AMD - Zen chitchat

[watercooled]

I'm not 100% certain on this but I don't think SMT can actually be broken in isolation (or it's so unlikely it's not worth actually binning for) - I remember Charlie at SA explaining this about Intel's implementation, the function is an integral part of the core and not a discrete block. Disabling it is more down to product segmentation either for retail cost or power reasons.

Edit: Managed to find it after submitting the post! https://semiaccurate.com/2012/12/13/...n-crystalwell/

HT is an integral part of the core pipeline, it is not a unit that can be turned off or on without preventing the CPU from working. If there is an error in the unit, the whole core is scrap. It has to be turned off artificially, basically by being told to not work programatically, the circuitry must continue to function for the core to at all.

[scaryjim]

... Disabling it is more down to product segmentation either for retail cost or power reasons. ...

I wonder if running with SMT enabled can affect the leakage of an imperfect die? At which point while you're not specifically looking at binning for functional SMT, you're looking at hitting a voltage/TDP bin that you can only achieve by disabling SMT, which has the same effective outcome - a die that gets binned as only being usable with SMT disabled.

TBH if Ryzen 3 mobile was 4C/4T with 384 Vega shaders, and it arrived in laptops costing ~ £400, that'd be a pretty much instant buy for me, as it should give a decent jump in performance from my A10 4600M...

[Millennium]

I wonder if running with SMT enabled can affect the leakage of an imperfect die? At which point while you're not specifically looking at binning for functional SMT, you're looking at hitting a voltage/TDP bin that you can only achieve by disabling SMT, which has the same effective outcome - a die that gets binned as only being usable with SMT disabled.

TBH if Ryzen 3 mobile was 4C/4T with 384 Vega shaders, and it arrived in laptops costing ~ £400, that'd be a pretty much instant buy for me, as it should give a decent jump in performance from my A10 4600M...

I think it would be an instant buy for me too, and I don't even need another laptop ! lol.
I would use it in place of a chromecast. However, the chromecast is remote controlled. So I may look at a solution to get it working on my whole setup and let off the upgrade... mmhmm

[DanceswithUnix]

I wonder if running with SMT enabled can affect the leakage of an imperfect die? At which point while you're not specifically looking at binning for functional SMT, you're looking at hitting a voltage/TDP bin that you can only achieve by disabling SMT, which has the same effective outcome - a die that gets binned as only being usable with SMT disabled.

SMT only increases power consumption because it increases throughput and hence means pipelines parts aren't idle as much, I can't think of how an imperfect die would make any difference there.

However, AIUI a quad core can be more power efficient way of getting 4 threads than dual core with threading turned on (presumably because you can drop the clocks). So yes, they can play with turning threading on and off to wring out performance per watt.

[scaryjim]

SMT only increases power consumption because it increases throughput and hence means pipelines parts aren't idle as much, I can't think of how an imperfect die would make any difference there. ...

I was just thinking along the lines that we know different dies off the same silicon line have different leakage - that's part of how you bin for server/desktop/HEDT/mobile, after all. Leakage increases with temperature, so if turning off SMT reduces overall power draw, it might be possible to run a leakier part at a lower voltage (or higher clock speed) with SMT off than with SMT on.

If you compare the 1300X and the 1500X power consumption (particularly the cores-only full load), there's a relatively small difference (~5%) given the lack of SMT and the halving of the L3 cache. It's far from conclusive, but it fits my general thesis that Ryzen 3 dies are binned because they require too much voltage/power to run with SMT enabled...

[CAT-THE-FIFTH]

I wonder if running with SMT enabled can affect the leakage of an imperfect die? At which point while you're not specifically looking at binning for functional SMT, you're looking at hitting a voltage/TDP bin that you can only achieve by disabling SMT, which has the same effective outcome - a die that gets binned as only being usable with SMT disabled.

TBH if Ryzen 3 mobile was 4C/4T with 384 Vega shaders, and it arrived in laptops costing ~ £400, that'd be a pretty much instant buy for me, as it should give a decent jump in performance from my A10 4600M...

ATM,you can get an FX9830P based laptop with a 128GB SSD,1TB HDD and a RX460 4GB GDDR5 for £500.

[scaryjim]

ATM,you can get an FX9830P based laptop with a 128GB SSD,1TB HDD and a RX460 4GB GDDR5 for £500.

That's a pretty good deal for the hardware, but the balance looks off to me - BR with an RX 460? 1366x768 screen?

Drop the RX 460 from that spec, drop the price by at least £50, and you'd have something that might interest me. Or give it a 1080p screen at the same price - they could even drop the pokey little SSD to balance the costs. I can always slap my own SSD in later if I decide it's necessary. But a dGPU with a 1366x768 screen makes no sense to me. Absolute deal breaker.

[CAT-THE-FIFTH]

That's a pretty good deal for the hardware, but the balance looks off to me - BR with an RX 460? 1366x768 screen?

Drop the RX 460 from that spec, drop the price by at least £50, and you'd have something that might interest me. Or give it a 1080p screen at the same price - they could even drop the pokey little SSD to balance the costs. I can always slap my own SSD in later if I decide it's necessary. But a dGPU with a 1366x768 screen makes no sense to me. Absolute deal breaker.

For gaming it does probably make more sense,as a 1080p is actually twice the resolution of 768p,so realistically you would need something much more higher end in a laptop,to get similar performance. In context my GTX960 4GB could get pushed quite easily at 1080p,so I don't think an RX460 is really that much OTT at a lower resolution.

Also the FX9830P is the highest BR SKU and by extension the fastest AMD mobile CPU prior to Ryzen based ones,so should make it relatively close to an Athlon II X4 845. So an Athlon X4 845 with a graphics card between an RX550 and a RX460 does not seem that unbalanced,certainly not as much as the higher end laptops with AMD CPUs from the previous generation which had faster graphics cards and worse CPUs.

Edit!!

Single core performance is close to a Intel Core i3-6100U and the multicore performance is slightly behind a Intel Core i5-7200U.

[watercooled]

I was just thinking along the lines that we know different dies off the same silicon line have different leakage - that's part of how you bin for server/desktop/HEDT/mobile, after all. Leakage increases with temperature, so if turning off SMT reduces overall power draw, it might be possible to run a leakier part at a lower voltage (or higher clock speed) with SMT off than with SMT on.

If you compare the 1300X and the 1500X power consumption (particularly the cores-only full load), there's a relatively small difference (~5%) given the lack of SMT and the halving of the L3 cache. It's far from conclusive, but it fits my general thesis that Ryzen 3 dies are binned because they require too much voltage/power to run with SMT enabled...

Yeah I think that would be possible - as DanceswithUnix says the SMT itself isn't what makes much difference to power consumption (there are a couple of statically partitioned units but they're a relatively tiny part of the core), but it being enabled does allow greater utilisation of the core, therefore resulting in higher consumption. So they (Intel/AMD) likely do bin based on that for whatever workloads they bin with (it's theoretically possible, AIUI, to get an SMT core to draw pretty much the same power with SMT turned off, with exactly the right workload/power virus, just not that common in normal workloads hence the reason for SMT in the first place) - having SMT enabled could mean that workload pushes the die above the threshold for a given TDP bin.

@DanceswithUnix: WRT power efficiency, IIRC there are certain workloads where SMT makes zero to negative impact on performance but may increase power draw anyway e.g. through cache thrashing. I'm not certain how much of the responsibility for managing such things lies with hardware, kernel and software?

[scaryjim]

For gaming it does probably make more sense. ...

For gaming you can run at a non-native resolution if you can't find settings you like at the native res - as it is most of the games I play ran fine at 1080p on the A6 3670k we won a few years back, so for me gaming performance isn't a big issue. My laptop gets casual games and things like Torchlight/Neverwinter Nights/XCOM, which look fine at lowest settings on a native 1080p monitor, but if it came down to it I'd happily drop the resolution too for games ... for me, any game that isn't worth playing at lowest quality settings isn't worth playing full stop.

For productivity, OTOH, you can't increase the resolution, and having the higher res makes a huge difference for documents/web browsing etc. My work laptop is a 14" 1080p and the difference switching between that and my HP is night and day.

And while that might be the fastest BR chip released on mobile, it's actually not a huge amount faster than my A10 in most circumstances. I was surprised when I looked at the comparison just how little progress was made in - what, 4 generations? I'd put money on there being a noticable difference between the FX + RX 460 and an i5 U + RX 460...

[CAT-THE-FIFTH]

And while that might be the fastest BR chip released on mobile, it's actually not a huge amount faster than my A10 in most circumstances. I was surprised when I looked at the comparison just how little progress was made in - what, 4 generations? I'd put money on there being a noticable difference between the FX + RX 460 and an i5 U + RX 460...

You need to realise on desktop,the later Intel chips matched the AMD ones in terms of clockspeeds,but in mobile those lower end U series CPUs don't really perform as well due to low clockspeed. That 9830P for instance has around the same single core performance as an FX8350. The X4 845 matches the performance of an A10 7870K for example.

Also compared to the A10 4600M(which I think you have),the BR CPUs do have a big performance jump:

https://www.notebookcheck.net/AMD-A-...r.74065.0.html
https://www.notebookcheck.net/AMD-Br....181434.0.html

The huge difference is clockspeeds - BR just boosts much higher. Look at the R15 ST scores,the A10 4600M scores around 65ish and the FX9830P around 96. That comes from having a 700MHZ higher base clockspeed. It seems in that Asus chassis the CPU runs at betweem 3.2GHZ~3.5GHZ,so at least the CPU is well cooled.

The Sunspider score is 35% faster and so on. I would expect in any lightly threaded game or scenario the performance jump should be noticable.

If anything,with BR its the ST performance which went up the most in the mobile arena,more than MT performance(due to lower amount of L2 cache).

Edit!!

The other alternative is something like a Core i5 and a MX150,but once you factor in even a cheapo SSD,its starting at £650ish onwards.

Its probably faster,but sadly the choice at £500 at under its usually crap Intel IGPs and spinny disks.

[scaryjim]

... in mobile those U series CPUs don't really perform as well due to low clockspeed. ...

i5 7200U is 2.5Ghz - 3.1GHz. That gives it ~ 17% lower clocks than the FX9830. But the IPC deficit of BR compared to Skylake is way, way more than 17%. Wasn't Ryzen meant to me a 52% increase in IPC and it's *still* 10% behind Intel?

Why don't we compare the FX 9830p and the i5 7200U directly? https://www.notebookcheck.net/7200U-....247596.0.html

The i5 7200U Cinebench R15 ST is 126 on notebookcheck - WAY ahead of the FX. In fact the comparison benchmarks basically show that the FX matches (and sometimes beats) the i5's multithreaded performance but drops behind on single threaded performance. So the mobile i5 is the better bet - better performance in lightly threaded tasks with roughly equal performance in heavily threaded tasks.

Yes, mobile BR was the pinnacle of AMD's construction CPUs. And to be fair the FX 9830p is further ahead of my A10 than I realised - although I was looking at the 15W BR chips; the FX is 25W - 45W. But even then - and despite its clockspeed advantage - it only matches a mid-range Intel part in heavily threaded workloads, and is still well behind on single threaded. So I'm going to stick by my suggestion that you'd notice the difference between the FX + 460 and the i5 + 460.

All of that's a bit moot though, for a number of reasons - not least that I reckon neither of those processors *should* be paired with an RX 460: I'd personally look for an i5 HQ as a minimum. Swap the RX 460 out for an RX 550/540, and I'd at least raise an eyebrow. Get rid of it completely and give me the FX and a 1080p screen, and I'd be tempted if the price was right. But then I'd know more Ryzen laptops are due soon, and why would I buy BR when I can wait for Ryzen, which should inevitably wipe the floor with BR in every metric?

As a package, that laptop makes little sense. And for me, the lack of a 1080p panel is a deal breaker, straight up. No matter how much hardware yo're otherwise getting for your money.

[scaryjim]

Finally a review site tested both AMD and Intel versions of the HP Envy X360:

https://www.laptopmag.com/reviews/la...-envy-x360-15t

Don't know if you've seen it yet CAT, but there's also a performance-centric review of the two laptops: https://www.laptopmag.com/articles/a...l-8th-gen-core

Ryzen comes out of it pretty damn well

[CAT-THE-FIFTH]

i5 7200U is 2.5Ghz - 3.1GHz. That gives it ~ 17% lower clocks than the FX9830. But the IPC deficit of BR compared to Skylake is way, way more than 17%. Wasn't Ryzen meant to me a 52% increase in IPC and it's *still* 10% behind Intel?

Why don't we compare the FX 9830p and the i5 7200U directly? https://www.notebookcheck.net/7200U-....247596.0.html

The i5 7200U Cinebench R15 ST is 126 on notebookcheck - WAY ahead of the FX. In fact the comparison benchmarks basically show that the FX matches (and sometimes beats) the i5's multithreaded performance but drops behind on single threaded performance. So the mobile i5 is the better bet - better performance in lightly threaded tasks with roughly equal performance in heavily threaded tasks.

Yes, mobile BR was the pinnacle of AMD's construction CPUs. And to be fair the FX 9830p is further ahead of my A10 than I realised - although I was looking at the 15W BR chips; the FX is 25W - 45W. But even then - and despite its clockspeed advantage - it only matches a mid-range Intel part in heavily threaded workloads, and is still well behind on single threaded. So I'm going to stick by my suggestion that you'd notice the difference between the FX + 460 and the i5 + 460.

All of that's a bit moot though, for a number of reasons - not least that I reckon neither of those processors *should* be paired with an RX 460: I'd personally look for an i5 HQ as a minimum. Swap the RX 460 out for an RX 550/540, and I'd at least raise an eyebrow. Get rid of it completely and give me the FX and a 1080p screen, and I'd be tempted if the price was right. But then I'd know more Ryzen laptops are due soon, and why would I buy BR when I can wait for Ryzen, which should inevitably wipe the floor with BR in every metric?

As a package, that laptop makes little sense. And for me, the lack of a 1080p panel is a deal breaker, straight up. No matter how much hardware yo're otherwise getting for your money.

The difference between your A10 4600M and an FX9830P is between 35% to 50% in lightly threaded tasks,and you are making a bigger noise about,what a 30% difference in ST performance and a matching of performance in MT. Which is all irrelevant - you are trying your best to make the deal I sound look bad - instead of moaning,find me a £500 laptop with better graphics performance?? You can't,and not one with a boost SSD and seperate storage drive.

Like I said I looked - a Core i5 7200U with a MX150,plus the cost of an SSD will be £650ish. Its faster but you won't get a better gaming laptop for under £500.

I tried an HD5850 1GB on a Core i3 2100,A6 3670K and a 3GHZ Q6600 and guess what?? For most intents and purposes there was hardly any realworld difference in gaming performance - very few games were massively ahead in reality despite a massive increase in ST performance on the Core i3. If that Q6600 rig hadn't started having issues I would have not got the Core i3 2100.

You are going to be mostly GPU limited with a RX460 in newer games,and having run newer games I can attest to that - I have had an HD5850,HD6870,HD7850,GTX660,GTX960,RX470 and GTX1080 and most of my games have been GPU limited outside FO4,and that was only after I severely modded the game with settlements having 100s of people in total,and that was even only noticeable when I got a GTX1080.

Heck,I have another playthrough without large settlements and I am mostly fine.

Guess what the ST performance of that IB Core i7 is?? Its around the same as a Core i5 7200U.

So honestly are you now trying to tell me that an FX9830P is going to be a limiting factor for 95% of gaming with a downclocked RX460? I would argue a bigger limitation is whether the RX460 would downclock in a laptop.

Edit!!

Also another issue you don't get,are you honestly now telling me all those SKL,Broadwell and HW Core i3 and Core i5 CPUs which are clocked lower,and have similar ST performance are suddenly unusable??

What about many of those having similar performance dGPUs like the GTX950M,etc.

Many Core M based laptops have worse performance overall especially under extended load and they are in premium laptops and tablets.

Second Edit!!

I can even get the 1080p screen argument,but honestly if CPU were all that,a Bristol Ridge APU shouldn't be thrashing the Core i5 8400 when it comes to IGP gaming.

[scaryjim]

... you are trying your best to make the deal I sound look bad ...

Look bad? No. I've said several times it's a lot of hardware for the money. You can't get a better dGPU in a laptop at that price.

But that's kind of the point I was making. It's unbalanced. It's very graphics heavy. I accept that the FX 9830p is faster than I thought, but it's *still* slower than a mid-range U series chip, even though, as you put it, "in mobile those U series CPUs don't really perform as well due to low clockspeed." Well, those U series CPUs *do* perform as well as an FX 9830p. Better, in lightly threaded tasks. So the FX is a compromise to hit a price point. That laptop would be faster in games with an i5 U.

But that's only part of the story. The 768p screen might be ideal for getting good frame rates from the RX 460 in medium to high quality gaming, but that's pretty much all its good for. For general use a 1080p screen is just WAY better. It's another compromise to hit a price point. Having small boot SSD and a spinning rust storage drive is another compromise. It's a way to make the spec look better without actually providing that much more usability, because you can hardly fit anything on a 128GB SSD.

That laptop is great for one thing, and that's playing games in medium to high quality. For that use, it's an excellent buy. No argument here.

But that's not what I use my laptop for, and it makes a lot of compromises to be what it is. So it's not a good deal for me (or for anyone else who uses their laptop the way I do).

[CAT-THE-FIFTH]

Look bad? No. I've said several times it's a lot of hardware for the money. You can't get a better dGPU in a laptop at that price.

But that's kind of the point I was making. It's unbalanced. It's very graphics heavy. I accept that the FX 9830p is faster than I thought, but it's *still* slower than a mid-range U series chip, even though, as you put it, "in mobile those U series CPUs don't really perform as well due to low clockspeed." Well, those U series CPUs *do* perform as well as an FX 9830p. Better, in lightly threaded tasks. So the FX is a compromise to hit a price point. That laptop would be faster in games with an i5 U.

But that's only part of the story. The 768p screen might be ideal for getting good frame rates from the RX 460 in medium to high quality gaming, but that's pretty much all its good for. For general use a 1080p screen is just WAY better. It's another compromise to hit a price point. Having small boot SSD and a spinning rust storage drive is another compromise. It's a way to make the spec look better without actually providing that much more usability, because you can hardly fit anything on a 128GB SSD.

That laptop is great for one thing, and that's playing games in medium to high quality. For that use, it's an excellent buy. No argument here.

But that's not what I use my laptop for, and it makes a lot of compromises to be what it is. So it's not a good deal for me (or for anyone else who uses their laptop the way I do).

BTW,my main boot drive is that 120GB SSD I won,which replaced the POS 480GB Sandisk. Its more than enough for general stuff and I certainly had Planetside 2 installed on it for years before - only PS2 and FO4 seemed to really benefit from having an SSD in my own usage. PS2 was more the case since it was quicker for respawns.

It also won't be faster with a number of those mobile Core i3/Core i5,because AT did a review of the similar X4 845 and compared it to the Core i3 6100TE. The FX9830P runs at close to 3.5GHZ,so the approximation is close,so is the Core i3 6100TE which is also clocked around 10% lower than the mobile Core i5,but I suspect the Turbo clocks will be quite similar in reality.







Those are all very CPU intensive games BTW and some like GTAV don't even like Ryzen that much(seems more an optimisation issue with AMD CPUs),and most are within a few percent margin of error,and in one situation the X4 845 wins big and the other the Core i3 does. But either way,look at the framerates. One game is unplayable,and every other one is at least 30FPS.

That is with an R9 285 and the RX460 in the laptops will run at a lower clockspeed especially under load. Your easily looking at a 30% maybe 40% performance increase in reality.

So in reality,a downclocked RX460 is still going to be the bigger limitation,and as games become more graphically intense over the next few years,its the GPU which will be taxed more and more anyway.

Its what I found with gaming in laptops in practice - its the GPUs which are the limiting factor here. Its why I lament the slow progress in upgradable and external graphics standards. There are many laptops with solid CPUs which could benefit from such tech,especially the quad Intel ones from more recent generations.

I mean if it were a mobile GTX1060 then I would agree with you,and sure a 1080p screen would be more useful for certain stuff,but then I find even most 1080p screens on laptops rubbish since they are not colour accurate enough and many are too glossy!!