Ryzen "B2" Stepping a False Rumor (apparently)

F

Formula.350

[H]ard|Gawd
Joined
Sep 30, 2011
Messages
1,102
Figured I'd pass this along... seeing as some would be rather bummed if a new stepping came out.

B2-or-not-B2-that-is-the-question.png

Original post linkage.


For anyone not familiar with chew*, he's a veteran extreme overclocker (now retired due to how the industry treats those guys), but more importantly he's a true-blue lover of AMD and used to be one of their top go-to guys for overclocking and events. Still has a couple records on the CPU-Z rankings. (As does The Stilt)

I trust chew*, but I also trust The Stilt, so it's hard to say whether or not a new stepping is due out soon, or not. I'd love to quote this post from Stilt if I could find it again, but around a month ago or so he had mentioned a revision to the Ryzen silicon that would address some things (pretty sure he didn't mention what), and he thought it'd be out in a couple months. He is also a big AMD guy and is in their good graces, having a Ryzen ES prior to launch...

All in all it's hard to say what the case may be. I wouldn't doubt the validity of a B2 Stepping existing and functional, but could very well be a case of AMD making improvements and didn't feel it was significant enough of one to replace the current stepping, or adjust the product stack in order to fit a new chip in (or just didn't want to get anyone's panties in a bunch over releasing a new stepping heh).

I trust what both of them say, to the point I would put what they claim right up there with that which Kyle tells us. I get the impression they are both right, and that sure there are new stepping chips, but they're not going to be released anytime soon. If I had to bet, I'd put my money on a new stepping debuting along side Threadripper, as otherwise it could potentially eat into the sale of those when they come out.
 
If you looked at something as Phenom line you would see that most of the things that needed improving badly were introduced in the Phenom II . Currently there is much going on but revisions of Ryzen will happen eventually that does not mean that everything has to change, same process limitation not being able to get higher clock speeds and so on .....
 
It's possible they got confused and the new chip being worked on is the next year Zen+ and not a B2 stepping. If they are planning a early launch next year, this would be the time to start running the fab for it. But who knows but the fact that Canard had 0 info on anything tells me they have no clue either and just made something up to sound good.
 
June has a full 30 day run a Global Foundries in june for 14 nm LPP. hoping for some process improvement.Stock base speed of 4ghz would be nice
 
BeepBeep2 said:
Just as chew* already replied to me, you are correct.

However, the initial CPC rumor never mentioned "Ryzen", only "Zen". It seems the B2 revision will be used for Epyc, at least for now.
Click to expand...

Problem is Canard just made crap up about bugs and fixes, he guessed and he was wrong. Why AMD calls EPYC B2 is only known to AMD.
 
Gideon said:
Problem is Canard just made crap up about bugs and fixes, he guessed and he was wrong. Why AMD calls EPYC B2 is only known to AMD.
Click to expand...
The validity of those statements remains to be seen as AMD has not yet released a public Revision Guide for Family 17h containing the known Product Errata list
 
If I had to guess B2 is uncore fixes to PCIe/Infinity to make multi die communications bug free

So there is no B2 for AM4 Ryzen.... or at least you'd never know the difference
 
ecmaster76 said:
If I had to guess B2 is uncore fixes to PCIe/Infinity to make multi die communications bug free

So there is no B2 for AM4 Ryzen.... or at least you'd never know the difference
Click to expand...
That would be my guess as well, based on the replies on the tweet. News sites ran with it because they have been spending the last several months correlating Naples and Summit Ridge being "the same die". chew* was addressing the rumors he saw that used modified language in comparison to the original statements from CanardPC.
 
You could just ignore "news website" to "sites which need hits" . Now everything gets reported as truth because it is canard PC.
 
BeepBeep2 said:
That would be my guess as well, based on the replies on the tweet. News sites ran with it because they have been spending the last several months correlating Naples and Summit Ridge being "the same die". chew* was addressing the rumors he saw that used modified language in comparison to the original statements from CanardPC.
Click to expand...

Naples and Summit Ridge definitely are the same die. No doubt about that at all. They are not, however, the same product. Ryzen only uses a very small subset of the uncore features that Epyc will have enabled
To name a few
  • External Infinity fabric (unless Vega will use that with Ryzen )
  • built in power regulation
  • programmable power modes
  • memory encryption
  • ARM security coprocessor
All that stuff is very new and any bugs on server parts would be devastating. So if B2 only touches that stuff then indeed there is no B2 Ryzen per se even if they switch to new masks
 
Formula.350 said:
Quite certain it does actually, as that seems to be a feature of this all that AMD has been hyping.
Click to expand...

But wouldn't that only be a feature on the parts that have integrated graphics (blanking on the name ATM)? That would have to be a separate die anyway. Outside that, I don't see any way that would apply to Vega, unless those parts are supposed to be MCMs like EPYC/Threadripper. Of course, that may be how they're doing that and I missed the news/rumours/"news" pointing to that.
 
That would be highly unlikely at 99.9% yields, the process has been going smooth for AMD but that is over the top.
 
carbide said:
But wouldn't that only be a feature on the parts that have integrated graphics (blanking on the name ATM)? That would have to be a separate die anyway. Outside that, I don't see any way that would apply to Vega, unless those parts are supposed to be MCMs like EPYC/Threadripper. Of course, that may be how they're doing that and I missed the news/rumours/"news" pointing to that.
Click to expand...
I'll try and dig up what it was, but currently I was under the impression that it piggybacks off the PCIe bus and that's how it works, being part of AMD's HSA like the R9 285 + A10-7850 + 88X combos kicked off (that developers seem to not have much interest in, annoyingly enough... am hopeful that Ryzen's success will turn that around.)

EDIT: Alright so in full disclosure I will admit that my brain may have read too far into all of this and been assuming that all the talk of Ryzen (all Zen-based products) and Vega both using the Infinity Fabric, was also implying/hinting at them also utilizing it for some level of communication to more or less unify them, despite being individual products (as apposed to an integrated part like the APUs).
That being said, seems as though S|A also was getting that same impression with Ryzen+Vega, as per the last paragraph:
http://semiaccurate.com/2017/01/19/amd-infinity-fabric-underpins-everything-will-make/

Then there's also this:
aYDcW0N.jpg


However... while at first blush it looks to do exactly what I'm hoping and was saying, I get the vibe that it isn't quite talking about Consumer platforms, at least in the conventional way that we've grown accustomed to. Yes, it says that a CPU (center) can link up to another CPU or GPU over Infinity Fabric, BUT it seems to be implying that this would be a Multi-Socket computer; thus, the dGPU would be in a socketed package instead of a PCIe-slot card.


And yet... I can't help but still be hopeful due to them adding things like the High-Bandwidth Cache Controller on Vega that can tap into system RAM or Storage (SATA, m.2)
This edit is already over an hour in the making while I tried to find pages with info I wanted, but I'll just plop this link here in hopes it has the info :\
http://techreport.com/review/31224/the-curtain-comes-up-on-amd-vega-architecture
 
Last edited:
At least on Epyc the same PCIe pins get used for IF on the 2S systems so they definitely share resources.

The trick will be is the socket and traces of a typical x16 slot compatible with the IF protocol and does Vega have the same flexibility?
 
lolfail9001 said:
Proceeds to correct it to 98% after understanding that he went overboard with bullshit.
Click to expand...


I mean, some of these chips are running XFR voltage over 1.55v and is barely stock stable at 1.35v. If you give it more voltage they may have good yields because the tolerances are extremely loose. Who knows, some chips are probably being binned as lesser core models etc so I would believe it. Probably getting a lot of Ryzen 3 stock out of those yields.
 
Rvenger said:
I mean, some of these chips are running XFR voltage over 1.55v and is barely stock stable at 1.35v. If you give it more voltage they may have good yields because the tolerances are extremely loose. Who knows, some chips are probably being binned as lesser core models etc so I would believe it. Probably getting a lot of Ryzen 3 stock out of those yields.
Click to expand...

XFR at 1.55V means single core boost , so that is pretty normal ?

Stock stable depends on what you are talking about so far I haven't seen anything that was unstable at default stock clocks with normal voltage.

About Ryzen 3 , it does not sound like anything else then shut down cores on Ryzen 7 not a separate production line, which means the Ryzen 3 is prolly a paper product and would not see much volume there.
 
I have seen some strange things in terms of voltage on multiple cpus and don't really know what to make of it.

Motherboard: Asus Prime B350 Plus and also X370 Pro, even the latest beta bios for each

cpu 1 - Ryzen 7 1700 - At stock speeds, all bios settings default, sometimes I would see voltage hit 1.35V. Asus has an auto overclocking thing in the bios. One click and you are set. With this enabled, the cpu was set to 36.5 multiplier and voltage was 1.1875V. So faster default speed and lower voltage. Seemed stable, no funny crashes, so I left it like this for months.

cpu 2 - Ryzen 5 1500X - At stock speeds, bios at defaults, sometimes it would hit 1.475V. I didn't like this and remembered the auto overlocking feature. One click in the bios later and it was overclocked to a 38 multiplier with 1.2375V. So faster overall and much lower voltage. No issues like this.

I don't think these cpus really require the voltages that the bioses are feeding them. Maybe AMD just needs to update the AGESA or something again to stop overvolting the crap out of these poor things.
 
The numbers I posted are from the bios and Ryzen Master only. They seemed to agree too. That doesn't mean they are correct, but it seemed believable to me at the time.
I agree with your comment in general, most other things are clearly way off. Cpu-Z is all over the place and shows numbers that are way too low to be real and those numbers don't agree with anything else. I haven't tried HWmonitor.
 
pgaster said:
The numbers I posted are from the bios and Ryzen Master only. They seemed to agree too. That doesn't mean they are correct, but it seemed believable to me at the time.
I agree with your comment in general, most other things are clearly way off. Cpu-Z is all over the place and shows numbers that are way too low to be real and those numbers don't agree with anything else. I haven't tried HWmonitor.
Click to expand...

I was a little baffled about this as well , saw some of the videos Chew posted and he even has boards standing on the side to be able to manually measure the voltage. And he tested some of the motherboard with taking the voltage to 1.4V on cpu he tests and see how much it really was from the pins on the back.

Even in the bios I have put values as 1.375V and it would not read this way if I was looking at the display (in the bios) which shows the voltage for dram. It would be scary to think that when you are setting voltages and it displays another number you did something wrong entering it.

But to comeback to the XFR value it really means nothing if the cpu scales back the other cores to lets say less then 1V who would care if there is one core running at 1.55V ?
 
pgaster said:
I have seen some strange things in terms of voltage on multiple cpus and don't really know what to make of it.

Motherboard: Asus Prime B350 Plus and also X370 Pro, even the latest beta bios for each

cpu 1 - Ryzen 7 1700 - At stock speeds, all bios settings default, sometimes I would see voltage hit 1.35V. Asus has an auto overclocking thing in the bios. One click and you are set. With this enabled, the cpu was set to 36.5 multiplier and voltage was 1.1875V. So faster default speed and lower voltage. Seemed stable, no funny crashes, so I left it like this for months.

cpu 2 - Ryzen 5 1500X - At stock speeds, bios at defaults, sometimes it would hit 1.475V. I didn't like this and remembered the auto overlocking feature. One click in the bios later and it was overclocked to a 38 multiplier with 1.2375V. So faster overall and much lower voltage. No issues like this.

I don't think these cpus really require the voltages that the bioses are feeding them. Maybe AMD just needs to update the AGESA or something again to stop overvolting the crap out of these poor things.
Click to expand...
According to The Stilt, these weren't really intended on being much more than 3.2GHz chips, and would've had much lower voltages than we're seeing.

I think a pretty good case-in-point on that is earlier today I was futzing with my setup and had downclocked the CPU to 22x and set the VID voltage (PState) to 0.9V (which is default for it's P3 state). I was able to drop it down to 0.85V (which was, under load, 0.812V), and RAISE the CPU multi up to 2.7GHz, all while under Prime95 load on all cores. At 2.8GHz it BSoD's as soon as I applied the multiplier. So I'd speculate that 2.65GHz would've been perfectly stable at 0.815V! That's pretty damn impressive IMO, for an 8C/16T chip, under full load, at DDR4-3333 (CPU-NB was 0.971V, for anyone curious).

Most of the time my chip is using 1.212V at 3.5GHz (which is XFR speed for my 1700X) and thermals are well under control at 70C (50C Tdie, ambient roughly 23C); however, if I move it to 1.35V which is what it'll set to in "Overclock Mode", those temps soar to 77C.
 
Those are about the thermals my 1700X had at stock. When I pushed to [email protected], it jumped to around 70°C, and the little extra to [email protected] (well, +.05V offset, anyway) moved the temp to almost 80°. I'm debating whether to keep it that high or drop it back to 3.8 or 3.85. I don't /think/ my cooler is the issue; a H100i V2 should be sufficient, as it kept my Xeon W3670 under control when overclocked.
 
carbide said:
Those are about the thermals my 1700X had at stock. When I pushed to [email protected], it jumped to around 70°C, and the little extra to [email protected] (well, +.05V offset, anyway) moved the temp to almost 80°. I'm debating whether to keep it that high or drop it back to 3.8 or 3.85. I don't /think/ my cooler is the issue; a H100i V2 should be sufficient, as it kept my Xeon W3670 under control when overclocked.
Click to expand...

Unless you have some special software installed , the offset for 1700X (or any X Ryzen 7 or 5 cpu from AMD) is -20C :)
AMD have done that offset to allow fan speed to ramp up faster.
 
Last edited:
carbide said:
Those are about the thermals my 1700X had at stock. When I pushed to [email protected], it jumped to around 70°C, and the little extra to [email protected] (well, +.05V offset, anyway) moved the temp to almost 80°. I'm debating whether to keep it that high or drop it back to 3.8 or 3.85. I don't /think/ my cooler is the issue; a H100i V2 should be sufficient, as it kept my Xeon W3670 under control when overclocked.
Click to expand...
Pieter3dnow said:
Unless you have some special software installed , the offset for 1700X (or any X Ryzen 7 or 5 cpu from AMD) is -20C :)
AMD have done that offset to allow fan speed to ramp up faster.
Click to expand...
Yep, like Pieter says, and that's why I mentioned the 50C Tdie reading since that's 'actual'. I hate that they've done it, and more so that the board makers don't just mask that temp and report Tdie instead. Oh well. heh
 
Oh, I know all about that. That 80° is with the offset taken into account. I'm using HWiNFO64 to monitor temps for the most part, though I did find the correct temperature value buried in AIDA64's temperature sensor list to display when running the stress test. The average value after a 7.5 hour stress test run was 78°C with an ambient temp around 25°C.
 
Oh, dang! :eek: See, I've never had my Tdie value go over ~65C (so 85C 'Ryzen X' Temp). My Water2.0 PRO is only a 120MM (140MM? whatever) Rad, but it's double the thickness, so is sort of like a 240MM, but not really since the water has a much longer time to be cooled, which is why you should be seeing much better temps than me! I only have minimal heavy load time on mine at 3.8GHz though, and my voltage also was 1.35V.... so I really don't know for certain what the temps would eventual reach after even a couple hours, let alone 7. I figured after an hour mine was properly heat soaked, but perhaps not *shrug*

I can't help but think perhaps one of these may be in play...
- your pump's cold plate is too convex or concave, causing improper contact with Ryzen's super flat heatspreader
- your mounting pressure is uneven and causing improper contact
- your thermal compound isn't spread even and not so heatspreader and cold plate aren't making any contact in those air pocket spots.

I used a razor blade to spread my paste on and did my best to get the most even application I could, and even I've been questioning how well that is, since I didn't bother to remove it and check. I mean temps don't skyrocket, so assumed it was pretty good. Though to be fair, mine is on an open-bench, not in a case, so I definitely have airflow on my side.
 
Yeah, I didn't see a big temp difference in open case vs. closed case; the H100i is acting as the case intake with a 120 (rear) and 140 (top) as exhausts. The liquid loop temperature didn't exceed 45C, if that tells you anything. The next thing I'm going to do is try remounting the pump/cold plate. I did the thin vertical line of paste technique originally as it's worked for me in the past, but I think I'll try pre-spreading it instead and see if that helps. I'm leaning against it being an issue with the cold plate itself, since it cooled an overclocked Xeon with a higher base TDP relatively well scarcely more than a week ago.
 
Yea I'd do the spread technique (I spread it on the heatspreader since that's the smaller surface).

Also AMD and Intel TDPs are not comparable. They classify them totally different and can only be compared to products from the same manufacturer unfortunately.
For example, 90W on Ryzen doesn't mean the same thing as 90W on an Intel. Unfortunately, due to the differences, I haven't even bothered to remember what the differences are lol Highest I've had my Ryzen is like 170W actual power consumption (Volt x Amps for Core and SoC[NB], adding those values together), which isn't the same as the "Package Power" that is displayed either. I don't frankly know WTF the Package Power refers to, if it's how much heat it's implied to be putting out or what, but sometimes it's higher than the total power, but most of the time it's quite a lot lower.

EDIT: Here's an example of what I mean in regards to Package vs Total (bottom)... Note the 40+ watt gap.
Example Readings.png
 
Last edited:
Formula.350 said:
Yea I'd do the spread technique (I spread it on the heatspreader since that's the smaller surface).

Also AMD and Intel TDPs are not comparable. They classify them totally different and can only be compared to products from the same manufacturer unfortunately.
For example, 90W on Ryzen doesn't mean the same thing as 90W on an Intel. Unfortunately, due to the differences, I haven't even bothered to remember what the differences are lol Highest I've had my Ryzen is like 170W actual power consumption (Volt x Amps for Core and SoC[NB], adding those values together), which isn't the same as the "Package Power" that is displayed either. I don't frankly know WTF the Package Power refers to, if it's how much heat it's implied to be putting out or what, but sometimes it's higher than the total power, but most of the time it's quite a lot lower.

EDIT: Here's an example of what I mean in regards to Package vs Total (bottom)... Note the 40+ watt gap.
Click to expand...
135 + 8 = 143 ("core" + soc = total)
135 - 100 = 35 ("core" - "package" = "?")
I'm thinking "package" here is uncore maybe, "core" is package, and "?" is core (else swap core and uncore above). I have no documentation and little experience to back that up, however.
 
I actually came across a small big of info regarding this last night, by sheer happenstance...

Originally Posted by Mumak

OK, so here another update: www.hwinfo.com/beta/hw64_547_3118.zip
Now HWiNFO can measure the voltage, current and power of the CPU and SoC rails straight from the VRM via telemetry.
Look for new values under the CPU sensor with (VDDCR_CPU) or (VDDCR_SoC).
Note that (VDDCR_CPU+SOC) is the sum of both CPU and SoC powers and doesn't reflect the total CPU package power as there are some additional rails (like PCI-E Phy, Standby, etc.) which can't be monitored.
The original "CPU Package Power" power probably adds some constant power for those unaccounted rails, so the Sum power might sometimes appear less than "CPU Package Power".
The advantage of these new values is that they should be much more precise than the "CPU Package Power" or ASUS EC "CPU Power" values, which in certain cases provided unrealistic values.
Another important note is that due to certain (yet unresolved) system specifics, these new values should work accurate for some boards, while some others might not (yet) give accurate values. Currently C6H should be OK, probably ASUS PRIME X370 too, ARICA board seems to work well too. I'll work more on this if I get required information how to make this reporting more universal for other boards.
Let me know how it works or if there are any issues.
Also please note, that I can't follow all posts in this thread, so if you want to make sure I see your post, please mention me or quote.
Click to expand...

I know that the recent updates to HWiNFO seem to have focused on improving the ASUS and MSI readings (as per the changelogs). However, I don't know what exactly that means for other boards such as from Gigabyte or BioStar. I'd suspect that the top models have had focus, as well as possibly the best-sellers. Nevertheless, little has appeared to change in regards to my Titanium, and a few readings leave me wondering why they're like that... Mainly the 'confusing' part for me is the fact of the VRM voltage output and CPU Core Voltage (VDDCR_CPU) align perfectly, or at least within a hundredth of a volt of each other, whereas the Motherboard column's VCore reports the same voltage that CPU-Z does and is waaaay different at idle and reads roughly 0.35-0.45V higher when under load.
As an example...
IDLE: the VRM would display 1.44V, the VDDCR_CPU would be 1.45V (yes, higher *shrug*), and Mobo VCore would be 0.468V (it usually jumps around, but .500 ±0.075V).
LOAD: VRM = 1.213V, VDDCR_CPU = 1.215V, MB VCore = 1.250V

PERSONALLY I would expect that the VRM would match up with the MB VCore since that's what I'd expect the PIN'S voltage to be at. Then, the VDDCR_CPU would be the one that's currently listed as Motherboard voltage, since that, when idle, drops to half a volt and thus I'd expect to be what the CPU's INTERNAL voltage is being configured to.

Granted, that wouldn't explain why the VCore reads 1.25V when the VRM is supposedly outputting 1.213V, unless the CPU is using it's internal Voltage Regulators to boost voltage by using some of the Amperage and converting it to the additional ~0.035V


Getting back on topic though... by my math, the CPU Package power should always read higher than the Total (CPU+SoC), based on what he said "Note that (VDDCR_CPU+SOC) is the sum of both CPU and SoC powers and doesn't reflect the total CPU package power as there are some additional rails (like PCI-E Phy, Standby, etc.) which can't be monitored. The original "CPU Package Power" power probably adds some constant power for those unaccounted rails, so the Sum power might sometimes appear less than "CPU Package Power".", assuming we really are figuring that there is some overhead calculated into it. And yet when at high load, as shown, it's waaay lower. I can only imagine that it is implying how much HEAT it is putting out (or at least a rough approximate). The highest I've ever managed to get the Package Power to reach (so far) is running just the AIDA Cache stress test, and then it'll hit upwards of 125-130W. I'll have to try and see what it reads for Total in that circumstance, as I don't recall off hand. There are times when the Package and Total are nearly identical I've noticed, and it has been when things are loaded in a way to result in around 110W.

Who knows, maybe AMD has detailed it in one of their Blog postings and I've just overlooked it heh Don't believe Ryzen Master monitors that sensor though, so outside of technical documents I don't think they'll make mention of it (which I believe are still on a by-request-and-approval only basis), since the average joe won't be looking at it through anything provided by AMD.
 
Well, I did screw up on the thermal paste application; though since I centered the line on it when I applied it, I'm wondering if you may have a point with the surface of the cold plate being uneven. I'm out of the paste I used when I applied it the first time (Cooler Master ThermalFusion 400), so I applied the only thermal paste I had left (AS5) evenly on the heatspreader and put it back together. My temps are *slightly* better, so I think I'm going to let it sit another day or two and then pop it back off and do a more thorough inspection of the cold plate surface and the mounting hardware in case something is off. I was planning on getting a better cooler at some point anyway, so I may just have to start planning to pull that forward a bit. I've had my eye on the alphacool Eisbaer 280.

3nF5hYa.jpg


On the power subject though, I was running a P95 test last night and decided to keep an eye on that number while it chewed away. At 3.8GHz, 1.3V (which eventually proved unstable), the total power was averaging 180W with periods of "swingyness" if the threads were transitioning between tests. During the 180W phases, Tdie was hovering around 70°C, the liquid temperature was around 38°C, and the room temperature was ~24°C. When the power was swinging, Tdie dropped down to ~66°C but the liquid temperature didn't move much.
 
Well based solely on your heatspreader it indeed wasn't making sufficient contact heh
I believe that AS5 still needs to "cure", and thus will require some thermal cycling to get it to that point, then it's temps will be better. Hopefully you had cleaned off both surfaces before applying the AS5?

As for the power, yea, when a test completes, the load drops a tad and causes both the temps and current (thus Total Power) to drop a bit.
The instability would depend on which Prime profile you used. It defaults to Blend with only 2GB, so if that's what was used then it could easily be the RAM as the cause (given you do have 4 sticks, running at 3200, a feat in itself for most). So changing it to the first test profile which doesn't touch RAM, would be needed to tell whether it was the CPU or the RAM / IMC to blame.


I did some testing of my own yesterday and concluded that mine is (y)(y)(y) for 3.7GHz with VID set to 1.25V (VID is also what Ryzen Master sets for voltage). Under a Prime Profile 1 LOAD, in HWiNFO, under the Ryzen section the Core Voltage was only 1.181V (same with CPU VRM), though the Motherboard VCore was reporting 1.225V. Idle that raised up to 1.237V on the Ryzen reading, but MB reading went into Cool'n'Quiet mode and was 0.482V. Either way, I'm not entirely sure what I'd have to input for a BIOS voltage, but I suspect it'd be around the same 1.25V I input into Ryzen Master (which is the same as K17TK's VID).

Nevertheless, the difference in temps and total power were significant. I started out at 1.337V on the Ryzen Voltage reading, which under load was probably ~1.31V, and temps were 75-76C with power at 173W. After tweaking voltages (all while Prime was still running), temps dropped to ~54C and power consumption to ~150W :)
 
You must log in or register to reply here.
Back
Top