making a cut-down 6 core allows AMD to compete with the cheaper Extreme processors without actually developing anything new. It makes too much business sense to keep it.

I agree but I disagree. It would be easy to cut out some cores but I think they pretty much have all performance levels covered with the HT or no HT option: 4 core w/ no HT, 4 Core + HT, 8 Core w/ no HT, and 8 Core + HT. I don't think they really need to add a 6 core in there. Its easier to scale this way than cut cores; however they might have a 6 core processor later down the line to keep yield numbers up. If we see a six core solution early, we can assume yields are not great?

[xlink]

This DOES make sense.

if HT improves multithreaded performance by a third...

4*1.33 = 5.33
8*1.00 = 8.0
8*1.33 = 10.66

These are fairly even steps. Even if HT is a dash slower it's alright.
Also note that there's relatively little room for 6core-no HT in such a paradigm since it'd be so close to 4-core with HT.

[StAndrew]

This DOES make sense.

if HT improves multithreaded performance by a third...

4*1.33 = 5.33
8*1.00 = 8.0
8*1.33 = 10.66

These are fairly even steps. Even if HT is a dash slower it's alright.
Also note that there's relatively little room for 6core-no HT in such a paradigm since it'd be so close to 4-core with HT.

Yeah, I think the only reason to release 6 cores is because 8 cores take too much power (ie Gulftown, but we're way beyond that) or to increase yields; but even then bad 8 core yields can become 4 cores

We may see some mid range muscling between AMD and Intel that may produce 6 core procs to fill minute gaps in the performance spectrum (much like we see with NVIDIA and AMD).

[xlink]

Yeah, I think the only reason to release 6 cores is because 8 cores take too much power (ie Gulftown, but we're way beyond that) or to increase yields; but even then bad 8 core yields can become 4 cores

We may see some mid range muscling between AMD and Intel that may produce 6 core procs to fill minute gaps in the performance spectrum (much like we see with NVIDIA and AMD).

My wag is that a 6 core, no HT model would end up right around a 4-core + HT chip in terms of performance... but with worse power characteristics.
My guess is that you'd end up with an 8 core chip with 20% of the power draw being cut by disabling 2 cores and a 10% reduction to thermal output due to no HT... so around ~70% of the full product.
Meanwhile a 4core chip with HT will have around 50% of the power draw and 88% of the performance... making the 4core+HT version around 20% more power efficient.

All of these numbers are ballpark guesses. Take them with a huge grain of salt. They're NOT informed by anything more than intuition.


As you suggested, I see 6 core chips becoming a way of improving yields... later. 6c-12t might be a good weapon against something intel puts out later.

[StAndrew]

I think I see what you are saying. And 6 core + HT will probably overlap the 8 core none HT.

[xlink]

I think I see what you are saying. And 6 core + HT will probably overlap the 8 core none HT.

assuming 33% multithreaded performance improvements...

6* 1.33 = 8
8* 1.00 = 8


yes, they'll be VERY close.
Now, if you have a 7 threaded task the 8 core chip WILL be a little faster but... that'll pretty much never happen.


I once again would like to note that I'm pulling 33% out of my rear and that it's being chosen because 1) close enough to intel's 30% 2) it makes for REALLY nice numbers mathematically (4C + HT + 10% higher clocks ~= 6C; 6C + HT ~= 8C; ALSO the relative difference between configurations is roughly constant if clocks are held equal)

[zanzabar]

Yeah, I think the only reason to release 6 cores is because 8 cores take too much power (ie Gulftown, but we're way beyond that) or to increase yields; but even then bad 8 core yields can become 4 cores

We may see some mid range muscling between AMD and Intel that may produce 6 core procs to fill minute gaps in the performance spectrum (much like we see with NVIDIA and AMD).

didnt they have the l3 link connect to 2 modules, i would bet if you are going to have a flaw it would be in the l3 thanks to using the close transistor style lithography like gpus typically use.

[AliG]

I take more stock in the rumors of the 4 core not having HT enabled to help further separate their lineup.

You gotta remember it's more profitable to rebadge 8-cores that are too power hungry into 6-cores than it is to create a robust 4 core. I would wager with about 90% certain the 6-cores will have SMT enabled to directly compete with the 6-core Skylake-E processors.

[StAndrew]

didnt they have the l3 link connect to 2 modules, i would bet if you are going to have a flaw it would be in the l3 thanks to using the close transistor style lithography like gpus typically use.

I've never seen any details on the architecture tbh so not sure.

I take more stock in the rumors of the 4 core not having HT enabled to help further separate their lineup.

You gotta remember it's more profitable to rebadge 8-cores that are too power hungry into 6-cores than it is to create a robust 4 core. I would wager with about 90% certain the 6-cores will have SMT enabled to directly compete with the 6-core Skylake-E processors.

Sure, I could see that as well. Time (and a release) will tell.

[drmrlordx]

FWIW it looks like CanardPC is parroting AMD marketing info which indicates 5 Ryzen SKUs at launch:

2 8c/16t SKUs
1 8c/8t SKU
1 4c/8t SKU
1 4c/4t SKU (?!?!?)

No 6c/12t SKUs.

[MirageSys]

Some of the low end harvested SKU's may be OEM & Foreign only. At least that is my suspicion based off of what has happened in the past.

[AliG]

FWIW it looks like CanardPC is parroting AMD marketing info which indicates 5 Ryzen SKUs at launch:

2 8c/16t SKUs
1 8c/8t SKU
1 4c/8t SKU
1 4c/4t SKU (?!?!?)

No 6c/12t SKUs.

I guess I have to say I'm really surprised by this. Either AMD is extremely confident in their manufacturing process, or the 4-core is the cutdown part (which would actually immensely boost their yields).

[drmrlordx]

Well it looks like CanardPC is contradicting itself today so it's hard to say what they're really going to do. But word is that all SKUs will be from the same die, with cores/cache disabled where necessary.

The base die will feature 2 CCXs.

[AliG]

Well it looks like CanardPC is contradicting itself today so it's hard to say what they're really going to do. But word is that all SKUs will be from the same die, with cores/cache disabled where necessary.

The base die will feature 2 CCXs.

I just don't see any value in an 8C/8T CPU. Why wouldn't you want 6C/12T instead? SMT doesn't scale perfectly, but we're talking 4 extra threads vs 2 extra cores.

Even if they all come from the same die, I still think it's more reasonable to expect 8C/16T, 6C/12T, 4C/8T and 4C/4T. That setup maximizes manufacturing efficiency and allows for the most separation across the product lineup.

[StAndrew]

I just don't see any value in an 8C/8T CPU. Why wouldn't you want 6C/12T instead? SMT doesn't scale perfectly, but we're talking 4 extra threads vs 2 extra cores.

Even if they all come from the same die, I still think it's more reasonable to expect 8C/16T, 6C/12T, 4C/8T and 4C/4T. That setup maximizes manufacturing efficiency and allows for the most separation across the product lineup.

I guess it will come down to how well the architecture supports it. You might wind up with an unbalanced processor if you just disable 2 cores

[AliG]

I guess it will come down to how well the architecture supports it. You might wind up with an unbalanced processor if you just disable 2 cores

I guess the only way I can see that being true is if the scheduler is extremely hard coded for 8 cores. But if that were the case I would assume 4 cores would be just as problematic.

Otherwise you simply just restrict the way it addresses computation tasks to ALUs and AGUs in cores 0-5 (as opposed to 0-7).

[drmrlordx]

I just don't see any value in an 8C/8T CPU.

Only reasons I see are:

a). They might have a non-trivial number of CPUs that don't clock very well with SMT active
b). 8c/8t would compete very nicely with Intel's i5-7600k (4c/4t).

[zanzabar]

i have a 5930k and run a plex server and sling dvr on it while also gaming. it is great to have those extra cores. i am not usre how the mass market appeal is but i feel like if you have a desktop still you might also do some sort of home server as well and might want to use the server while gaming.

[AliG]

i have a 5930k and run a plex server and sling dvr on it while also gaming. it is great to have those extra cores. i am not usre how the mass market appeal is but i feel like if you have a desktop still you might also do some sort of home server as well and might want to use the server while gaming.

But why would you want 8C/8T over 6C/12T? No one denies extra threads can't be useful; I'm just saying a 6 core cutdown processor makes far more sense from both a performance and manufacturing perspective.

Eitherway, the other thread suggests that this news was mostly poppy and that the initial lineup will feature a 3.3 GHz 6C/12T SKU, and 3.4 + 3.6 GHz 8C/16T SKUs. That 6C clockspeed definitely suggests it being a cutdown processor that didn't make the voltage leakage requirements.

[drmrlordx]

It does look like 6c/12t is happening instead of 8c/8t. There was a leak today showing an alleged breakdown of Ryzen chips, with 4 6c/12t SKUs (two "Pro" and two not) with price points aimed squarely at the i5-7600k.

So while it does look like AMD is going to try to maul the i5s with superior core/thread count, they're going to use SMT to do so.

[dasa]

maybe cpu without split threads perform better in games due to there pore multithreading
since consoles have 8 cores i doubt we will see any games that can make good use of more than 8 cores till the cpu is so old it wont really matter