XFR Extended Frequency Range; Incomplete WIP?

Does anyone really think that AMD spent time, money, and promotion on a feature that will only give a boost of 80Mhz to a single core? I'm of the opinion that XFR isn't even close to complete yet, and that XFR is one of the interesting concepts of this new architecture. My conjecture is in part due to the revelation(if true) that the 1400X has a higher XFR Boost of up to 200Mhz doubling that of the 1800X.

What that tells me is that XFR is a parameter that can be adjusted by AMD through code updates.
We know that XFR is engaged by detection of thermal headroom, this headroom has been artificially reduced by the 20 degree Celcius off-set reported by the sensors on X chips. What would happen if this off-set was removed? It would seem likely that XFR Boosts would occur more often.

Does manual overclocking disable XFR, or would XFR Boosts then be applied on top of a manual 24/7 OC due to 20C more headroom found?

Will AMD in the near future enable XFR on each core?

Will AMD remove the limitations on XFR Boosts to only that imposed by the Silicon Lottery?

There are other questions I'd like answers to, but for now I'm very interested in what the community's thoughts are.

What do you think?

We will have to wait and see if AMD does decide to do this. All they would realistically need to do is ask their R and D team how much of a boost can they add to the chip with stock cooler.

That would be fantastic if they do.

See now, the X chips do not come stock coolers because in order to best take advantage of XFR, AMD expects you use better cooling solutions. This is according to Dr. Su in her speech announcing this feature.

2 Likes

Oopsie. As you can see i never got into Ryzen. :stuck_out_tongue_winking_eye:

I then guess its up to what can the chip hit while still being stable.

That isn't to say that the AMD stock coolers are bad, they are very good on the non-X chips and even allow some reasonable overclocking, just that the X chips have a higher TDP and detected thermal headroom is what XFR works on.

1 Like

XFR is mostly just a marketing gimmick and is only used on one core loads. It isn't automatic overclocking. It is just another, higher boost state.

The reason the others boost higher is because they have less cores and lower clocks to begin with.

Manual OCing disables it.

No. It won't get higher. If you're hoping for xfr pushing a Ryzen CPU to 4.3 then you're gonna be disappointed. Even with manual OCing Ryzen can barely reach 4.2. It is a hard architecture limitation. Temperature has nothing to do with it. Just takes way too much voltage.

The CPU knows it's true temperature. It just reports it higher to ensure adequate cooling. Removing it would change little.

Also btw that chart you posted from lovely WCCFShitTech is completely wrong. Like the CPU names are wrong, the core counts are wrong, cache levels are wrong and clocks are wrong lmao.

3 Likes

@OP

Literally that entire graph is wrong for one.
For two, it gives a boost of 100-300mhz depending on the CPU.

So yeah.

To be honest, XFR was for me the only letdown of Ryzen. I wachted AMD's livesteam in December and while it was not specifically said, I did have the impression that XFR would automatically overclock higher than the additional 100MHz. Sure, Ryzen isn't an overclocking beast in general, but I didn't know this back then.
Still, I like my Ryzen CPU and I don't care to much about its, in my opinion limited, XFR.

Actually as low as 50Mhz I think right?

On one core should be at least 100 iirc

Dat graph do.

I think the reason we both got the impression that XFR would be more than it is currently, is that it hasn't been fully implemented yet. I don't think Lisa Su intentionally exagerrated the feature, I believe that it will improve, just as I believe the current low overclock ceiling will be lifted in the future, not as high as Kaby Lake, but higher than 4080Mhz.

AMD are known for and expecting to once again play the long game on AM4 , just like 2 and 2+ & 3 and 3+ before it. So while the current Ryzen chips hit their limit at about 4.1ghz realistically the chips that come after it, the next generation not the R5's, will likely be more able to leverage XFR. It is a nice marketing gimik at the moment but is held back by the fact that Ryzen ships at near max speed already.

Proper overclockers will force a manual OC at the max frequency, 4 or 4.1ghz, on all cores and call it a day. Regular users will let XFR speed up one core occasionally. The next generation should take better advantage of it just like Intel core boost on version 3 now or whatever. Improvements over time.

1 Like

It might not even be an architechure limitation as much as a limitation of the 14nm FinFET process that Ryzen is produced on. The Ryzen core really isn't that big, so in theory it might have been possible for clocks to have been higher had the process been better. Intel really has an advantage here regarding the manufacturing process, since their fabs are regarded as maybe the best around right now.

Ehh the Samsung process is actually pretty good. It's very close to Intel. Yes Intel has a smaller process for sure but their yields are actually pretty poor on it. Worse than Glo Flo. Their 10nm is a joke and AMD will likely beat them to 7nm.

It's less a process defect Imo and more specific design choice. The process is built for low power consumption. Ryzen is built for low power consumption. It has aggressive clock gating and other features. So while I'm sure process plays a part it is much more conscious design decisions than anything else.

Also of course it is a brand new architecture. Look at Intel for example when they introduced the first gen i parts. The top end 4/8 i7 was only like a 3.2ghz part. By the next gen it was closer to 3.8.

Clocks will improve I'm sure. Again this is a brand new part. I'm sure by next generation we will have AMDs 2500K. Ie a processor that is extremely long lasting in the market.

1 Like

I'll just cross my fingers and hope that AMD uncouples the infinity-fabric from the memory speed in their next design iteration. Get it running ridiculously fast disregarding memory speed as a factor. But yeah your probably right it being a design choice of the process.

2 Likes

Sames.

That would honestly solve so many problems lol.

As it is now though at least RAM is getting better. Mays update should fix it for good.

1 Like

I kinda like the memory being integrated, just because for the first time in a long time the RAM speed actually matters.

The speed not being very fast to start is a result of the motherboards, not the CPU's.

I was under the impression that XFR (extended frequency range?) was to kick up the FSB when needed to get a little more kick to it. And if you don't want to do any (FSB overclocking) it will do it for you... If you are an over clocker it makes no sense as you could get the r7 1700, and a motherboard with BCLK to set the frequency to whatever you want, based on the mobo's limitations.

I think XFR was supposed to be a bigger deal with this launch too but Intel happened to drop a very fast line of CPU's with Kaby Lake so AMD had to push their products more, effectively shrinking the headroom for XFR to work at.
I'm don't think that uncoupling the infinity fabric from the RAM speed is possible. But as Intel's push for Crosspoint made JDEC come out with the DDR5 specs, maybe we will see an early adoption. We will see.