Paper Launch for AMD Ryzen 9 3900X

Well, I was at Microcenter for a cable, ended up buying a water block. Anyway, the Dallas Microcenter is out of Ryzen 9 3900X's. Again.
I’ve had a 3900x on preorder with B&H for a week and a half now .Expect to wait a couple of more weeks. I know B&H Blankets the web with adds but doesn’t actually stock these and it takes a while to get but they do allow preorder so. At tleast I don’t have to monitor stock morning, noon and night. Here iin Western Massachusetts we have no local high tech option sbut best buy who I’d rather not do business with
 
i'm seeing a 100% improvement in many workloads over my 1800x where smp scaling can occur. Not bad for 50% more cores == 100% more performance. Single core performance is a massive improvement too in most cases.

I was not expecting such a drastic difference. Insane L2/l3 cache helps i guess.

I wish i could do something about the full load temps but it looks like i'd have to go custom water loop to make much of a difference. Idles at 29c and throttles at 76-78c @ 4.2Ghz across all cores. Would be interesting to see what kind of cooler is required to keep temps below 70C at full load with ambient temps around 75-80f.
 
I’ve had a 3900x on preorder with B&H for a week and a half now .Expect to wait a couple of more weeks. I know B&H Blankets the web with adds but doesn’t actually stock these and it takes a while to get but they do allow preorder so. At tleast I don’t have to monitor stock morning, noon and night. Here iin Western Massachusetts we have no local high tech option sbut best buy who I’d rather not do business with

Yeah, but BestBuy has them in stock very often and now they restricted them to local pickup it seems, and if that's the case, you should be able to pick one up pretty quickly. Good luck!

i'm seeing a 100% improvement in many workloads over my 1800x where smp scaling can occur. Not bad for 50% more cores == 100% more performance. Single core performance is a massive improvement too in most cases.

I was not expecting such a drastic difference. Insane L2/l3 cache helps i guess.

I wish i could do something about the full load temps but it looks like i'd have to go custom water loop to make much of a difference. Idles at 29c and throttles at 76-78c @ 4.2Ghz across all cores. Would be interesting to see what kind of cooler is required to keep temps below 70C at full load with ambient temps around 75-80f.

What make/model CPU cooler do you have? What motherboard? Keep an eye on CPU voltage as that's the main cause of overheating if it gets too much of it at stock settings.
 
i'm seeing a 100% improvement in many workloads over my 1800x

This sounds like you had a broken 1800X... that's still a damn fine CPU. Though perhaps we can give you a pass because the numbers might be close in some cases (more cores + more clockspeed + more IPC) that apply to your workload, and because you're justifiably excited about using a new CPU?

;)
 
  • Like
Reactions: STEM
like this
Microcenter in Parkville/Baltimore had more than 10. I decided to be more practical and bought a 3700x
 
This sounds like you had a broken 1800X... that's still a damn fine CPU. Though perhaps we can give you a pass because the numbers might be close in some cases (more cores + more clockspeed + more IPC) that apply to your workload, and because you're justifiably excited about using a new CPU?

;)

I got my 1800x on launch day. I think some workloads just love the extra cpu cache. I'm sure if the 1800x had the same amount I'd be seeing a much closer comparison that reflects the 50% core count increase for 50% better performance + the ipc improvements.

I've seen the effect even on the 1800x with certain parallel operations. Where encoding ogg files on all 16 cores is faster per core than the speed you'd see doing them one at a time.
 
Well, 3900X is not like 3800X. Tried 4.4ghz @1.37 @1.38 @1.40. All BSOD. I'm kinda sad, but I wasn't going to do any permanent ocing in the first place.

That sucks. I finally finished my 3900x build. Got it to [email protected]. I tested 4.4ghz but it needed more than 1.35v and eh, it wasn't worth it to keep pushing.
 
People in Canada. Canada computers have plenty. Just ordered one

And the plenty is now none at all. Thanks for blabbing in front of the internet. Oh well... only 20-30 more sleepless sleeps until the 3950X is out of stock everywhere too. :D
 
And the plenty is now none at all. Thanks for blabbing in front of the internet. Oh well... only 20-30 more sleepless sleeps until the 3950X is out of stock everywhere too. :D


AMD might as well offer a even bigger chip say a 3975x with s 20 cores and sell it for 200 bucks. What's to lose , they won't actually have to have any available for sale just have a sku for it and release it. How can a company talk and talk about a line of CPU's for over a year, then release it and have minimal product actually available for sale unless it's a marketing strategy or they actually have problems making them. I'll say it again AMD has been responsible for selling more Intel 9900K's in the last month than iNTEL has been. It's still faster and quite available
 
AMD might as well offer a even bigger chip say a 3975x with s 20 cores and sell it for 200 bucks. What's to lose , they won't actually have to have any available for sale just have a sku for it and release it. How can a company talk and talk about a line of CPU's for over a year, then release it and have minimal product actually available for sale unless it's a marketing strategy or they actually have problems making them. I'll say it again AMD has been responsible for selling more Intel 9900K's in the last month than iNTEL has been. It's still faster and quite available

Uhmm, the 3800X is the direct competitor to the 9900K. And the 3800X is available anywhere for nearly $100 less and the $330 3700X is basically the same chip with a similar supply. There not selling more 9900K's because of 3900X not being available.. The 3900X specs very close to the 9920X, not the 9900k, should you want the Intel counterpart it's going to cost you $1,200

Simple fact is, the 3900X is basically 2 high binned 3600x dies, so they have to sort dies and test them to make their top of the line chip, there making a lot of 3600's and less 3900X's through binning. This combined with the 3900X being the fastest consumer chip that you can buy relatively speaking, demand exploded. Stock is moving in and out quickly because that many are selling
 
I bet those scalpers are having a hard time moving stock. Right about once per week pre order or stock available comes up for it. Anyone who wants one can get one pretty easy right now.. Albit, even at a $629 price it would be fairly priced I think
 
I bet those scalpers are having a hard time moving stock. Right about once per week pre order or stock available comes up for it. Anyone who wants one can get one pretty easy right now.. Albit, even at a $629 price it would be fairly priced I think

Thank God for credit cards, right? Because if people couldn't slap it on a card and pay it over time, I doubt that they would think that paying anything over MSRP is a fair price. Even if I wanted to be understanding that those scalpers have to pay 10% to eBay and 3% to PayPal, I don't make them sell on eBay. It'll be funny when they get stuck with 6 CPUs or so each when they become wildly available. This entire release has been a money grab. Motherboard pricing is absolutely disgusting and out of control. I saw folks on forums saying that $500 to $700 is a fair price for high-end motherboards. Again, if they couldn't charge it to a credit card, I doubt they would think that. Just look at the X370 and X470 releases. The most expensive and high-end board that was actually worth paying for was the ASUS Crosshair VI Extreme at $359. And that thing had a 8+4 phase VRM which to this day is overkill. Oh, and the 10Gbit Aquantia 107 NICs? You can get one for $80 brand new, and it will still hook up to your southbridge and they are PCI-E 3.0. So yes, overpriced. However, as long as enthusiasts think that the prices are fair, it's all that matters for board manufacturers because they have no incentive to lower prices. I don't even want to think what next-gen HEDT boards will cost. It's almost like X570 motherboards that are between $200 to $300 are made in such a way to justify the $500 to $700 motherboards. I tell ya, it's ridiculous.

FYI, you can pick up a 3900X at your local Best Buy. IIRC Best Buy restricted the 3900X for local pickup only due to so many orders. Not bad.
 
Last edited:
I meant if AMD had set the MSRP at $629 it would be fair.. I think it's part of why the 3900X is selling so well at $500. The cost differential from midrange 3600 to 3900X is only $250 more, and from an Intel perspective, the cost from 8C/16t thread 9900K to the $3900K 12c/24t is $20. I think the cost offset is actually too cheap.. Not that I'm complaining by any means
 
nobody with a microcenter nearby complains about not being able to get a 3900x for very long.

best physical computer store still in existence. Even though i go there mostly for micro-controller odds and ends.
 
  • Like
Reactions: STEM
like this
AMD might as well offer a even bigger chip say a 3975x with s 20 cores and sell it for 200 bucks. What's to lose , they won't actually have to have any available for sale just have a sku for it and release it. How can a company talk and talk about a line of CPU's for over a year, then release it and have minimal product actually available for sale unless it's a marketing strategy or they actually have problems making them. I'll say it again AMD has been responsible for selling more Intel 9900K's in the last month than iNTEL has been. It's still faster and quite available

This is nonsense. These processors are hardly vaporware. First off, you need to understand that AMD isn't Intel. Even via third-parties, it doesn't have the capacity to build as many of these things as Intel does. You are also talking about needing two CCD's that meet specs for the 3900X compared to one in each of the other models. I've been at Microcenter a bunch lately, and every couple of days they have more 3900X's in stock. They are just hot sellers right now.

Do you have some sort of evidence that there has been a spike in Core i9 9900K sales? The price drop on 9900K's would be the only potential reason for an increase in 9900K sales. That and Ryzen 3000 is now a known quantity, and given that people are now aware that the 9900K is faster in games, anyone on the fence about what to buy has potentially made a decision to go Intel. Some people held out to see if Ryzen would ultimately end up being faster in gaming.

Uhmm, the 3800X is the direct competitor to the 9900K. And the 3800X is available anywhere for nearly $100 less and the $330 3700X is basically the same chip with a similar supply. There not selling more 9900K's because of 3900X not being available.. The 3900X specs very close to the 9920X, not the 9900k, should you want the Intel counterpart it's going to cost you $1,200

Simple fact is, the 3900X is basically 2 high binned 3600x dies, so they have to sort dies and test them to make their top of the line chip, there making a lot of 3600's and less 3900X's through binning. This combined with the 3900X being the fastest consumer chip that you can buy relatively speaking, demand exploded. Stock is moving in and out quickly because that many are selling

No, it isn't. What generally defines the comparison is usually price point, not specifications. In this context, we are looking at AMD's flagship processor (for now) in the mainstream market vs. Intel's. Both are around the $500 price point. Even AMD defined it this way in its slide deck.

upload_2019-8-15_10-22-14.png


Yes, I know AMD has the 3600X listed twice. Their mistake. Regardless, you can see that the price of each AMD processor is designed to coincide with one of Intel's price points. Intel has adjusted their prices somewhat in an effort to boost sales, but the comparisons are basically outlined in the slide. The Ryzen 9 3900X is to be compared against the Intel Core i9 9900K and so on. The fact is, AMD's pricing makes a lot of sense because it provides additional value that favors the Ryzen 3000 series. At nearly $500, AMD gives you 12c/24t compared to Intel's 8c/16t. The only reason we compare the 3800X and 3700X as alternatives to the 3900X is for gaming machines where the 3900X's extra cores and CCD aren't needed. They are almost detrimental to performance rather than beneficial. This allows someone to save a considerable amount of money in a pure gaming build, but gaming isn't the only metric by which these processors are priced and judged.

Every reviewer I know of compared the Ryzen 9 3900X to the Core i9 9900K and this is because we were looking at each companies flagship offering and their shared price point.

Lastly, let's talk about binning. The Ryzen 9 3900X is not a pair of highly binned 3600 or 3600X chiplets. At least, we have no evidence this is the case since most of the cores in a Ryzen 9 3900X are incapable of a 4.4GHz or greater clock speed. Many people report 4.4GHz all core overclocks on 3600 / 3600X's, which isn't what we can get out of a 3900X. At least, very few of them can achieve this.
 
Last edited:
AMD might as well offer a even bigger chip say a 3975x with s 20 cores and sell it for 200 bucks.

Aside from the (hopefully) facetiousness, it should be noted that AMD isn't putting more than eight cores on each CCD (chiplet), and they're not putting more than two CCDs / chiplets on each Ryzen package.

It's likely not possible for AMD to increase the cores per CCD or CCDs per package until their next die shrink, which should be TSMCs 5nm node.
 
I'm sure they'll stick to whatever number makes sense seeing as how at some division, they have to share memory bandwidth ...both from cache and from system ram and they need access to IO. Cramming too many into that shared pool eventually does more harm than good on many levels.

Before even thinking about 5nm though. A novel cooling solution needs to be made to move the wattage out from smaller surface areas effectively. operating voltages can only go so low. The wall of wattage will need to be addressed, otherwise what's the point of shrinking the die size down if it means never being able to hit max clock rates for any length of time before having to drop frequency due to overheating.

It'll be interesting to hear how it's eventually dealt with. My bet is on some kind of carbon nanotube fabric that can transport heat horizontally on the die so conventional heatsinks and such can continue to be used.
 
i dont see them doing this currently. What evidence do you have that they do?

A given wattage in a given area is harder to cool than that same wattage spread out over a larger area.

This is exactly what we see with die size reductions in cpu's and it is seen with the move to 7nm in these chips compared to similar wattage chips using larger die sizes.

There's nothing "ryzen" about it ..it's physics.

edit: the only thing being done so far is using solder to connect the ihs cover to the die's. Rather than old school tim's. Obviously that only gets you so far since it's easy to hit thermal throttling temps at relatively low wattages compared to larger (older) cpu's with the same cooling systems.

between 12nm and 7nm is a 43% reduction in die size on the Zen2 cpu's compared to zen+ ...So basically the same amount of power must be dissipated from half the surface area (since tdp's are basically the same). These are physics problems that will also happen on the intel side when they eventually catch up :)
 
Last edited:
i dont see them doing this currently. What evidence do you have that they do?

They spread out the heat generation. Evidence: Any screenshot of a Ryzen 3000 die (with Zen 2 cores).

The reason you don't see it is because AMD is terrible at efficiency in the first place.
 
if the heat was spread out across the entire IHC then we'd see no difference in temps for the same output wattage in a zen2 than we do with a zen+ since the IHC size is the same and we do. It's obvious the heat transport has not improved alongside the die shrinkage.

as far as amd being terrible at efficiency According to what ?

https://www.tomshardware.com/reviews/ryzen-9-3900x-7-3700x-review,6214-3.html

amd is using less wattage and putting out more performance across more cores than intel

terrible efficiency is no longer a valid argument.

I think chiplets help with this versus a larger monolithic die like Intel uses.

I think the intel dies end up being similar in area to the multi-chiplet design from amd just considering the cpu and not their integrated graphics (compared to the zen+ dies anyway). Still, it does supread the amd ones out a little bit. You still have up to 8 cores in half the surface area to deal with.
 
Last edited:
One of Germany's largest e-tailers, Mindfactory, has received a shipment that allowed them to fulfill 1/5 of their existing 3900X preorders:

 
terrible efficiency is no longer a valid argument.

Zen can't get faster without extremely excessive binning- and trying to push it faster, where possible, requires enormous cooling.

Drop the voltage on an Intel CPU to where they're most efficient and do the same for Ryzen, and watch Intel run circles around AMD with an older, larger process :D

Still, it does supread the amd ones out a little bit.

This is what I'm referring to. Spreading out the heat generation underneath the heatspreader makes getting the heat away from the CPU easier. But it still takes a lot to make any meaningful difference with Zen- the arch is already at its limit.
 
Zen can't get faster without extremely excessive binning- and trying to push it faster, where possible, requires enormous cooling.

Correct. AMD pushed it to the wall this time. There is nothing left on the table. Albeit, I do think the best bins were probably reserved for the 3950X, so it will be interesting to see how that one performs. Though it may be a situation similar to the 3900X where there is one golden chiplet, and one shitty chiplet (a shitlet? Lol.).

Drop the voltage on an Intel CPU to where they're most efficient and do the same for Ryzen, and watch Intel run circles around AMD with an older, larger process :D

On this one, The_Stilt has demonstrated that Zen 2 is more efficient than CFL/Sklyake variants even at lower clocks/voltages. In fact, the picture is remarkably similar if you lower clocks/voltages with both uarchs. Relative positions don't change much.


This is what I'm referring to. Spreading out the heat generation underneath the heatspreader makes getting the heat away from the CPU easier. But it still takes a lot to make any meaningful difference with Zen- the arch is already at its limit.

Heat is a serious limiter for Zen 2, even when at otherwise reasonable clocks and voltage. The chiplets are so damned small, it's difficult to exhaust that heat quickly enough. We've seen a lot of mixed messaging around boost clocks, where some people are hitting them, or pretty close to them, and others are way off. I suspect heat is a big part of that, since precision boost will boost until a stock limit is reached... and one such limit is heat.

But as you say, even if you eliminate the waste heat, the uarch is near its limits. Overclocking these is essentially pointless. Even PBO/Auto-OC gets what... a percent or two? Not really worth even bothering. Which is kind of sad to this old OCer, in a way.
 
Zen can't get faster without extremely excessive binning- and trying to push it faster, where possible, requires enormous cooling.

Drop the voltage on an Intel CPU to where they're most efficient and do the same for Ryzen, and watch Intel run circles around AMD with an older, larger process :D

If you think undervolting stock cpu's is going to let intel beat performance / power use numbers with zen2 then you're dreaming. Zen2 can undervolt and unless you can pull way more off the cpu voltage than .1 volts on intel, you're not going to drop power usage remotely enough to make a difference.

If you're arguing that you can undervolt to beat zen1+ then you're having a different discussion than what's going on.

This is what I'm referring to. Spreading out the heat generation underneath the heatspreader makes getting the heat away from the CPU easier. But it still takes a lot to make any meaningful difference with Zen- the arch is already at its limit.

spreading the heat out from tinier and tinier cores has absolutely nothing to do with zen or intel archs. It's a physical consequence of making tiny circuits. Nothing about intel or amd makes any difference with the problem that becomes ever more problematic as die sizes shrink. It's not like intel has a arch that magically works at half the voltage when they drop to 7nm that amd's chips run at. They're both going to end up putting out the same wattage for relatively the same performance but over a smaller surface area.

So far the tech used to move that heat out from where it's generated hasn't changed. Just throw some highly conductive metal on top ..interfaced with solder or TIM. Which has been adequate up until now but it seems unlikely we'll see 150watt + cpu's with these die sizes able to sustain that power use since 110 watts or so can't be moved away quickly enough with avg water cooling even without yielding 85C temps.

There are experimental ideas floated around to create little microscopic heat pipes out of carbon nano tubes and the like within cpu dies to transport that heat faster than a slab of copper on top does but I dont think that's going into anything yet. Other ideas also exist, none of which seem to be getting implemented yet. But it seems unlikely we'll be able to avoid something along those lines once we go to 5nm. Otherwise we'll be limited to something like 65watts max cpu's at 5nm ... as the cores would thermal throttle when they reach 90C since we're pushing 65watts into half the area we were with 7nm and not improving how fast copper can move heat
 
On this one, The_Stilt has demonstrated that Zen 2 is more efficient than CFL/Sklyake variants even at lower clocks/voltages. In fact, the picture is remarkably similar if you lower clocks/voltages with both uarchs. Relative positions don't change much.

Without context, I'd accept this as likely true- I'm speaking of running both at their 'most efficient' voltage and clockspeed ranges, and then testing performance, and then measuring task energy for comparison.
 
no matter what context you want. intel wont have power to efficiency crown again until they actually start selling a 7nm chip.

even then, we're not even talking about server processors...where power to efficiency isn't even close to favoring intel. It would be surprising is intel's next gen server cpu is even close to approaching epyc performance / watt for anything remotely close to their price. I'm sure they could drum up something that can do it the buyer has no problem spending twice the amount though. Intel has basically infinite money to produce a cpu that has a 0.005% wafer yield and not really blink.
 
spreading the heat out from tinier and tinier cores has absolutely nothing to do with zen or intel archs.

I didn't make this argument.

They're both going to end up putting out the same wattage for relatively the same performance but over a smaller surface area.

Not likely, given AMD has just more or less caught up with Intel's five year old arch.

If you think undervolting stock cpu's is going to let intel beat performance / power use numbers with zen2 then you're dreaming. Zen2 can undervolt and unless you can pull way more off the cpu voltage than .1 volts on intel, you're not going to drop power usage remotely enough to make a difference.

I know we're talking about desktop parts here, but when talking about efficiency, we can use mobile CPUs for comparison.

Put simply, AMD is so innefficient here that OEMs don't bother with their parts for thin and light.. AMD has a performance advantage, given that their CPUs are certainly fast enough for what mobile users need and their GPUs are as fast as you're going to get in a single socket (without going exotic), but battery life for like tasks is around half at best. And that's before you compare Intel at 10nm.
 
no matter what context you want. intel wont have power to efficiency crown again until they actually start selling a 7nm chip.

See Ice Lake. It's an imperfect comparison because it's going in ultrabooks and the like, however, it's also posting performance numbers that challenge Zen 2 core for core- while running in a 25w TDP envelope.

Intel 7nm?

Let's hope TSMC has 5nm+ ready soon.
 
Without context, I'd accept this as likely true- I'm speaking of running both at their 'most efficient' voltage and clockspeed ranges, and then testing performance, and then measuring task energy for comparison.

Yes, The Stilt did this.

https://www.overclock.net/forum/10-amd-cpus/1728758-strictly-technical-matisse-not-really.html

Relevant graph:

yjYDYVx.png


Note he explains the left side of the graph (where the 9900k appears to be getting more efficient, albeit it still less than the 3900X and 3700X) that he could not lower the voltage any further on the 3900X due to a motherboard limitation, so the efficiency gains flatline < 3Ghz. Also no results for Zen 2 or the 9920X at 4.5GHz, because neither could sustain an all-core OC at this level within any reasonable power envelope.
 
  • Like
Reactions: N4CR
like this
fanboy whichever company you want.

when intel drops it's die size, it's going to still have the same tdp envelopes that amd cpu's have more or less depending on how they decide to market their processors. producing 110 watts at half the size of their current chips will yield the same issue we're seeing with zen2 in keeping the cores at below thermal throttling temps. Unless they chunk their cores into even smaller groups and spread them around on a much larger area without performance losses due to distance apart. Which is an unlikely design choice.

Going to 5nm with 110watt cpu's doesn't appear to be possible without a novel approach to moving heat out of those dies. Obviously, this is assuming we'll want ever faster cpu's instead of just ever more efficient ones at current day performance levels.
 
Back
Top