Speculation: Ryzen 4000 series/Zen 3

Page 70 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
21,570
10,763
136
Sadly they only did it for one bench. That or it only made a difference on one bench. It's a bit off-topic so I'll stop now.

To segue: Sub-ambient cooling might be tempting for Matisse since it gets to be such a hassle to cool it with just a little bit of a clock bump. Hopefully 7nm+ will be easier to deal with than 7nm, but it's doubtful. Maybe someone could come up with a nitrogen-sealed, chilled case for PCs? Most non-HDD components can handle temps down to -60C without malfunction. Nitrogen-sealed means: no condensation. The only trick would be external peripherals. Also, there are some things you don't want to add to your heat load when running an active-cooled environment, like the PSU. Also NVMe drives might not be happy at those temps either.
 

soresu

Platinum Member
Dec 19, 2014
2,582
1,778
136
I lost my phone in the snow for an evening, and it was single digit in Fahrenheit at the time. No damage. But the charge was near empty. I somewhat worry that a fully 100% charged battery could get shocked when quickly going from 40C to 45C (which it might easily coming fresh off a fast charger) to -20C. The energy that a battery can hold at room temperature is significantly more than what it can hold at -20C. So what happens to that energy that it now all the sudden cannot hold any more?
From what I understand it has to do with the chemistry and structure of the electrodes and electrolyte - especially the electrolyte as solid electrolyte cell batteries can function well at somewhat lower temps than liquid electrolyte cells can.

The energy stored in a rechargeable electrochemical battery cell is not like bottled lightning so much as ions induced to move from one electrode to the other during charging - even without use it will discharge anyway over time as those ions move back to their natural states, seemingly a drop in temperature accelerates it.

It seems very counter intuitive as cold should slow down chemical processes - but I probably have the wrong end of the electrochemical stick, so don't take me at my word.

Edit:

Found this link which takes a stab at an explanation without exact specifics.

Sounds like the battery does not discharge so much as simply stops working well enough to detect a charge coming from it after the electrolyte temp dips to a critical level which impedes ion movement - like a power wire that suddenly develops a massive resistance between the two electrodes in each cell.

The device (phone) makes an educated guess at the current level of charge in its battery based on the output voltage I think.

As it drops during discharge, the device interprets the battery voltage output as an equivalent percentage of remaining charge until it gets to an unacceptable voltage for the device minimum spec to run at - which the device interprets as 0% on the battery charge, rather than the absolute discharge of the battery (repeated absolute discharge is not a good idea for long term battery health anyway).

When the liquid electrolyte freezes, the battery can no longer discharge properly, so the output voltage drops dramatically - leading the device to think the battery has actually discharged instead.

Solid electrolyte circumvents this problem by having sort of engineered fixed channels for ion movement - not quite as mobile/fast as liquid electrolyte, but less temperature dependent.
 
Last edited:
  • Like
Reactions: amd6502

uzzi38

Platinum Member
Oct 16, 2019
2,556
5,531
146
More Renoir for you lot, and this comes with a lot of info:

So from what we can see, it's 8C16T, sports a base clock of 1.8GHz and - in this device - is maintaining an average clock speed of 3.65GHz over the duration of the test. Clocks are likely higher for SC, lower for MC.

L3 Cache per CCX appears to be cut down to 4MB, and it looks to be capable of supporting DDR4-3200. Drivers have also shown it should be capable of supporting LPDDR4X-4266 as well, though we're yet to see a device using such memory.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
I lost my phone in the snow for an evening, and it was single digit in Fahrenheit at the time. No damage. But the charge was near empty. I somewhat worry that a fully 100% charged battery could get shocked when quickly going from 40C to 45C (which it might easily coming fresh off a fast charger) to -20C. The energy that a battery can hold at room temperature is significantly more than what it can hold at -20C. So what happens to that energy that it now all the sudden cannot hold any more?
Nothing happens. The chemical reactions in the battery just slow down.
 

jpiniero

Lifer
Oct 1, 2010
14,487
5,154
136
Forget exactly where I saw it, but Van Gogh was mentioned as RDNA2. I wonder if it's actually not an APU but comparable to Kaby-G like a discrete GPU on package.
 

uzzi38

Platinum Member
Oct 16, 2019
2,556
5,531
146
Forget exactly where I saw it, but Van Gogh was mentioned as RDNA2. I wonder if it's actually not an APU but comparable to Kaby-G like a discrete GPU on package.
Yes it is, gfx103X indicates it's Navi.

But, personally I think it'll be an APU instead. They're easier to configure and tend to be a bit cheaper, especially important for something like a mobile device.

Though I could be wrong. I just don't see it happening yet.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Actually, with indications of Van Gogh using Navi 21/23 GPU, it'll be way too big of an APU.

Sounds like AMD's version of Kaby-G having fat GPU on package. Or its a custom chip meant for specialized markets.

Navi 21 = 500mm2 die?
Navi 23 = hardware ray tracing?
 
Last edited:
  • Like
Reactions: lightmanek

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I remember they had something like that for Chinese market last year (or maybe couple of years ago).

Yes, that's what I'm talking about. 500-600mm2 just for the GPU portion is way overkill, but it might make sense for the custom market, like in consoles.

More Renoir for you lot, and this comes with a lot of info:

The GPU in the particular system is 40% faster than the GPU in Picasso.

The CPU is 10-15% better in multi-thread than the one in Cometlake, and 20-25% faster than Picasso in single thread. Icelake is 8-10% faster in ST, but is behind due to having only 4 cores.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Is that IPC you're talking, or are we talking absolute performance?

It's impossible to separate the result.

I looked at enough results to know it seems to be a decent benchmark as long as you know what you are looking for and isolate some weird results.

The key point is they have caught up mostly but likely still fall behind latest Intel cores in uarch.
 

uzzi38

Platinum Member
Oct 16, 2019
2,556
5,531
146
Yes, that's what I'm talking about. 500-600mm2 just for the GPU portion is way overkill, but it might make sense for the custom market, like in consoles.



The GPU in the particular system is 40% faster than the GPU in Picasso.

The CPU is 10-15% better in multi-thread than the one in Cometlake, and 20-25% faster than Picasso in single thread. Icelake is 8-10% faster in ST, but is behind due to having only 4 cores.
Excuse me what? Where did you get those numbers from?

CPU side:




And on the GPU side:


 

Gideon

Golden Member
Nov 27, 2007
1,598
3,520
136
We definitely need more benchmarks before concluding anything, but so far looks quite promising. 9% faster ST performance, 66% faster MT and roughly 60% faster GPU than Ice Lake.

I'm sure that some people here will continue to claim that AMD's mobile chips are useless garbage but hopefully this will convince OEMs to finally make premium laptops with these chips.
 

juergbi

Junior Member
Apr 27, 2019
12
14
41
[...] 66% faster MT [...] than Ice Lake.

That's just the 8-thread performance. UserBenchmark is hiding the 64-thread performance further down and, according to that, Renoir 4800U is 121% faster than Ice Lake 1065G7. The Ice Lake 64-thread score is oddly lower than with 8 threads, though, while I'd expect it to be about the same. Correcting that still results in a 112% lead for Renoir, though.

I assume the scores on both sides are at maximum boost. Sustained performance at the 15W and 25W limits would also be interesting.
 

uzzi38

Platinum Member
Oct 16, 2019
2,556
5,531
146
Actually, with indications of Van Gogh using Navi 21/23 GPU, it'll be way too big of an APU.

Sounds like AMD's version of Kaby-G having fat GPU on package. Or its a custom chip meant for specialized markets.

Navi 21 = 500mm2 die?
Navi 23 = hardware ray tracing?

Also, this is completely wrong, it's FP6, just like Renoir.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136

That's why I said you need to know what you are looking for.

When you are comparing through Geekbench, you can't use average results, because its user submitted and with potentially million different configurations, some bad, means it can easily fluctuate 10%. In the extremes, it can fluctuate 50-100%.

It's same with userbenchmark. The top 1065G7 gets ~145 points in ST, and ~650 points in max thread. I discard anything in between because who knows what's logical threads and what's single thread?

The 4800U gets 138 points in ST and 1074 points in MT. Do you really believe the 1065G7 is only 10% faster in ST compared to 3700U? The compared Icelake system is underperforming. The ~30% difference approximates the high end of the single threaded difference we see with real world systems.

Are you interested in facts, or something that's convenient to fit your belief system? You agree that user submitted results suck for that reason and why we clamour to see benchmarks from reputable sites right? Enthusiasts cry about seemingly insignificant detail such as minor revisions of graphics driver versions when reviews are out. Yet you take user submitted results as concrete evidence?

Yes, some results show the 64-thread(or max thread) result is slower than the 4-threaded result? Why? Who knows why? It's called an outlier.

I use rules when looking at user submitted results.
-Discard outliers
-Look at the average peak results, because its less likely tainted with wrong software/hardware configurations
-If individual results are low, look why. Single vs Dual channel memory? CPU underperforms which hold back the GPU? The result itself sucks?
-Again, discard outliers. 5-10% variations are fine. If its 30-40% faster either way then its wrong.

Same thing can be done with the GPU.

The Iris Plus and Vega 10 configuration gets 45-50 points for the GPU in 3D DX9. ~70 points for 4800U is 40% faster. Get it now? The UHD 620 in the Whiskeylake 8565U gets 22-25 points for the GPU. Again, reflects real world results. I see some UHD 620 results with 40+ GPU points. What do I do? Trash bin! Probably malfunctioning optimus switching driver is keeping an MX 250 on or something. I don't care!

Most poor results are due to badly configured system or using single channel memory. Yes, it affects the CPU too.

I didn't say things in detail because I try to give people benefit of the doubt, and assume people know all these things.

So absolute performance? Because that's what you get when you can't control for clockspeed.

Yes. That's all we can do. 2/4/8 core results can't be taken seriously because scaling up to "64 cores" means in userbenchmark land they are not talking about cores but threads.

And that's what matters in a laptop anyway. You are buying a whole system, so you can have the best CPU/GPU, but you might not consider it for $600 because the design is straight from the late 90s, the battery life lasts 1 hour idling, the keyboard has a weird layout, the WiFi is flaky, and the whole device feels like cheap plastic and the screen only has a 90 degree FOV and its brightness is only at 200 nits.

You are not going to get IPC Performance per clock comparisons using a laptop.
 
Last edited:

uzzi38

Platinum Member
Oct 16, 2019
2,556
5,531
146
Well, the you've got the basic reasoning down, I'll grant you that, and you're justified to have made the choice you did. But, there's some things I want to point out. Firstly

I discard anything in between because who knows what's logical threads and what's single thread?

Schedulers always prioritize physical cores over logical cores. When it says single core - it means single thread, and that single thread will - for the most part - be running on a single physical core, provided the user isn't performing a different heavy load in the background.

The 4800U gets 138 points in ST and 1074 points in MT. Do you really believe the 1065G7 is only 10% faster in ST compared to 3700U? The compared Icelake system is underperforming. The ~30% difference approximates the high end of the single threaded difference we see with real world systems.

Than the 4800U you mean? Because the average 3700U is listed as being 44% slower than the 1065G7 and the best performing ones are 31% slower.

See: https://cpu.userbenchmark.com/Compare/Intel-Core-i7-1065G7-vs-AMD-Ryzen-7-3700U/m888368vsm765724

If we're talking the 4800U, then yes, I do believe the 4800U will be at worst 10% slower on a single thread than ICL-U on two systems that allow for the same power budget etc, and let me back that up too.

Firstly, it's important to note that unlike Geekbench, Userbenchmark does actually scale quite well with extra cores. It's pretty much linear in fact, but the workload is so light it tends to exhibit greater SMT scaling than you'll often find in different tests. That's me getting off track a little, so back on track again:

Most Ryzen laptops are configured with their cTDP set between 22 and 25W. This laptop is no different, and if you look at the chassis it appears to be in, you can be certain it's not allowed a greater power budget. It simply wouldn't be capable of cooling it.

So in any case, we have a 4800U device that is most likely configured to run at 25W and maintains near perfect scaling going from a single thread to dual threaded workload. What does that imply to you? Because to me, what it imply's is the the silicon is capable of maintaining it's 4.3GHz boost (which I'm guessing will be the same for the 4800H and 4800U - the 4800H's 4.3GHz boost was already leaked by TUM_APISAK). If nothing else, what can we most certainly agree on? That's right, that at least one thread on the Renoir system is achieving it's maximum boost.

Now, remind me, what IPC advantage does Zen 2 claim over Skylake? That's right, about ~3%. Let's assume the cut cache on Renoir removes that advantage (remember, the vast majority of Zen 2's IPC uplift comes from the TAGE Branch Predictor - see Wikichip's article of Zen 2 from it's launch for that point).

So, what does Renoir have over the 1065G7? Well, erm, about an extra 400mhz in max turbo. What does the 1065G7 have over Renoir? Well, lets just say the full 18% IPC advantage. Knock off 10% of that advantage for the 10% higher boost clock on Renoir and oh would you look at that, Renoir has a <10% lead. How shocking.

Now, I briefly touched upon the topic of power efficiency, and this is precisely where I want to point out neither Ice Lake nor Tiger Lake have even a dream of touching Renoir. Oh, and don't get me wrong, I don't mean battery life, alas, it's too soon to talk about that. Power efficiency on the other hand, well I pointed out how Renoir appears to be maintaining (or extremely close to maintaining) it's maximum clocks in workloads that push two threads when provided a 25W power budget?

Well, let's take things a step further. Ice Lake is incapable of maintaining a single thread's 3.9GHz boost within a 15W power budget. With a 25W power budget, the gap between Ice Lake and Renoir will shrink when it comes to ST workloads, because, well, Renoir doesn't need that power budget. More than the RAM, or user error with Userbenchmark (as there are no 1068G7 devices that ship with single channel RAM to my knowledge and I didn't use it as a comparison point on the GPU side anyway), one of the biggest discrepancies with differing ICL-U results is actually because of this. Devices like the new Razer Blade Stealth simply cannot actually boost to their max on a single core.

On the other hand, we have Renoir. And this will be one of the biggest selling points of the chip - the sheer amount of work that Renoir will be capable of doing in limited power and thermal budget situations (read: LAPTOPS) gives it one hell of a niche. Back when the SMT disabled 4700U leaked, I wanted to simulate Renoir's power efficiency ahead of time. Take a gander at these: https://cdn.discordapp.com/attachme.../Screenshot_20191230-225505_Samsung_Notes.jpg

That's correct, within a simple 15.5W power budget for all 6 cores in my dirt-tier silicon bad 3600 (which is fully incapable of even a 4.1GHz all core OC even at 1.325v), each core was capable of maintaining 3.2GHz during Cinebench R20. On the other hand, we have ICL-U:


Yeah. Oh, and if you don't believe the table, then I can happily point you towards Hardware Unboxed or alternatively LTT tested in Blender.

Within a limited power and/or thermal budget - which is laptops as a whole in a nutshell, Renoir will outperform it's competitors. Simple as. So I wouldn't exactly call a <10% ST lead positive in any regard. Also, it'll probably the best thing since sliced bread for the -H market, imagine gaming laptops that don't throttle for once! Now that's the dream.

Well, for me personally, I'm just happy to see someone else AMD hit the mobile market hard like they will with Renoir. About darn time they had a product actually capable of competing.

PS: I dunno why you rambled about the iGPU, I only provided comparisons on the iGPU side to the 3700U and to the MX250, and it outdid both. I didn't even compare Gen 11, figured beating the MX250 was enough. And it did so on both the average and best results. And this isn't going to be Renoir's best showing on the iGPU, no that will be when LPDDR4X is in play.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Man, I don't know why you are going off on a tangent, especially when you ended up agreeing with my original comment(which you tried to counter). Your response was specifically to my comment:

The GPU in the particular system is 40% faster than the GPU in Picasso.

The CPU is 10-15% better in multi-thread than the one in Cometlake, and 20-25% faster than Picasso in single thread. Icelake is 8-10% faster in ST, but is behind due to having only 4 cores.

That's why I mentioned the GPU. Did you forget? You also responded to my reply with bunch of Icelake results that are underperforming. And replies by others agree with you based on that.

Schedulers always prioritize physical cores over logical cores. When it says single core - it means single thread, and that single thread will - for the most part - be running on a single physical core, provided the user isn't performing a different heavy load in the background.

Did you skim through my reply? I admit I type long replies. I said nothing about schedulers. It's about the userbenchmark itself. Here:

It's same with userbenchmark. The top 1065G7 gets ~145 points in ST, and ~650 points in max thread. I discard anything in between because who knows what's logical threads and what's single thread?

Obviously we can all see that userbenchmark's "8 cores" isn't really 8 physical cores otherwise the max, "64 cores" won't improve. There's no way to distinguish between cores and threads when the program supports anything between 1 and Max. For simplicity's sake we take note of ST results, and MT.

as there are no 1068G7 devices that ship with single channel RAM to my knowledge and I didn't use it as a comparison point on the GPU side anyway

There are laptops beyond the Spectre and XPS you know? You just didn't look hard enough.

The HP Pavillions, Inspirons, and the unnamed Lenovos you can find with Icelake. Many don't use LPDDR4x memory. Heck, many don't even include a second memory channel! Yes and they are DDR4-2666. And I can find them in userbenchmark too. Just look at bandwidth figures. And they will impact CPU performance.
 
Last edited:
  • Like
Reactions: Tlh97 and IEC