Speculation: PC enthusiasts will migrate to HEDT

[Vattila]

With the latest rumour that EPYC 2 will use eight 8-core dies, it seems likely that the upcoming Ryzen 3000 series will be using one of those dies, and that we'll still have a maximum of 8 cores in the mainstream desktop segment next year.

Perhaps 8 cores on the mainstream platform will remain with us for a long time and — along with memory bandwidth and connectivity (PCI-lanes, etc) — define the divide between mainstream and HEDT.

Now consider:

1. Clock frequency is no longer increasing much.
   
2. Current frequency and core count in the mainstream desktop segment are more than enough for typical PC usage and will continue to be for the foreseeable future.
   
3. Manufacturing processes are getting more expensive and longer lasting, so Intel and AMD are not keen on selling big pieces of silicon at low margin where it is not needed.
4. Arguably, more than 8 cores need more than dual-channel memory bandwidth, and hence a more costly system architecture (motherboard).
5. The minority that really need more performance is well served by the HEDT (high-end desktop) segment, which provides the needed bandwidth, connectivity and advanced features.
6. With AMD's entry into the segment, HEDT pricing is rapidly coming down.

Are we looking at a future where the mainstream desktop platform will be increasingly optimised for typical PC usage, in particular, optimised for low power and cost, while all kinds of users needing more compute performance — PC enthusiasts, over-clockers, gamers, content creators, developers, etc. — will migrate to HEDT?

 

[R0H1T]

So perhaps 8 cores on the mainstream platform will remain with us for a long time and — along with memory bandwidth and connectivity (PCI-lanes, etc) — define the divide between mainstream and HEDT.

It depends on where or when the next HEDT platform(s) evolve, the current offering start from 6 to 8 cores IIRC ~ Intel & AMD respectively. If the next gen starts from 10 & 12 cores, or 10 & 16 cores, for Intel/AMD then we'll see a clear difference between MSDT & HEDT. Otherwise the kind of overlap we see today, will make it hard to "define" the divide.

 

[Vattila]

It depends on where or when

I should have been a little more specific about the time frame to make the poll more useful. The question is whether there is or will be a trend towards HEDT going into the future, but I guess many will vote in the poll according to the current time and preference.

An interesting question to consider: Will there come a time when gamers will have a better experience on an HEDT platform? If so, how long into the future is that?

the current offering start from 6 to 8 cores

Yeah. To be precise, 8-core is an upper limit in the mainstream, not a lower limit in HEDT. However, I doubt it makes sense to offer any less in the HEDT segment going forward.

 

[arandomguy]

I'll again say that these types of questions are too abstract and open ended. Since the criteria's are not strictly set all the following arguments end up just arguing semantics of what constitutes a "PC Enthusiast" and their usage.

Say we take gamers at least to my definition unless the HEDT actually offers more performance in games (not seeing it) then why would they move to HEDT? They'll just take the CPU/platform cost savings (as they have been doing) and put their money in the GPUs (as Nvidia is stating GPU percentage of system builds has been growing and they are taking full advantage).

 

[Vattila]

these types of questions are too abstract and open ended. Since the criteria's are not strictly set all the following arguments end up just arguing semantics of what constitutes a "PC Enthusiast" and their usage.

That is a danger, true. However, you can still have an interesting discussion, as long as you are clear about what you mean when you say "enthusiast", and you are open to the possibility that someone else may have another definition in mind.

Say we take gamers at least to my definition unless the HEDT actually offers more performance in games (not seeing it) then why would they move to HEDT?

That's a good point. The mainstream desktop platform already covers a wide range of performance, well beyond what is needed for typical PC usage. And the mainstream desktop platform is currently the target for PC game developers.

However, the HEDT platform offers more compute performance, with higher memory bandwidth and core count. With gaming and HEDT being the (only) growing segments of the desktop PC market, would it make sense for game developers to target HEDT in the future? After all, gaming is a high-performance compute application. The most impressive games and most impressive hardware have traditionally gone hand-in-hand.

 

[beginner99]

Who really needs more than 8 relatively fast cores? Most don't even not enthusiasts. Pcie lanes is somewhat of an issue and if we are lucky but I doubt ti very much zen 2 will come with pcie 4 which would solve the issue. RAM is also only a minor problem next 2-3 years till we get ddr5 in mainstream. Also games are more sensitive to latency than bandwidth so if you do need the BW, you probably also need the cores so HEDT.

Zen 2 consuemr being 8 core was IMHO to be expected. Surprising they gonna do a 8-die MCM with epyc. I have trouble believing that as 4-x mcm is trivial in comparison and the package size???hm...

 

[arandomguy]

That's a good point. The mainstream desktop platform already covers a wide range of performance, well beyond what is needed for typical PC usage. And the mainstream desktop platform is currently the target for PC game developers.

However, the HEDT platform offers more compute performance, with higher memory bandwidth and core count. With gaming and HEDT being the (only) growing segments of the desktop PC market, would it make sense for game developers to target HEDT in the future? After all, gaming is a high-performance compute application. The most impressive games and most impressive hardware have traditionally gone hand-in-hand.

How fast is HEDT sales even increasing? I thought for example the Mindfactory sales data (that one German source) was showing that both AMD and Intel HEDT sales were just 2% of each companies respective sales. That is far away from becoming mainstream at least in terms of market share.

As for developer targeting it will be a tricky issue because features that target the CPU tend to either not scale very well. So this is much more problematic than developing advanced graphics as those you can scale more so while leaving the game also playable for those with the graphics power.

Then if we circle back to that the developers are also further more likely to target the GPU in terms of compute power, as the raw compute power is higher and more available. Something interestingly related to this is the now popular post process anti-aliasing techniques were first initially pushed by Intel (with MLAA) as a way to do anti-aliasing on their CPUs and increasing the marketing benefit of CPUs for gaming.

 

[french toast]

With the latest rumour that EPYC 2 will use eight 8-core dies, it seems likely that the upcoming Ryzen 3000 series will be using one of those dies, and that we'll still have a maximum of 8 cores in the mainstream desktop segment next year.

Perhaps 8 cores on the mainstream platform will remain with us for a long time and — along with memory bandwidth and connectivity (PCI-lanes, etc) — define the divide between mainstream and HEDT.

Now consider:

1. Clock frequency is no longer increasing much.
   
2. Current frequency and core count in the mainstream desktop segment are more than enough for typical PC usage and will continue to be for the foreseeable future.
   
3. Manufacturing processes are getting more expensive and longer lasting, so Intel and AMD are not keen on selling big pieces of silicon at low margin where it is not needed.
4. Arguably, more than 8 cores need more than dual-channel memory bandwidth, and hence a more costly system architecture (motherboard).
5. The minority that really need more performance is well served by the HEDT (high-end desktop) segment, which provides the needed bandwidth, connectivity and advanced features.
6. With AMD's entry into the segment, HEDT pricing is rapidly coming down.

Are we looking at a future where the mainstream desktop platform will be increasingly optimised for typical PC usage, in particular, optimised for low power and cost, while all kinds of users needing more compute performance — PC enthusiasts, over-clockers, gamers, content creators, developers, etc. — will migrate to HEDT?

This is what I have been saying (cores not HEDT)..not enough bandwidth to service 12+ cores with dual channel memory, for desktop, customer's would rather have 8 super fast cores that can effectively be served by its bandwidth with affordable ram.
If zen2 goes wider and faster, then 3200mhz ram is the minimum ram you would want with 8 cores, perhaps faster.
12-16 cores?..16 cores is a definite no, especially if AMD also increase the simd engines, 12 cores is an outside bet imo and something I bet on a while ago, I am not sure now, alot of the potential performance will be left on the table or be inaccessible without super expensive 4000mhz ddr4...that if AMD will support those speeds in the first place.

8 faster wider cores with large cache, improved infinity fabric and perhaps topology.
Leave the many core to enthusiast's, on die GPU to laptops and budget segment.
That way Ryzen 3000 can keep die size to sub 150mm2...desirable considering expensive 7nm wafers, needed capacity.

On topic, I think more and more enthusiast's will be using HEDT platforms yes, 8 fast cores is sweet spot for mainstream and for available bandwidth and affordable ram to feed it.
HEDT has been made more affordable and accessible to more enthusiast's, so yes is my answer.

 

[NTMBK]

Developers will continue to focus on the platform where 99% of their customers are, i.e. not HEDT. We'll start seeing better usage of 8C16T with the recent core increases from AMD and Intel, but don't expect games to start utilizing 64 cores spread over 4 NUMA nodes efficiently.

 

[dnavas]

Developers will continue to focus on the platform where 99% of their customers are, i.e. not HEDT.

Yep. And I expect hardware manufacturing to do the same. AMD needs more inroads into the laptop arena. Idle and low-core power use isn't helping them here, but it isn't clear to me if that's the only problem, or if there are more business-related matters involved. Laptops aren't going HEDT, but then that wasn't the question :>

Gamer enthusiasts will want higher single-thread frequencies -- the higher, the better. They certainly don't need the igpu (which is a holdover from the laptop market), so if you deliver a cheaper, higher frequency HEDT ... then they might. I don't expect 7nm to be cheap in the first round though, and Intel has never shown any interest in competing between their teams (too bad).

 

[whm1974]

Yep. And I expect hardware manufacturing to do the same. AMD needs more inroads into the laptop arena. Idle and low-core power use isn't helping them here, but it isn't clear to me if that's the only problem, or if there are more business-related matters involved. Laptops aren't going HEDT, but then that wasn't the question :>

Gamer enthusiasts will want higher single-thread frequencies -- the higher, the better. They certainly don't need the igpu (which is a holdover from the laptop market), so if you deliver a cheaper, higher frequency HEDT ... then they might. I don't expect 7nm to be cheap in the first round though, and Intel has never shown any interest in competing between their teams (too bad).

Gamer enthusiasts come in all stripes. I'm one and I live on a fixed income and so I have to settle for I can afford to spend on.

 

[IntelUser2000]

was showing that both AMD and Intel HEDT sales were just 2% of each companies respective sales.

It's way lower than 2%. 2% suggests 5 million out of 250 million CPUs. They are 1-2 million at most.

Problem with multi-die cores is higher latency. That goes against majority of client including gaming.

 

[Avalon]

I think unless the price of HEDT comes down, or mainstream ceases to handle top end games, then probably not. As you said, manufacturing processes are getting more expensive, so AMD/Intel probably do not want to lower the barrier to entry that is price any further. Performance wise...given that the GPU is the greatest bottleneck to gaming performance, it's going to be a long time before CPUs become such a bottleneck that enthusiasts no longer accept their performance level. While the rate of mainstream CPU performance advancement generation over generation is relatively slow, so far it seems like it's been enough to keep up.

 

[DrMrLordX]

The only way most PC builders will move to HEDT is if the vast majority of them give up on midrange/high-end desktop and stop buying hardware. There's still a possibility of that happening, though for right now (with chips like the upcoming 9900k and Ryzen 3 series) it doesn't seem to be an issue.

Sales for midrange hardware would have to dry up to next-to-nothing. It would get to the point that AMD, Intel, etc. would have to ask themselves, "why bother?" while migrating to extremely high-margin products for their limited remaining userbase. There would have to be a massive downturn in the AAA PC gaming market as well. It would mean the death of the PC as we know it as people completed the move to phones/tablets/Chromebooks. x86 would have to go back to being what it used to be: for business first, pleasure second. And price-of-entry would (by necessity) go wayyyyy up.

My parents paid over $3000 for my first PC: a 386-dx40 with 4 MB of RAM. That was in . . . 1993? I think? So yeah if we could somehow take the PC market back to 1993, I could see it happening.

 

[PeterScott]

If they start going beyond 8 cores on the the mainstream sockets, they might convince more HEDT buyers to stick with mainstream sockets. Even faster 8 cores might do that.

 

[whm1974]

If they start going beyond 8 cores on the the mainstream sockets, they might convince more HEDT buyers to stick with mainstream sockets. Even faster 8 cores might do that.

Most of us would benefit from speedier 6 and 8 cores then more cores beyond that anyway. And most use case don't need lots of cores/threads or NUMA.

 

[Insomniator]

All comes down to what you use it for. Can be a complete non-enthusiast/not know anything about PC's and still make great use of a 20 core HEDT platform. Can also be an total enthusiast/nerd that runs their desktop as a Linux VM installed on a server core installation and have no use for anything over 4 cores.

How many "enthusiasts" on this board have 16 thread chips that sit idle 90% of the time while playing rocket league and arguing about benchmarks showing 1% difference between different DDR4 timings?

 

[Vattila]

Thanks for the replies and votes so far!

Looking at the poll results and comments, I think it is clear that HEDT is currently not living up to its name — it is not providing the high-end ultimate performance for all enthusiasts; gamers in particular. Most PC enthusiasts are gamers, and I suspect that a majority of those that are voting "no" do so because they do not see the benefit in migrating to HEDT for gaming.

It is clear what AMD and Intel need to do. They need to make sure the HEDT platform is superior even for gamers. Of course, for some gamers, those that do parallel streaming or other multitasking, the HEDT platform is already attractive. But HEDT needs to become persuasive on gaming benchmarks as well.

It is not hard to see how this can happen. Future processors and games may make even better use of the superior memory bandwidth that HEDT offers. If that happens, and HEDT scores better on gaming benchmarks, a HEDT migration will make more sense. The most expensive gaming processors already achieves the majority of sales, so cost is not an issue.

If this happens, and HEDT becomes an undisputed high-performance platform for all, then I think we will see motherboard makers shift their resources to the HEDT space for their premium and gaming-focussed products. That makes more sense than spending that effort on a platform made for the mainstream.

 

[whm1974]

Thanks for the replies and votes so far!

Looking at the poll results and comments, I think it is clear that HEDT is currently not living up to its name — it is not providing the high-end ultimate performance for all enthusiasts; gamers in particular. Most PC enthusiasts are gamers, and I suspect that a majority of those that are voting "no" do so because they do not see the benefit in migrating to HEDT for gaming.

It is clear what AMD and Intel need to do. They need to make sure the HEDT platform is superior even for gamers. Of course, for some gamers, those that do parallel streaming or other multitasking, the HEDT platform is already attractive. But HEDT needs to become persuasive on gaming benchmarks as well.

It is not hard to see how this can happen. Future processors and games may make even better use of the superior memory bandwidth that HEDT offers. If that happens, and HEDT scores better on gaming benchmarks, a HEDT migration will make more sense. The most expensive gaming processors already achieves the majority of sales, so cost is not an issue.

If this happens, and HEDT becomes an undisputed high-performance platform for all, then I think we will see motherboard makers shift their resources to the HEDT space for their premium and gaming-focussed products. That makes more sense than spending that effort on a platform made for the mainstream.

There is the issue of cost to deal with here as HEDT platforms are not cheap and both the CPUs and motherboards are are costly. Quite a few gamers are gaming on budget boxes because they can't afford anything better.

 

[Vattila]

There is the issue of cost to deal with here as HEDT platforms are not cheap and both the CPUs and motherboards are are costly. Quite a few gamers are gaming on budget boxes because they can't afford anything better.

Yeah. There will be gamers that cannot afford HEDT, of course. And adoption all depends on how much the cost of entry to the HEDT platform comes down.

But it seems there is much money willing to be spent in the gaming market. I am astonished how well the top gaming processors have sold in the DIY channel. They sell vastly better than any lower-end processors according to the Mindfactory data:

https://imgur.com/a/35ItoEz

 

[cellarnoise]

For those that can afford it, and have even a remote amount of many CPU core needs from time to time the high end will be attractive.

I built a ryzen 1700x system a year ago (got a great deal on the 1700x re-box, about what they are going for over a year later -cause I want power and am cheap...). Love it so much. Can run audio and video software without an issue and then can let it produce a final product over night (when I'm away). It's ability to give a fairly accurate preview and then a good final product is fantastic. But it is my hobby machine, and I do value my "hobby time", though I do earn some coin every now and then from this box.

I still use my 4700mq laptop that my ryzen build replaced for high cpu use, as it only has a hard limit of 47W tdp. With avx software such as video conversion / rendering it throttles over 2.7 gz on its 4 cores. Not good enough!

Even with the 4700mq sometimes I would get software hangs with a fast ssd waiting for heavy cpu threads to run. I don't like feeling like a crash is about to happen and it really does not feel the same with twice the cores available on the 1700x.

If non-gaming performance keeps up like it has, I will jump to a 16 or greater core box when zen 3 comes out!

 

[PeterScott]

For those that can afford it, and have even a remote amount of many CPU core needs from time to time the high end will be attractive.

For many it's not a case of afford, but a distaste for wasting money.

If your need for high counts is only occasional, it really isn't much of a need, since uses for high core counts beyond 6-8 mainstream socket, tend to be non real-time encoding/rendering.

Say you do 4 hours of video encoding/week. Does it makes sense to spend a lot of extra money so it can happen in 2 hours, when it's isn't real time activity and can take place in the background, or even when you are away from your computer?

As mainstream socket core counts increase, HEDT will get less attractive to those without serious rendering/encoding needs. While many users can convince themselves (incorrectly) that they could benefit from more than 4 cores for doing trivial loads like watching video while web browsing (but LOTS of tabs...), as mainstream socket core counts rise, they will have a much harder time convincing themselves they need more cores.

HEDT will push further into the realm of hard core encoders/renderers.

 

[Vattila]

HEDT will push further into the realm of hard core encoders/renderers.

Yes, if HEDT cannot provide superior gaming performance, that is probably its destiny, and high-end gaming will be stuck on the mainstream platform. But, as Mindfactory data shows, gamers in great numbers are willing to spend what it takes to have the very best gaming processor. If that processor was on the HEDT platform, I think a good number of gamers would follow.

 

[darkswordsman17]

If HEDT is going to be a gaming platform, then its going to be something like a home server, with mixed use and/or multiple users. Which that would actually be a great fit for arcades, gaming cafes, expos, and other events (like competitive gaming), where you'd have multiple players (but wouldn't want to spend quite as much for the full on servers). Of course arcades and internet/gaming cafes aren't exactly doing well (but maybe this could help them carve out some new niche, especially the VR ones).

But it'll take a company willing to make the case that you put a bunch down, but then can save on the clients. And they'd have to develop some attractive thin client setup that gets people to actually want it. I think that'd take something like AR headsets or something. I think the only company that could realistically pull that off is Apple. Microsoft maybe (especially with Xbox and Mixed Reality/HoloLens). Where the devices/clients become very I/O focused, and you put more towards that and battery life and general usability with no longer having to be as concerned with processing performance of the device itself.

Considering the price of the Surface Studio, it'd probably be a $10,000+ thing and that's gonna be a very tough sell for even families. And Microsoft would probably be better of selling subscriptions to their cloud services, but I don't know maybe they could figure out some other aspect (like they subsidize the cost some in exchange for it operating as a semi-public server, and they setup a huge Microsoft network, which could benefit their cloud services.

I think Apple is the only brand that could legitimately get people to consider it, but even they would need something more compelling than what they have now. But for creative pros, a Mac Pro that doubles as home server could be potentially interesting.

 

[moinmoin]

Personally think it's purely on the definition of who are the enthusiasts. We like to generalize that group as a homogenous bleeding edge one, but in the case of PCs a heavy focus on gaming is nearly always implied. HEDT always had been more bleeding edge than mainstream platforms and may intuitively be a good fit, but both Threadripper and Skylake X showed that they are excellent options only for specific cases of which gaming as of now clearly is not one. So enthusiast gamers that would readily jump on to GTX 1080 ti, RTX 2080 and the likes are more likely going for 8086K, 9900K etc. for CPUs instead, even preferring their exclusivity through high prices over the more mainstream pricing of e.g. 2700X.