It appears the one big-er 'Prime' core concept is paying dividends in ST performance, numbers floating around are a leap from 2400 in GB4 single core, to 3700, not bad for one generations jump.
Do we know EXACTLY what is going on with that Prime core? The way it's being sold is very strange. Suppose the pitch were 4 A76 (sorry, Kryo 485 Gold), maximum frequency of 2.84 GHz. No-one would be upset if that means when ONE core is running, it hits 2.84 GHz, and this drops when two or more are running. Everybody does that -- other mobile, Apple, Intel, etc.
So what's different? The obvious answer is that only ONE of the cores can hit 2.84GHz. But even that raises a whole bunch of questions like
- why draw attention to this? It's a weakness, not a strength; but it can be hidden by just having the OS always place the single thread on the Prime core, not on just any core.
- is this something detected by testing? ie design to 2.84 GHz, then choose whichever of the 4 runs that at lowest power? There might actually be some good sense to that (generically, rank ALL cores by power burned, and have the OS use that in scheduling) but again why make this public?
- or is this core deliberately different? Like some part of it (I assume not much, just a few critical paths) uses high performance rather than low power transistors? Is that effort worth an extra 15% performance (hopefully at not too much energy cost)? I'd have to see the actual performance/power tradeoff, but it IS (IMHO) an intriguing idea. On Twitter I've mulled over the idea of Apple offering a faster than Vortex core (but higher power) with the AX SoCs offering something like 1V+ & 3V & 4T cores, which is what QC appear to be doing here. What I had in mind was a more aggressively designed super-core, but even a tweaked+recompiled version of a Vortex core (ie, like I said, upgrade just the transistors in the critical paths) does the job and 15% is 15% --- that's about a half-generation performance boost which ain't nothing.
So, yeah, IMHO can the whining! There are some very nice ideas here. The beefed up DSP, for example, remains a nice feature and I keep wondering if that's another block of silicon that Apple will eventually (these things take time!) drop onto its SoCs. Certainly the sorts of things they are offering with this tech (like photography feature, available to video) are good ideas, and I wonder if Apple will be offering them next year.
It's possibly also the best binned for higher clocks, like Intel is starting to do now? Not sure. But yeah, the extra cache and highest turbo alone still make for a pretty interesting ST performance bump.
Vague handwavy specs to outright lies to... I don't even know what some of this press conference was. "Our hardware supports PBR!" is the most nonsensical statement I've ever seen. It's like congratulating yourself on releasing a car that can fit things inside it.
Who the hell was this conference for? What was the point? I think tech companies that make internal, non consumer facing components have gone batshit for weird PR conferences like this. They tell tech people nothing true or relevant. They tell consumers nothing they care about. So who are they selling too? These things seem to exist only so PR people at the company can justify having a job.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
7 Comments
Back to Article
tipoo - Wednesday, December 5, 2018 - link
It appears the one big-er 'Prime' core concept is paying dividends in ST performance, numbers floating around are a leap from 2400 in GB4 single core, to 3700, not bad for one generations jump.name99 - Wednesday, December 5, 2018 - link
Do we know EXACTLY what is going on with that Prime core? The way it's being sold is very strange.Suppose the pitch were 4 A76 (sorry, Kryo 485 Gold), maximum frequency of 2.84 GHz. No-one would be upset if that means when ONE core is running, it hits 2.84 GHz, and this drops when two or more are running. Everybody does that -- other mobile, Apple, Intel, etc.
So what's different? The obvious answer is that only ONE of the cores can hit 2.84GHz. But even that raises a whole bunch of questions like
- why draw attention to this? It's a weakness, not a strength; but it can be hidden by just having the OS always place the single thread on the Prime core, not on just any core.
- is this something detected by testing? ie design to 2.84 GHz, then choose whichever of the 4 runs that at lowest power? There might actually be some good sense to that (generically, rank ALL cores by power burned, and have the OS use that in scheduling) but again why make this public?
- or is this core deliberately different? Like some part of it (I assume not much, just a few critical paths) uses high performance rather than low power transistors? Is that effort worth an extra 15% performance (hopefully at not too much energy cost)? I'd have to see the actual performance/power tradeoff, but it IS (IMHO) an intriguing idea.
On Twitter I've mulled over the idea of Apple offering a faster than Vortex core (but higher power) with the AX SoCs offering something like 1V+ & 3V & 4T cores, which is what QC appear to be doing here. What I had in mind was a more aggressively designed super-core, but even a tweaked+recompiled version of a Vortex core (ie, like I said, upgrade just the transistors in the critical paths) does the job and 15% is 15% --- that's about a half-generation performance boost which ain't nothing.
So, yeah, IMHO can the whining! There are some very nice ideas here.
The beefed up DSP, for example, remains a nice feature and I keep wondering if that's another block of silicon that Apple will eventually (these things take time!) drop onto its SoCs. Certainly the sorts of things they are offering with this tech (like photography feature, available to video) are good ideas, and I wonder if Apple will be offering them next year.
tipoo - Wednesday, December 5, 2018 - link
The Prime core has twice the L2 cache for onehttps://www.anandtech.com/show/13680/snapdragon-85...
It's possibly also the best binned for higher clocks, like Intel is starting to do now? Not sure. But yeah, the extra cache and highest turbo alone still make for a pretty interesting ST performance bump.
Frenetic Pony - Wednesday, December 5, 2018 - link
What a bunch of disappointing nonsense.Vague handwavy specs to outright lies to... I don't even know what some of this press conference was. "Our hardware supports PBR!" is the most nonsensical statement I've ever seen. It's like congratulating yourself on releasing a car that can fit things inside it.
Who the hell was this conference for? What was the point? I think tech companies that make internal, non consumer facing components have gone batshit for weird PR conferences like this. They tell tech people nothing true or relevant. They tell consumers nothing they care about. So who are they selling too? These things seem to exist only so PR people at the company can justify having a job.
Meteor2 - Wednesday, December 5, 2018 - link
Did you get out of bed on the wrong side this morning?WasHopingForAnHonestReview - Wednesday, December 5, 2018 - link
Next he complain how it's Trump's fault.BurntMyBacon - Thursday, December 6, 2018 - link
@Frenetic Pony: "These things seem to exist only so PR people at the company can justify having a job."I believe you now understand fully half of why marketing employees do the things they do.