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

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Shivansps

Platinum Member
Sep 11, 2013
2,342
132
126
#26
In price, AMD has a excellent deal, in performance im petty sure that a 28C Intel cpu with full AVX256 and AVX512 will have advantage over the 32C TR2 witch is divided in 4 NUMAS with half of them having no direct memory access or any other I/O, maybe even for non AVX loads as well.
 
May 11, 2008
18,156
10
126
#27
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 :|
I sure hope they are not going to go the SMT4 road for desktop processors. I rather see an extra execution unit specialized in (perhaps 2x256 as 512) AVX and SSE only instead of a combined fpu/simd unit as is now. If they are going to keep the 4 core CCX philosophy, it is better to widen the paths to a dedicated avx /sse simd unit to 256 bit. Maybe use even some of the experience from the gpu to use a special local qeue or cache and fifo/multiplexer of 512 bits wide and a dedicated 512 bits pipe for this unit.
 
May 11, 2008
18,156
10
126
#28
As long as HBM apu's and the software support for it are not existent and common, AMD will continue to have to support avx / avx512.
When finally HBM apu's are common and the software support is emerging to see the strength of such a device for algorithms that use simd operations that benefit from a embarrassingly parallel architecture more than from avx / avx512, avx and avx512 can slowly go the way of the original sse and 3dnow instructions.
 

StinkyPinky

Diamond Member
Jul 6, 2002
6,323
13
126
#29
In price, AMD has a excellent deal, in performance im petty sure that a 28C Intel cpu with full AVX256 and AVX512 will have advantage over the 32C TR2 witch is divided in 4 NUMAS with half of them having no direct memory access or any other I/O, maybe even for non AVX loads as well.
That is true, but the real concern for Intel is they just cannot get their 10nm out the door. AMD will be having 7nm HEDT cpu's next year and Intel will probably be trying to compete with a 14nm part
 

krumme

Diamond Member
Oct 9, 2009
5,643
40
136
#30
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.
What excactly is the software where a 7960x makes sense vs a 32c TR or a 6c kbl with higher freq?
In reality it for 90% resides to extreme wider fpu use and therefore the software where Intel HEDT portfolio makes sense is as slim as mindfactory numbers indicate.
Intel HEDT line is in no mans land.
And the reason is excactly the monolitic die and resulting cost. And no - 16c monolitic die and wide 512b fpu isnt suddenly the perfect sweetspot. Its probably exactly what is dragging Intel down in that market.
 

krumme

Diamond Member
Oct 9, 2009
5,643
40
136
#31
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.
Why should Intel even bother with this minor segment?
 

jpiniero

Diamond Member
Oct 1, 2010
5,929
90
126
#32
Why should Intel even bother with this minor segment?
They made a nice profit when they were selling smallish dies for a ton. The 28 core will do fine, but yeah the regular Skylake-X Refresh would have been cancelled if not for their desire to unload all the Skylake-SP dies.

I guess we will have to see if Tigerlake ends up being a scalable design. Probably won't see anything from Intel after Skylake-X refresh until then.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
64
106
#33
What excactly is the software where a 7960x makes sense vs a 32c TR or a 6c kbl with higher freq?
In reality it for 90% resides to extreme wider fpu use and therefore the software where Intel HEDT portfolio makes sense is as slim as mindfactory numbers indicate.
Intel HEDT line is in no mans land.
And the reason is excactly the monolitic die and resulting cost. And no - 16c monolitic die and wide 512b fpu isnt suddenly the perfect sweetspot. Its probably exactly what is dragging Intel down in that market.
I got the 7960X because I encode videos almost daily basis using HEVC.
The typical scenario is to encode 80-160k frames using constant rate control and aggressive motion estimation and rate distortion settings.
Most often the material still has =< 1080 vertical resolution, 4K is quite rare.

Due to the codec characteristics, you can efficiently utilize < 12 cores while encoding material with 1080 vertical resolution.
In theory you should be able to utilize ~ 17 cores at 1080 resolution (64 CTU), however in practice the scaling takes a nose dive beyond 10 cores.

Thanks to the AVX2 / AVX512 implementations in HEVC, Skylake-X has almost 50% higher IPC in it than Summit or Pinnacle Ridge.
That's a vast difference, even without mentioning the slightly higher clocks Skylake-X can simultaneously achieve.

The more modern video encoders (HEVC, VP9, etc) generally do not favor higher core count.
To maximise the efficiency the tile sizes have been increased, and that limits the number of cores you can utilize based on the vertical resolution of the video.
Obviously 4K and higher resolutions solve the issue, however such material is still extremely rare compared to FullHD or lower resolutions.

So it should be pretty obvious why I got myself a 7960X for which I paid 1700$, instead of using the 1950X which I got for free.
 

krumme

Diamond Member
Oct 9, 2009
5,643
40
136
#34
I got the 7960X because I encode videos almost daily basis using HEVC.
The typical scenario is to encode 80-160k frames using constant rate control and aggressive motion estimation and rate distortion settings.
Most often the material still has =< 1080 vertical resolution, 4K is quite rare.

Due to the codec characteristics, you can efficiently utilize < 12 cores while encoding material with 1080 vertical resolution.
In theory you should be able to utilize ~ 17 cores at 1080 resolution (64 CTU), however in practice the scaling takes a nose dive beyond 10 cores.

Thanks to the AVX2 / AVX512 implementations in HEVC, Skylake-X has almost 50% higher IPC in it than Summit or Pinnacle Ridge.
That's a vast difference, even without mentioning the slightly higher clocks Skylake-X can simultaneously achieve.

The more modern video encoders (HEVC, VP9, etc) generally do not favor higher core count.
To maximise the efficiency the tile sizes have been increased, and that limits the number of cores you can utilize based on the vertical resolution of the video.
Obviously 4K and higher resolutions solve the issue, however such material is still extremely rare compared to FullHD or lower resolutions.

So it should be pretty obvious why I got myself a 7960X for which I paid 1700$, instead of using the 1950X which I got for free.
Yes agree that is a usecase and we can find a few other similars. And HEVC 1080 encoding (and not x264 and 4k encoding) - where you cant wait for a 32c TR or 6C kbl - is excactly such a irrelevant miniscule segment and usecase it shows skl HEDT makes no sense today.
HEDT be it TR or SKL outside of IO makes only sense as rendering machines. Cheapish calculations machines for highly threadded workloads. Its just big calculators on sale. 32c TR is simply far ahead of the nearrest compettitor the 14c skl for 90% of the workloads here.
Not that it really matters outside of PR and brand value. Its also why we see that reuse of existing server solutions for both Intel and Amd.

Intel HEDT made sense when the nearrest compettitors was their own 4c core with limited pci and some dog slow 8c toaster. When the alternative is a 32c handicapped serversolution on crack its just dead in the water.
 
Last edited:
Mar 24, 2017
147
0
61
#35
I think people really need to understand that a lot of the benchmarks run by mainstream sites do not translate into reality with HEDT on Professional applications. We looked at TR for a bunch of workstations for Solid works, CAD and Photoshop and while TR performed well it lagged severely behind skylakeX. I believe TR and TR2 will continue to do very well with enthusiasts but, the majority of businesses running professional applications will still look to xeon\skylakex. AMD Really needs to get its 1-4 core clocks up as their performance in many cases is dismal. Engineers and Designers are not always running 12-16 cores full tilt. In many cases we see skylakex with 10-12 cores being faster than TR16. I would like to see Intel bring the prices of their 10-12 core parts down. The 7900x & 7920x are direct competitors to TR16 and they should be priced accordingly. I have no doubt the 28core part Intel is coming out with will be a beast and will most likely perform better than TR2.
 

moonbogg

Diamond Member
Jan 8, 2011
9,738
14
126
#36
Honestly it won't matter how well the Intel chip performs. Seriously, it won't matter. They will be priced so radically apart from each other that it will be meaningless to compare the two. AMD's chip will be attainable by people who want a TON of cores at around $1,700 with an already decently priced X399 board. Intel's solution won't be attainable by anyone, not to mention a 6 channel DDR4 motherboard. Anyone who can afford whatever Intel will be asking will already have a high core count Xeon system or something. It won't be priced at all like a high end consumer chip while AMD's chip will be priced like that. I expect Intel's chip to cost nearly as much as a small budget car (seriously).
That monolithic die strategy is obsolete as all hell right now. Its dead. It can't compete.
 

ub4ty

Senior member
Jun 21, 2017
749
307
96
#37
Simply put : No they won't. They eventually will once they go MCM, cut out the shenanigans and accept they can't charge a premium anymore. Seems that's some years out.


That monolithic die strategy is obsolete as all hell right now. Its dead. It can't compete.
Indeed. Whoever came up with that mesh architecture needs to be taken out back. It's like someone was hitting the crack pipe too hard before and after Xeon Phi added in with the business dev group influencing a costly meme architecture to justify keeping Intel's ludicrous pricing.

Rest in pieces Intel.
 

krumme

Diamond Member
Oct 9, 2009
5,643
40
136
#38
I think people really need to understand that a lot of the benchmarks run by mainstream sites do not translate into reality with HEDT on Professional applications. We looked at TR for a bunch of workstations for Solid works, CAD and Photoshop and while TR performed well it lagged severely behind skylakeX. I believe TR and TR2 will continue to do very well with enthusiasts but, the majority of businesses running professional applications will still look to xeon\skylakex. AMD Really needs to get its 1-4 core clocks up as their performance in many cases is dismal. Engineers and Designers are not always running 12-16 cores full tilt. In many cases we see skylakex with 10-12 cores being faster than TR16. I would like to see Intel bring the prices of their 10-12 core parts down. The 7900x & 7920x are direct competitors to TR16 and they should be priced accordingly. I have no doubt the 28core part Intel is coming out with will be a beast and will most likely perform better than TR2.
For normal usecases of the application you mentions a 8700k is faster anyway. Sure you can dig up some rare filters what not but still st speed comes first. And a 8700k is faster than any HEDT here anyway. Its a better processor for those task.
A lot of people dont realize it doesnt need to have the stamp HEDT on the box to solve the problem.
 

ub4ty

Senior member
Jun 21, 2017
749
307
96
#39
For normal usecases of the application you mentions a 8700k is faster anyway. Sure you can dig up some rare filters what not but still st speed comes first. And a 8700k is faster than any HEDT here anyway. Its a better processor for those task.
A lot of people dont realize it doesnt need to have the stamp HEDT on the box to solve the problem.
For normal use cases, a basic 4 core from intel from any year is good... Which is an example of how long Intel sat on their behinds. 8700k is great for gaming... It's not good for HEDT because it isn't an HEDT processor and has nowhere near the I/O, core count, or memory support.

My $170 Ryzen 7 1700 beats an 8700k for my workflow btw. I game on an intel 4 core that clocks higher than my 1700 and 4 cores is just about all I need for gaming.
I surf the web/general use on a low power 2 core intel processor equipped machine.
For more casual browsing, this drops down to an ARM processor equipped machine.
4k netflix on an even lower power ARM equipped media box [hardware accelerated decoding].. don't need much processing power.

For serious multi-threading, I use AMD processors exclusively. Better performance/Better Value hands down. This can change in the future and I hope Intel does. Until then, they're nowhere on the radar.
 
Last edited:

BigDaveX

Senior member
Jun 12, 2014
314
27
101
#40
That monolithic die strategy is obsolete as all hell right now. Its dead. It can't compete.
God, I feel old now. Seems like only yesterday Intel were the ones whose crudely slapped-together MCM solutions were being widely jeered and ridiculed, and AMD's monolithic designs were being heralded as the way of the future. :p
 

JoeRambo

Senior member
Jun 13, 2013
632
35
136
#41
God, I feel old now. Seems like only yesterday Intel were the ones whose crudely slapped-together MCM solutions were being widely jeered and ridiculed, and AMD's monolithic designs were being heralded as the way of the future. :p
That is cause of Intel marketing. Had they called their solution "Infinity Bus" instead of FSB, things would have been different.

Jokes aside, monolithic was, is, and will always be better for performance.
 

Hans Gruber

Senior member
Dec 23, 2006
452
15
91
#42
I never worry about Intel. Remember they pushed skylake back 6 months because AMD had nothing. Now that Ryzen is a success what will Intel do? No more tick tock stuff from Intel I am guessing. Probably good times again for do it yourself computer builders in the near future.
 
Dec 31, 2016
182
18
71
#43
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.
So aren't you trying to define the HEDT (in a way that is beneficial to AMD)? Isn't that bias? If the workloads rely on AVX number crunching then the Intel is ahead. Those are rarely consumer workloads at the moment. That's a real use case for some and in that case Threadripper wouldn't offer the best performance. Perf/€ mightbe a different story though. It just depends on the use case and wether or not one cares about perf/€.
 

maddie

Platinum Member
Jul 18, 2010
2,252
169
136
#44
That is cause of Intel marketing. Had they called their solution "Infinity Bus" instead of FSB, things would have been different.

Jokes aside, monolithic was, is, and will always be better for performance.
Nope.

Not in all cases. The more cores running together, the greater the chance of slower, lower performing ones included in the mix. Exactly where the crossover point is depends on many factors, but it exists.
 

PeterScott

Platinum Member
Jul 7, 2017
2,503
78
96
#45
What is HEDT? I know what the acronym stands for, but what is the purpose of HEDT? Or perhaps what is your individual need for HEDT.

Because it really looks like for any modern comparison, it depends heavily on what applications you need to run.

Is "winning" HEDT merely having the most cores/$, then AMD wins.

But is the i9-9900K also HEDT? If the clock speed rumors are correct, I bet it is will be outselling other HEDT processor by a large margin.
 

maddie

Platinum Member
Jul 18, 2010
2,252
169
136
#46
HEDT was, and is, a pure marketing creation.

Computers lie on several curves, perf/W, # of cores, max F, etc. Marketing assigns terms to various areas, but there will always be individuals that want some non-conventional combination. These fora probably have a lot from this group.
 

BigDaveX

Senior member
Jun 12, 2014
314
27
101
#47
Nope.

Not in all cases. The more cores running together, the greater the chance of slower, lower performing ones included in the mix. Exactly where the crossover point is depends on many factors, but it exists.
All else (core count, clockspeed, memory channels, cache size etc.) being equal, a monolithic chip is going to perform better than an MCM chip, but the difference between the two now would likely be smaller than it was a decade ago. You wouldn't get the same situation you had with, for instance, Quad FX, where NUMA and cache coherency issues ended up killing single-thread performance.

(At least, I'm assuming you wouldn't. Waiting to see what single-thread performance on the 4-die Threadrippers is like)
 

maddie

Platinum Member
Jul 18, 2010
2,252
169
136
#48
All else (core count, clockspeed, memory channels, cache size etc.) being equal, a monolithic chip is going to perform better than an MCM chip, but the difference between the two now would likely be smaller than it was a decade ago. You wouldn't get the same situation you had with, for instance, Quad FX, where NUMA and cache coherency issues ended up killing single-thread performance.

(At least, I'm assuming you wouldn't. Waiting to see what single-thread performance on the 4-die Threadrippers is like)
This is the thing, "all else being equal". A mythical beast.
 

CatMerc

Golden Member
Jul 16, 2016
1,111
32
106
#49
I'm not intending to enter this debate as there are a lot of angles to look at it from and I don't have the time right now, but Mindfactory sales are a good guage of ENTHUSIAST and DIY users. Both are very small minorities compared to the real money makers.

Skylake-X is clearly unpopular in DIY as can be easily seen from both Amazon CPU best sellers and Mindfactory, but in OEM these parts sell absurdly more. Probably generating the same revenue as the entirety of AMD right now.

So yes, TR is a great success for AMD, but don't lose perspective.
 

eek2121

Senior member
Aug 2, 2005
289
0
116
#50
All else (core count, clockspeed, memory channels, cache size etc.) being equal, a monolithic chip is going to perform better than an MCM chip, but the difference between the two now would likely be smaller than it was a decade ago. You wouldn't get the same situation you had with, for instance, Quad FX, where NUMA and cache coherency issues ended up killing single-thread performance.

(At least, I'm assuming you wouldn't. Waiting to see what single-thread performance on the 4-die Threadrippers is like)
IPC will likely be the same as 1st gen Threadripper. The boost algorithm will improve performance.
 

Similar threads



ASK THE COMMUNITY