Speculation: Will intel be able to challenge AMD in the HEDT market?

piesquared

Golden Member
Oct 16, 2006
1,598
17
136
#1
With the release of Threadripper 2, AMD will unequivocally claim the HEDT crown. intel's huge die 28 core chip is unlikely to challenge the 32/64t 2990WX in either performance or price. That being the case intel may lose the HEDT crown not just this year but for multiple years with Zen2 Castle Peak due out next year.
 

NTMBK

Diamond Member
Nov 14, 2011
8,213
171
126
#2
Intel will still have a lead in a few benchmarks with that 28 core part- anything that heavily uses AVX-512, and some memory bandwidth heavy workloads (6 channel Vs 4 channel). Plus TR2 is extremely NUMA, with 8 NUMA nodes in one socket (half of which have no direct access to a memory controller), which will mess up scaling in some workloads. It will vary wildly from benchmark to benchmark which part wins, I suspect.
 

maddie

Platinum Member
Jul 18, 2010
2,373
319
136
#3
With the release of Threadripper 2, AMD will unequivocally claim the HEDT crown. intel's huge die 28 core chip is unlikely to challenge the 32/64t 2990WX in either performance or price. That being the case intel may lose the HEDT crown not just this year but for multiple years with Zen2 Castle Peak due out next year.
Difficult, at least from a core count perspective, until Intel goes chiplet. We can safely assume that TR can equal whatever the contemporary EPYC CPU has available. 32,48,64.
 

csbin

Senior member
Feb 4, 2013
816
8
136
#4
Intel will still have a lead in a few benchmarks with that 28 core part- anything that heavily uses AVX-512, and some memory bandwidth heavy workloads (6 channel Vs 4 channel). Plus TR2 is extremely NUMA, with 8 NUMA nodes in one socket (half of which have no direct access to a memory controller), which will mess up scaling in some workloads. It will vary wildly from benchmark to benchmark which part wins, I suspect.

32 core TR2 is 4 NUMA

 

csbin

Senior member
Feb 4, 2013
816
8
136
#5
32 core EPYC

 
Dec 28, 2013
53
0
66
#6
Intel will still have a lead in a few benchmarks with that 28 core part- anything that heavily uses AVX-512, and some memory bandwidth heavy workloads
On the other hand the server version of that part currently sells for $9,000+ vs the 2990X for $1800. Coming on a brand new socket platform Intel has to match cost/performance to get any traction. Going to be a helluva lot of very upset Xeon Platinum 8176 buyers out there.
 

NTMBK

Diamond Member
Nov 14, 2011
8,213
171
126
#7
On the other hand the server version of that part currently sells for $9,000+ vs the 2990X for $1800. Coming on a brand new socket platform Intel has to match cost/performance to get any traction. Going to be a helluva lot of very upset Xeon Platinum 8176 buyers out there.
Price wise AMD are going to win, hands down! It's great value.
 

SPBHM

Diamond Member
Sep 12, 2012
4,763
27
106
#8
I'm still very curious to see how the memory configuration will affect performance on more memory intensive tasks (not Cinebench or synthetic tests), but for now things are looking good for AMD.
 

Vattila

Senior member
Oct 22, 2004
363
59
136
#9
Thanks for this speculation thread. It made me ask "what is the current situation" and go back to Ingebor's data based on MindFactory sales for July:

Model Qty
TR 1900X (8C16T) 30
TR 1920X (12C24T) 10
TR 1950X (16C32T) 100
Total 140
---
i7-7800X (6C12T) 10
i7-7820X (8C16T) 90
i9-7900X (10C20T) 20
i9-7920X (12C24T) 10

i9-7940X (14C28T) 20
i9-7960X (16C32T) 0
i9-7980XE (18C36T) 0
Total 150


Over the course of the last 12 months at MindFactory, TR 1950X has sold 1130 units alone, while i9-7960X and i9-7980XE combined have sold 90 units. That's a relative score to AMD of 13x the unit sales at and above 16C32T.

If MindFactory sales are typical, it looks like Threadripper already dominates at the top of the HEDT segment. With the Threadripper 2000-series now launching, and with AMD having a strong 7nm roadmap, Intel really has a challenge on their hands. They will need to move to MCM solutions, and likely need a better process than 14nm, to compete in this segment.
 
Last edited:
May 11, 2008
18,245
25
126
#10
While perhaps a lot of people think chiplets are an AMD idea, the idea has been around for some time.
IBM uses chiplets as well for their power cpu's.
AMD has been smart, because they started using chiplets at 14nm already to learn from the experience when going down to smaller nodes.
AMD does not have as much production facilities and thus must harvest as much as possible. This will be an increasing problem when going towards smaller nodes.

Intel has a crazy number of sku's. And that is not to confuse customers, it is to harvest as much usable dies as possible. Why so much dies, well because they produce so much dies.
But that does notwork for large dies at small nodes like 10nm.

Intel is going the chiplet road as well , that is why they developed EMIB.
And to let all these little chiplets communicate properly , a sound communication network is needed.
Intel has a lot of experience, but not enough. And that is where i think Jim Keller comes in.


This is an interesting article, i still have to read it myself.
https://semiengineering.com/chiplet-model-gaining-steam/

Discussion about chiplets is growing as the cost of developing chips at 10/7nm and beyond passes well beyond the capabilities of many chipmakers.

Estimates for developing 5nm chips (the equivalent 3nm for TSMC and Samsung) are well into the hundreds of millions of dollars just for the NRE costs alone. Masks costs will be in the double-digit millions of dollars even with EUV. And that’s assuming that IP will be available for the most advanced nodes, which isn’t a sure thing because a number of IP vendors are debating which nodes to support and from which foundries.

This doesn’t mean that billions of 5/3nm chips won’t be produced and sold. Digital logic still benefits from scaling, and density increases generally mean faster processing, even if that density comes from more than one processing element. But analog circuitry does not benefit from scaling, and a planar SoC is becoming too expensive to continue scaling. Its successor is likely to be some sort of advanced package, with a mix of components developed at different process nodes and connected by some high-speed interconnect, whether that is a bridge, an interposer, TSVs, or even wirebonds. So rather than a 3nm SoC, it’s highly likely that a 3nm processor platform will be included inside a package.

This is where chiplets begin to look much more interesting. If IP can be hardened at the node that makes the most sense—which in the case of analog may be 250nm or 40nm, depending upon the IP—then IP blocks or subsystems can be fully characterized to the point where they can be used in multiple designs, almost like off-the-shelf components. That’s a much more attractive business model for IP vendors, because they can develop IP once and sell it for years to come to more customers in more market segments, and it levels the playing field for chipmakers around the globe.

Marvell and Kandou Bus were the first to jump on this shift. They announced a deal in 2016 under which Marvell would use Kandou’s chip-to-chip interconnect technology to tie multiple chips together. Since then, DARPA has established a chiplet program, and other companies say privately they are working on similar approaches.

The chiplet approach always has made sense from a conceptual level. Building chips from components such as standard memories, SerDes blocks, I/O and even processor logic should be relatively quick and much less expensive. That isn’t true today, mainly because there is no standardized way to make this all work. But the shift to advanced packaging is already in full swing, and that will only continue to gain steam at each new node. It takes too much power and time to drive signals across a large die, in which wires are scaled to the latest node and where contention for resources creates floorplanning and timing nightmares.

While the biggest chipmakers will continue to leverage density improvements at new nodes, they also are being forced to change direction in order to improve performance and lower power consumption. Intel’s EMIB and Samsung’s RDL bridge are indications of these shifts. So are the growth of fan-outs, systems in package, 2.5D, and a number of other packaging variants. There are more packaging options available today than ever before. Not all of them will survive, but the direction is at least clear. Advanced packaging is here to stay.

The next step is to reduce the cost and development time for advanced packaging, while also improving the predictability of these different packaging options. This is where the chiplet approach begins to look particularly interesting. How the marketplace for hardened IP will evolve, and who will run it, are unknown at this point. But the fact that more companies are discussing this option, and looking at how to harden IP into modules or chips, is the latest indication that change is on the way.

The chiplet approach is one of a number of options on the table, but it’s certainly an interesting one. And if it lives up to its promise, it could have significant implications for the entire chip industry.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
72
106
#11
It pretty much depends how does one define HEDT.

If the definition of HEDT involves workloads which the end-users typically run on their system, then the question should be the other way around and the answer would be blatant no.

Threadripper only shines in scenarios where the legacy workload can be executed in extremely parallel manner.
Any workload, especially the modern ones which cannot fully utilize all of the cores in TR will perform better on Intel HEDT CPUs.

Until AMD is able to provide a monolithic die with sixteen or more cores, and larger execution resources (for faster 256-bit code execution) I personally considered the HEDT segment to belong to Intel.
Until Intel gets their newer manufacturing nodes in order AMD has the chance to actually take the HEDT crown. But ultimately that is down to the structure and the design of Zen 2.

That's my personal opinion as the owner of both 7960X and 1950X CPUs.
 

JoeRambo

Senior member
Jun 13, 2013
640
42
136
#12
32 core TR2 is 4 NUMA
Gonna be interesting to see how AMD configures NUMA, how will the distance matrix for nodes look like and how will memory controller less nodes will be represented.
 

french toast

Senior member
Feb 22, 2017
918
2
106
#13
Well, if those sales numbers are representative of the wider market, then it looks like the HEDT segment already belongs to AMD, this will only exacerbate with the new 2990wx and family.

If I had a choice I personally would go with the 28core intel product, the faster single core and better overclocking capabilities would suit my needs better, but it will likely cost thousands more.

That means the 28 core intel is not directly comparable to the threadripper 2...even with a new socket the price difference is going to be enormous, for many HEDT type scenarios the TR2 is going to offer comparable performance, for less than half the price...no contest... It's more like comparing a Xeon Vs AMD HEDT ...different categories.
 

Vattila

Senior member
Oct 22, 2004
363
59
136
#14
the question should be ["can AMD challenge Intel in HEDT"] and the answer would be blatant no.
The only thing "blatant" is your bias. I see that you are one of the most popular posters here with the, by far, highest ratio of likes-to-posts that I have observed, and I guess it is fair to say that you are respected for your technical knowledge. I think it is highly dangerous to factfullness for authority figures to have hidden biases. Hence, I appreciate that you are so open and frank about it.

That said, the definition of HEDT is not about your preferred use-case. It is defined by the products that Intel and AMD offer in this category, and the success of these products — challengers, if you will — is simply determined by their sales. Over the last three months, MindFactory hasn't sold a single i9-7960X or i9-7980XE, while they have sold 210 1950Xs alone. If you still cannot accept AMD as a challenger in the HEDT space, then you should refrain from taking part in the discussion.

Sorry for arguing and not speculating.
 
Last edited:

Abwx

Diamond Member
Apr 2, 2011
8,792
122
126
#15
Threadripper only shines in scenarios where the legacy workload can be executed in extremely parallel manner.
Any workload, especially the modern ones which cannot fully utilize all of the cores in TR will perform better on Intel HEDT CPUs.

.
Yes, we all know that ancient workloads were highly parralelized but thanks to the progress modern ones are most often lowly if not single threaded, hence the need fo more and more cores in HEDT...
 

aigomorla

Cases and Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
17,170
50
126
#16
Thanks for this speculation thread. It made me ask "what is the current situation" and go back to Ingebor's data based on MindFactory sales for July:
how is this even relevant?

I can show you sales graph for Intel from Dell and Supermicro which slaughter those mindfactory graphs in every catigory by a factor of at least 10x.

You guys are completely not accounting for Xeon's... and doing a straight i9 vs TR2.

Everyone who knows anything is aware for every i9 sold, at least if not 20-100 Xeon's were sold to the enterprise sector...

Dont get me wrong, i am glad for the competition from AMD. We honestly need it.. i also hope to see some for Nvidia as well from AMD.

But these speculation threads where its foreshadowing post apocalypse for intel are starting to get hilarious and redundant.
 
Last edited:

tamz_msc

Platinum Member
Jan 5, 2017
2,197
137
106
#18
I don't see how The Stilt is wrong in his assessment of HEDT capabilities of Intel's Skylake-X offerings when he talks about the actual workloads a typical HEDT user might run on his system.

Let's take two threads in this forum and analyze how Intel fares in terms of performance in the HEDT space. In the thread that polls what type of workloads people need more cores for, 3D-rendering is one of the least popular use cases. Incidentally Threadripper does very well in rendering as long as it is not AVX2 heavy, primarily due to the fact that AMD provides more cores for a given price point. In the industry, modern renderers make heavy use of AVX2 and the gap is much closer than what something like Cinebench would suggest.

If we look at the other end, then one of the most popular use-case scenarios is no doubt video encoding. Now looking at the data from the 4K Handbrake HEVC benchmarking thread, it is clear that x265 favors wider faster cores above all else, and Intel enjoys an advantage here since they can execute 256b ADD and MUL operations simultaneously, which AMD cannot. That's why a quad core 4970K isn't far behind a Ryzen 5 1600, or why an 8700K is comfortably ahead of the 2700X. With AVX512 support being slowly incorporated into x265, Skylake-X enjoys a strong lead in terms of performance over Threadripper; the only reason why the latter is more viable at this moment is because of Intel's retarded pricing that allows AMD to offer more cores for less $$$.

The other two popular use case scenarios is virtualization and scientific computing. Now, owing to the relative immaturity of the AMD platform, there is no doubt that many people would be inclined to choose Intel because of the software situation on AMD, but this is changing slowly, and both AMD and Intel will hopefully be equally viable for virtualization in the future.

As for scientific computing, many modern scientific code can take advantage of AVX2 and AVX512, and sometimes the scaling is almost linear. It is only in certain CFD/CFD-like workloads which are completely memory-bound that this advantage evaporates for Intel. For those situations, EPYC is more viable due to the eight memory channels, but you won't have that on HEDT. Having said that, the general consensus is that the strong point of Zen is in integer workloads, while Intel is much stronger in floating-point workloads(due to AVX2/512). Since most scientific workloads are of the latter type, Intel still has the performance edge here.

So like The Stilt says, AMD's HEDT challenge hinges on how they address the problem of weaker vector-SIMD performance with Zen 2. According to the rumor mills, the future looks promising.
 

french toast

Senior member
Feb 22, 2017
918
2
106
#19
how is this even relevant?

I can show you sales graph for Intel from Dell and Supermicro which slaughter those mindfactory graphs in every catigory by a factor of at least 10x.

You guys are completely not accounting for Xeon's... and doing a straight i9 vs TR2.

Everyone who knows anything is aware for every i9 sold, at least if not 20-100 Xeon's were sold to the enterprise sector...

Dont get me wrong, i am glad for the competition from AMD. We honestly need it.. i also hope to see some for Nvidia as well from AMD.

But these speculation threads where its foreshadowing post apocalypse for intel are starting to get hilarious and redundant.
Yes, but xeons and epyc are not HEDT...which is the thread topic.
That is a separate subject altogether.
This is a valid thread discussing HEDT sales and upcoming technology that may affect it, one way or another.
 

Vattila

Senior member
Oct 22, 2004
363
59
136
#20
I don't see how The Stilt is wrong in his assessment of HEDT capabilities
No one said that Stilt was wrong in his assessment of his preferred HEDT use-case. He is wrong to confuse personal preferences with the definition of HEDT and to hint that AMD is not yet a challenger in this space, when Threadripper reception, reviews and sales data indicate otherwise.
 

french toast

Senior member
Feb 22, 2017
918
2
106
#21
I don't see how The Stilt is wrong in his assessment of HEDT capabilities of Intel's Skylake-X offerings when he talks about the actual workloads a typical HEDT user might run on his system.

Let's take two threads in this forum and analyze how Intel fares in terms of performance in the HEDT space. In the thread that polls what type of workloads people need more cores for, 3D-rendering is one of the least popular use cases. Incidentally Threadripper does very well in rendering as long as it is not AVX2 heavy, primarily due to the fact that AMD provides more cores for a given price point. In the industry, modern renderers make heavy use of AVX2 and the gap is much closer than what something like Cinebench would suggest.

If we look at the other end, then one of the most popular use-case scenarios is no doubt video encoding. Now looking at the data from the 4K Handbrake HEVC benchmarking thread, it is clear that x265 favors wider faster cores above all else, and Intel enjoys an advantage here since they can execute 256b ADD and MUL operations simultaneously, which AMD cannot. That's why a quad core 4970K isn't far behind a Ryzen 5 1600, or why an 8700K is comfortably ahead of the 2700X. With AVX512 support being slowly incorporated into x265, Skylake-X enjoys a strong lead in terms of performance over Threadripper; the only reason why the latter is more viable at this moment is because of Intel's retarded pricing that allows AMD to offer more cores for less $$$.

The other two popular use case scenarios is virtualization and scientific computing. Now, owing to the relative immaturity of the AMD platform, there is no doubt that many people would be inclined to choose Intel because of the software situation on AMD, but this is changing slowly, and both AMD and Intel will hopefully be equally viable for virtualization in the future.

As for scientific computing, many modern scientific code can take advantage of AVX2 and AVX512, and sometimes the scaling is almost linear. It is only in certain CFD/CFD-like workloads which are completely memory-bound that this advantage evaporates for Intel. For those situations, EPYC is more viable due to the eight memory channels, but you won't have that on HEDT. Having said that, the general consensus is that the strong point of Zen is in integer workloads, while Intel is much stronger in floating-point workloads(due to AVX2/512). Since most scientific workloads are of the latter type, Intel still has the performance edge here.

So like The Stilt says, AMD's HEDT challenge hinges on how they address the problem of weaker vector-SIMD performance with Zen 2. According to the rumor mills, the future looks promising.
That is a good technical breakdown of the advantages of both architecture, but in the end what matters is the sales...if we can gather more information the better.
Is AMD outselling intel in HEDT or not?
I would like to know.
Mainstream intel is killing it, server intel owns 95% of the market, but solid numbers of HEDT SKUs is something I have not seen.
 

dnavas

Senior member
Feb 25, 2017
234
12
71
#22
...So like The Stilt says, AMD's HEDT challenge hinges on how they address the problem of weaker vector-SIMD performance with Zen 2. According to the rumor mills, the future looks promising.
Hmm, have there been rumors of wider vector cores in Zen2?
It's definitely on my wish-list, but all I've heard thus far is core-count increases.
 

Vattila

Senior member
Oct 22, 2004
363
59
136
#23
rumors of wider vector cores in Zen2?
No leaks or rumours, as far as I know, but speculation, definitely. As part of widening the Zen 2 or 3 core (perhaps including SMT4, as rumoured, and more execution units), my guess is that AMD will widen AVX registers to 256 bits — and add support for 512-bit operations, split in two, like they now do for 256-bit operations, as I understand it.
 

dnavas

Senior member
Feb 25, 2017
234
12
71
#24
No leaks or rumours, as far as I know, but speculation, definitely
Yes, I'm suspicious of this speculation precisely because we've heard rumors of SMT4 (of all things) and not wider vectors. It makes sense given the widening data paths, it's just surprising that nothing concrete has leaked. Half-width support of AVX512 would probably be one of the first things *I* would want added to the HEDT part, but the last time I pointed this out, I got (justifiable) negative reaction as to its applicability in server / consumer settings (AKA: AMD's target markets). Our market is small, and the impetus to get this done may be lacking :|
 

Vattila

Senior member
Oct 22, 2004
363
59
136
#25
Our market [for AVX512] is small, and the impetus to get this done may be lacking
The impetus is full ISA support. From a marketing standpoint, AMD will not want to be seen to be lacking features compared to Intel. Also, they will want better throughput on AVX2 software, which is becoming more widespread.

So I guess they will support AVX512 by performing the operations in 256-bit registers and join the result (as with the 256-bit AVX2 support currently). But I doubt they will ever widen the registers to 512 bits. It seems un-zen. I think they will instead focus on doing wide vectors more efficiently on GPU units.
 
Last edited:

Similar threads



ASK THE COMMUNITY

TRENDING THREADS