Thank you AMD for strong arming the serial-milker Intel. Price cut from $1900 to $900. Hard to believe Intel would cut down on the core count milking. Am I dreaming?
Mind it was Intel and without any competition who lowered mainstream price from $500 to $300 back in 2011. 3700X would cost $500 now if not that. So thank you, Intel.
Balderdash. Near the end of 2010 you could get a 6-core AMD Phenom II CPU for $270. By 2011, an "8 core" 4-module Bulldozer cost the same and Intel's products were priced to compete with that.
Sorry, but you may want to dig back into the review archives for some untinted perspective on that info. In fact, you're both incorrect. That mainstream performance pricing shift happened in 2008, with the introduction of the Yorkfield Core 2 Quad chips, and more importantly the Bloomfield Core i7 (with the i7-920 becoming that $300 darling entry point for the enthusiast platform).
Those mid-2010 Thuban Phenom II X6 chips? Passably competitive with the lowest end quad-core Bloomfield chips (which were already on the market for a year and a half prior). And let's just forget about the dumpster fire that was Bulldozer, given that fabled FX-8150 had trouble matching the old Thuban Phenom II X6 at launch.
The FX 8350 & FX 8320 Piledriver CPUs saved AMDs bacon and proved to be more than enough for modern PC Gaming back in the day. They held great price/performance and were highly cost effective. Among Steamroller, then Excavator for the APU markets, they've held AMD afloat just in time for the superior ZEN launch. Now sit back and watch Intel finally struggle, deservingly so.
According to the official Spec submissions nope, Zen 2 is on pair with Skylake (bypassing the huge L3). More or less AMD enlarged the L3 just to have the lead, this have a cost obviously, the 7nm silicon is pretty expensive. Try to image a 9900K with 32MB of L3 with the same latency. You will have a winner.
gondalf sorry but zen 2 does have better ipc then intel does now.. why else does intel need such high clocks to compete with lower clocked chips ?? explain that one.. clock for clock.. zen 2.. has better ipc...
gondalf : According to the official Spec submissions " what spec submissions page ? also.. if you think having a large L3 cache is the reason why zen has more ipc then intel, then there is something wrong. if that was the case, WHY didnt intel do the same with the 10xxx series they just released ? keep in mind, 10xxx series, as 10 megs more of L2 to play around with... lets see you explain that as well.
Adding L3 cannot increase processing. The L3 can only improve feeding of data, further the L3 is a victim cache, the data has to be expelled from the L2 first.
It doesn't matter how big the fuel line is on your 4 cylinder, it's only going to burn so much gas. Same for the L2 and L3. If the size of the cache increases the IPC that is *only* because the cache was too small for the design in the first place.
keep in mind, the comment is from gondalf, he will say any thing to make his beloved intel look better, as you can see, he DIDN'T answer my question to him as well...
You are delusional. 2011 is the year for 2500K/2600K release, and since then Intel has been charging 300+ for quad core till 2017 Ryzen release. It was also the six darkest years in CPU history where we see like 5% increase in IPC every year, I kept my 4.5GHz overclocked 2600K for 6 years because there was no reason to upgrade.
Yeah that was part of the issue. Sandy Bridge had so much overclocking headroom, you could put a good AiO on it, crank it up to 4.8-5.0 GHz, and generations later the competition would just barely catch up. The percentage of difference between the two was very small, and Bulldozer was chasing Core i3s.
You're not alone buddy. I've held on to my Icy Bridge 3700K until Ryzen 39**X because Intel was offering no innnovation to the market.
I distinctly remember the Anandtech article for IIRC the Kaby Lake Intel processors where they basically said this was the first generation to be 20% better than Sandy Bridge/Icy Bridge which made is worth upgrading. That was 6 years without any performance increases.
Make no mistake, without AMD competition we wouldn't have moved beyond 8 cores on the desktop or 12 cores in the HEDT. Intel was happy to sit on their fingers and rake in the money with 2-5% improvement per year. In fact 3 solid years of AMD competition have doubled core counts on both the desktop and server and at the same time lowered prices across the board. Without AMD there is no innovation at Intel because they don't have competition. Thank god for Lisa Su.
Bollocks. Pulling arbitrary dollar values of nameless CPUs out of your behinds and linking even more arbitrarily 2011 CPUs to 2019 CPUs is an extremely poor tactic. Your suck at this (-->Intel apologetics). Be better so we can have meaningful arguments :)
Read the article your posting spam at. The author mentions the 1900 and 900 numbers. I'll let you guess which page. You might actual read the review then.
Do not thank Intel for anything CPU related. They've been milking consumers for many years, confusing people with several different sockets, chipsets, price points etc., this processor not compatible with that socket and so on. They've caused an industry mess with very expensive CPUs and miniscule IPC increases. Who to thank? AMD for launching ZEN and catching Intel by the surprise.
While your at bringing up irrelevant points from over 7 years ago, might as well thank AMD for X64 as well. Are you going to fellate Intel for everything they've done in the past?
evernessince dont forget the on die memory controller that intel also copied from amd :-) not irrelevant points, they are valid points.. it kinda proves intel likes to milk people for all they are worth, while stagnating the cpu industry, and over charging for their cpus as well. come on, the top sku for 10xxx cpus is 1k less then the cpu it is replacing, and it cant even do that for the most part, may as well just stick with the 9xxx cpus....
While technically Intel was first to 64bit X86, AMD beat intel to X86-64bit, meaning it was both x86(32bit) and x86(64bit) compatible. Intel tried to go 64bit only, but it backfired on them hard, and so they scrambled to do both 32 and 64 bit, but AMD beat them to the punch, so much so Intel had to license that from AMD, and is the sole reason it's still known today as AMD-64
Xyler94 i remember reading that good old microsoft didnt want to have to code windows for 2 different x86-64 instruction sets, so they made intel drop theirs and adopt AMDs instead, as they had already started programming windows for amd64...
You realize the only reason Intel lowered the mainstream bracket to the $300 level was competition from AMD's Black Edition CPU's. And after Ivy Bridge, they shot right back up to $400.
So stop pretending Intel doesn't price their products based on competition from AMD. That's just ridiculous.
Yeah well, lower prices are nice and all, but the product isnt that good. Its still the Skylake architecture, still ancient 14 nm, still has dozens of security flaws.
I don't care about process nodes, as long as they're delivering competitive prices, core counts, and performance per core. Intel's not quite out of the game yet since AMD's HEDT goes higher than Intel's, but they've gotten smashed at the halo spot, and they won't be able to deliver on price and performance if they can't get something in order.
No they haven't been "smashed at the halo spot". The 3900X and 3950X are both beasts and both shred in most of the important benchmarks. For video rendering both the 3900x and 3950X hand out with both the threadrippers and the intels. You get 90% of the performance for 1/4th the price. 12-16 cores is also a very important number for programmers, since you have enough CPUs for compiling, and running 2-3 VMs comfortably.
Why is it that Intel gains so incredibly much more from AVX512 than AMD gains from AVX2?
In the 3DPM2 test, the AMD CPUs gain roughly a factor of two in performance, which is exactly what I'd expect given that AVX2 is twice as wide as standard SSE. The Intel CPUs, on the other hand, gain almost a factor of 9, which is more than twice what I'd expect given that AVX512 as four times as wide as SSE.
What causes this? Does AVX512 have some other kind of tricks up its sleeves? Does opmasking benefit 3DPM2?
Basically, AVX-512 is double the performance of AVX2 (or another way to see it, 256bit vs 512bits, which 512 is double 256). So anything optimized for 512 will be about double in speed from 256, even on the exact same processor.
Zen does not support AVX-512 instructions. At all.
AVX-512 is not simply AVX-256 (AKA AVX2) scaled up.
Something to consider is that AVX-512 forces Intel chips to run at much slower clock speeds, so if you're mixing workloads, using AVX-512 instructions could easily cause overall performance to drop. It's only in an artificial benchmark situation where it has such a huge advantage.
Obviously, AMD just caught up with Intel's 256-bit AVX2, prior to Ryzen 3 AMD only had 128-bit AVX2 AFAIK. It was the only reason I bought into a cheap Ryzen 3700X Desktop (under $600US complete and prebuilt). To get the same level of AVX support, bitwise.
I've been using Intel's Fortran compiler since 1983 (back then it was on a DEC VAX).
So I only do math modeling at 64-bits like forever (going back to 1975), So I am very excited that AVX-512 is now under $1KUS. An immediate 2X speed boost over AVX2 (at least for the stuff I'm doing now).
I'd be curious how much the AVX512 is used by people. It seems to be a highly tailored for only big math operations which kinda limits it's practical usage to science/engineering. In addition the power use of the module was massive in the last article I read, to the point that the main CPU throttled when the AVX512 was engaged for more than a few seconds.
I'd be really curious what percentage of people buying HEDT are using it, or if it's just a niche feature for science/engineering.
If you don't need AVX512 you probably don't need or even want a desktop computer. Not when you can get an 8-core/16-thread MacBook Pro. Desktops are mostly built for show and playing games. Most real work is getting done on laptops.
Video processing and image processing can also benefit from AVX512. Many AI algorithms can benefit from AVX512. Problem for Intel is that in many cases where AVX512 gives good speedup, GPU would be even better choice. Also software support for AVX512 is lacking.
Not so! https://software.intel.com/en-us/parallel-studio-x... It compiles and runs on both Intel and AMD. Full AVX-512 support on AVX-512 hardware. You have to go full Volta to get true FP64, otherwise desktop GPU's are real FP64 dogs!
There are tools and compilers for software developers, but not so much end user software actually use them. FP64 is mostly required only in science/engineering category. Image/video/ai processing is usually just fine with lower precision. I'd add that also GPUs only have small (<=32GB) RAM while intel/amd CPUs can have hundreds of GB or more. Some datasets do not fit into a GPU. AVX512 still has its niche, but it's getting smaller.
I asked about this a couple of months ago. Apparently the 3DPM2 code uses a lot of 64b integer multiplies; the AVX2 instruction set doesn't include packed 64b integer mul instructions - those were added with AVX512, along with some other integer and bit manipulation stuff. This means that any CPU without AVX512 is stuck using scalar 64b muls, which on modern microarchitectures only have a throughput of 1/clock. IIRC the Skylake-X core and derivatives have two pipes capable of packed 64b muls, for a total throughput of 16/clock.
I do wish AnandTech would make this a little more clear in their articles though; it is not at all obvious that the 3DPM2 is more of a mixed FP/Integer workload, which is not something I would normally expect from a scientific simulation.
I also think that the testing methodology on this benchmark is a little odd - each algorithm is run for 20 seconds, with a 10 second pause in between? I would expect simulations to run quite a bit longer than that, and the nature of turbo on CPUs means that steady-state and burst performance might diverge significantly.
For me the most important CPU features are AVX512, very high Ghz, and lots of fast memory. Plenty of PCI lanes are a significant plus, because they feed all the Volta GPUs ... which of course are PCIe-3. I don't care much about high core counts because that's the point of the GPUs. Mainly I need a few very fast cores to handle all the stuff that can't be massively parallelized. So, this chip checks all the boxes, especially at half the price! :-)
Looking at the retailers today, the price landed at about $60 higher than prices listed here (and everywhere else leading up to release). 10900x is $649, 10920x is $749, etc
Yes I saw that, but thought it might be helpful to some to know actual retail pricing. Also worth noting that the markup landed at the high end of the range given and then some.
‘consumers don’t care about process nodes, so you shouldn’t either’
The ex-CEO said, Intel processors function as intended, no bug, so using the design shortcut of partial addresses, causing many more security holes than AMD's, is intended, and enterprise and consumers don't care, so you shouldn’t either!
The more the security holes, the more the demand; what an amazing company and business!
Intel Processors even dating back to 2008+ all have massive amounts of security and malware vulnerabilities nightmare. Not only is ZEN superior technologically, its faster, securer and much more cost effective.
You state that the AMD loses on PCI lanes but those PCI lanes are 4.0 vs 3.0. They are twice as fast per lane. With the right hardware the total bandwidth is the exact same.
Which means nothing if you can't use your RAID controller or 10g network card because there aren't enough lanes to create the required expansion slots.
I don't think there's a single X570 board capable of running a computer with a RAID controller or 10g network card, both of which require x8 slots. You could, in principle, bifurcate lanes to create more slots, but no one does that. So the fact that the lanes are PCIe 4.0 is utterly irrelevant.
If you want a non-toy computer, you need either Intel's -E/-X series processors, or, since 2017, AMD's Threadripper processors.
That summary makes no sense. The desktop AMD chip is beating the HEDT one with a $500 savings when you factor in motherboard cost. It's even beating it in gaming.
If you picked up an older threadripper part for productivity it will walk all over the HEDT part and still be cheaper.
It doesn't matter what Intel does there's an AMD part available for cheaper.
TR 2000 is now cheaper, but its per-core performance lags behind Cascade Lake-X. It also has a higher-latency topology. If you want a real computer capable of running a RAID controller and 10g network card, for example, that also games reasonably well, and don't need really high core counts, then you'll get better results with Cascade Lake-X at the lower end.
AMD has blundered. They should have released a 16-core Zen 2 chip, made it compatible with X399, and made it no more expensive than the MSRP of the 2950X.
If you're going HEDT you need the cores that's the whole point. Further more if you need ECC you won't get that with these HEDT parts while you will on all Ryzen CPUs from the bottom to the top.
For HEDT ECC can be mandatory. If you want that with Intel you'll spend an extra $1000. Nope not joking.
i7-7740x anyone? That CPU was laughed out of existence. Even the 8 core AMD TR died a quick death. That's how many people need a low core count HEDT processor.
AMD NEEDS 20-core TR right under $1000 to fill the gap between $750 3950X and $1400 3650X and take the ground from under Intel's 18-core part. Unfortunately, they are almost out of numbers. 3955X? ;) They would benefit from a $800 16-core part too, perfect for those who are limited by PCI lanes or memory on 3950X.
My guess is that the vast majority of people willing to pay for a Threadripper motherboard, especially a TR4 board, went for the higher core count TR CPU's. I wouldn't be surprised if it was a case of AMD deciding not to waste money on making such low volume SKU.
He can't be certain what the markup will be, so he's using the tray prices. Not that I recommend that, but it's about as fair as he can be for the time being.
Message for intel, the only people who are willing to upgrade are those who already invested in the x299 lga2066 socket motherboards. So when you release a 10nm make sure you release at least one on LGA2066 socket. I am done with Intel if they go after the money and do one more socket change to annoy those who stuck with intel despite their stale processes. For it will make more sense to just move to AMD if you need to buy a new motherboard anyway, Atleast they will support the socket for a few generations of CPUs
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
79 Comments
Back to Article
milkywayer - Monday, November 25, 2019 - link
Thank you AMD for strong arming the serial-milker Intel. Price cut from $1900 to $900. Hard to believe Intel would cut down on the core count milking. Am I dreaming?regsEx - Monday, November 25, 2019 - link
Mind it was Intel and without any competition who lowered mainstream price from $500 to $300 back in 2011. 3700X would cost $500 now if not that. So thank you, Intel.Spunjji - Monday, November 25, 2019 - link
Balderdash. Near the end of 2010 you could get a 6-core AMD Phenom II CPU for $270. By 2011, an "8 core" 4-module Bulldozer cost the same and Intel's products were priced to compete with that.So, uh, thanks again AMD..?
karmapop - Monday, November 25, 2019 - link
Sorry, but you may want to dig back into the review archives for some untinted perspective on that info. In fact, you're both incorrect. That mainstream performance pricing shift happened in 2008, with the introduction of the Yorkfield Core 2 Quad chips, and more importantly the Bloomfield Core i7 (with the i7-920 becoming that $300 darling entry point for the enthusiast platform).Those mid-2010 Thuban Phenom II X6 chips? Passably competitive with the lowest end quad-core Bloomfield chips (which were already on the market for a year and a half prior). And let's just forget about the dumpster fire that was Bulldozer, given that fabled FX-8150 had trouble matching the old Thuban Phenom II X6 at launch.
nt300 - Tuesday, November 26, 2019 - link
The FX 8350 & FX 8320 Piledriver CPUs saved AMDs bacon and proved to be more than enough for modern PC Gaming back in the day. They held great price/performance and were highly cost effective. Among Steamroller, then Excavator for the APU markets, they've held AMD afloat just in time for the superior ZEN launch. Now sit back and watch Intel finally struggle, deservingly so.yeeeeman - Wednesday, November 27, 2019 - link
They were so crap that AMD could boast 50% better ipc with zen over bulldozer, while still being lower than Skylake.Korguz - Wednesday, November 27, 2019 - link
too bad zen has better ipc then intel now...Gondalf - Wednesday, November 27, 2019 - link
According to the official Spec submissions nope, Zen 2 is on pair with Skylake (bypassing the huge L3). More or less AMD enlarged the L3 just to have the lead, this have a cost obviously, the 7nm silicon is pretty expensive. Try to image a 9900K with 32MB of L3 with the same latency.You will have a winner.
Korguz - Wednesday, November 27, 2019 - link
gondalfsorry but zen 2 does have better ipc then intel does now.. why else does intel need such high clocks to compete with lower clocked chips ?? explain that one.. clock for clock.. zen 2.. has better ipc...
Qasar - Wednesday, November 27, 2019 - link
gondalf : According to the official Spec submissions " what spec submissions page ? also.. if you think having a large L3 cache is the reason why zen has more ipc then intel, then there is something wrong. if that was the case, WHY didnt intel do the same with the 10xxx series they just released ? keep in mind, 10xxx series, as 10 megs more of L2 to play around with... lets see you explain that as well.Korguz - Thursday, November 28, 2019 - link
yep.. i knew gondalf wouldnt answer my question...0ldman79 - Thursday, December 5, 2019 - link
That is ignorant.Adding L3 cannot increase processing. The L3 can only improve feeding of data, further the L3 is a victim cache, the data has to be expelled from the L2 first.
It doesn't matter how big the fuel line is on your 4 cylinder, it's only going to burn so much gas. Same for the L2 and L3. If the size of the cache increases the IPC that is *only* because the cache was too small for the design in the first place.
Korguz - Sunday, December 8, 2019 - link
keep in mind, the comment is from gondalf, he will say any thing to make his beloved intel look better, as you can see, he DIDN'T answer my question to him as well...airdrifting - Monday, November 25, 2019 - link
You are delusional. 2011 is the year for 2500K/2600K release, and since then Intel has been charging 300+ for quad core till 2017 Ryzen release. It was also the six darkest years in CPU history where we see like 5% increase in IPC every year, I kept my 4.5GHz overclocked 2600K for 6 years because there was no reason to upgrade.eek2121 - Monday, November 25, 2019 - link
Yeah that was part of the issue. Sandy Bridge had so much overclocking headroom, you could put a good AiO on it, crank it up to 4.8-5.0 GHz, and generations later the competition would just barely catch up. The percentage of difference between the two was very small, and Bulldozer was chasing Core i3s.rahvin - Monday, November 25, 2019 - link
You're not alone buddy. I've held on to my Icy Bridge 3700K until Ryzen 39**X because Intel was offering no innnovation to the market.I distinctly remember the Anandtech article for IIRC the Kaby Lake Intel processors where they basically said this was the first generation to be 20% better than Sandy Bridge/Icy Bridge which made is worth upgrading. That was 6 years without any performance increases.
Make no mistake, without AMD competition we wouldn't have moved beyond 8 cores on the desktop or 12 cores in the HEDT. Intel was happy to sit on their fingers and rake in the money with 2-5% improvement per year. In fact 3 solid years of AMD competition have doubled core counts on both the desktop and server and at the same time lowered prices across the board. Without AMD there is no innovation at Intel because they don't have competition. Thank god for Lisa Su.
Santoval - Monday, November 25, 2019 - link
Bollocks. Pulling arbitrary dollar values of nameless CPUs out of your behinds and linking even more arbitrarily 2011 CPUs to 2019 CPUs is an extremely poor tactic. Your suck at this (-->Intel apologetics). Be better so we can have meaningful arguments :)milkywayer - Monday, November 25, 2019 - link
Read the article your posting spam at. The author mentions the 1900 and 900 numbers. I'll let you guess which page. You might actual read the review then.milkywayer - Monday, November 25, 2019 - link
Whups. Meant it for RegsEx.milkywayer - Monday, November 25, 2019 - link
Out of the kindness of their heart. How generous and kind of them./s
nt300 - Tuesday, November 26, 2019 - link
Do not thank Intel for anything CPU related. They've been milking consumers for many years, confusing people with several different sockets, chipsets, price points etc., this processor not compatible with that socket and so on. They've caused an industry mess with very expensive CPUs and miniscule IPC increases.Who to thank? AMD for launching ZEN and catching Intel by the surprise.
evernessince - Tuesday, November 26, 2019 - link
While your at bringing up irrelevant points from over 7 years ago, might as well thank AMD for X64 as well. Are you going to fellate Intel for everything they've done in the past?Korguz - Tuesday, November 26, 2019 - link
evernessince dont forget the on die memory controller that intel also copied from amd :-)not irrelevant points, they are valid points.. it kinda proves intel likes to milk people for all they are worth, while stagnating the cpu industry, and over charging for their cpus as well. come on, the top sku for 10xxx cpus is 1k less then the cpu it is replacing, and it cant even do that for the most part, may as well just stick with the 9xxx cpus....
Xyler94 - Wednesday, November 27, 2019 - link
While technically Intel was first to 64bit X86, AMD beat intel to X86-64bit, meaning it was both x86(32bit) and x86(64bit) compatible. Intel tried to go 64bit only, but it backfired on them hard, and so they scrambled to do both 32 and 64 bit, but AMD beat them to the punch, so much so Intel had to license that from AMD, and is the sole reason it's still known today as AMD-64Qasar - Wednesday, November 27, 2019 - link
Xyler94 i remember reading that good old microsoft didnt want to have to code windows for 2 different x86-64 instruction sets, so they made intel drop theirs and adopt AMDs instead, as they had already started programming windows for amd64...Samus - Wednesday, November 27, 2019 - link
You realize the only reason Intel lowered the mainstream bracket to the $300 level was competition from AMD's Black Edition CPU's. And after Ivy Bridge, they shot right back up to $400.So stop pretending Intel doesn't price their products based on competition from AMD. That's just ridiculous.
Beaver M. - Monday, November 25, 2019 - link
Yeah well, lower prices are nice and all, but the product isnt that good.Its still the Skylake architecture, still ancient 14 nm, still has dozens of security flaws.
ksec - Tuesday, November 26, 2019 - link
Thanking AMD for Intel's Price Cut and then continue to buy Intel?JlHADJOE - Tuesday, November 26, 2019 - link
Now W3175X needs to drop down to $2000 for the 32-core TR3 to have any kind of competition.UglyFrank - Monday, November 25, 2019 - link
This could make a great workstation processor for me but AMD is still much compelling.Santoval - Monday, November 25, 2019 - link
Wait for the prices of both to adjust first.Drumsticks - Monday, November 25, 2019 - link
I don't care about process nodes, as long as they're delivering competitive prices, core counts, and performance per core. Intel's not quite out of the game yet since AMD's HEDT goes higher than Intel's, but they've gotten smashed at the halo spot, and they won't be able to deliver on price and performance if they can't get something in order.Braincruser - Tuesday, November 26, 2019 - link
No they haven't been "smashed at the halo spot". The 3900X and 3950X are both beasts and both shred in most of the important benchmarks. For video rendering both the 3900x and 3950X hand out with both the threadrippers and the intels. You get 90% of the performance for 1/4th the price. 12-16 cores is also a very important number for programmers, since you have enough CPUs for compiling, and running 2-3 VMs comfortably.Dolda2000 - Monday, November 25, 2019 - link
Why is it that Intel gains so incredibly much more from AVX512 than AMD gains from AVX2?In the 3DPM2 test, the AMD CPUs gain roughly a factor of two in performance, which is exactly what I'd expect given that AVX2 is twice as wide as standard SSE. The Intel CPUs, on the other hand, gain almost a factor of 9, which is more than twice what I'd expect given that AVX512 as four times as wide as SSE.
What causes this? Does AVX512 have some other kind of tricks up its sleeves? Does opmasking benefit 3DPM2?
Xyler94 - Monday, November 25, 2019 - link
Basically, AVX-512 is double the performance of AVX2 (or another way to see it, 256bit vs 512bits, which 512 is double 256). So anything optimized for 512 will be about double in speed from 256, even on the exact same processor.Xyler94 - Monday, November 25, 2019 - link
To note: That's a highly overly simplistic view of it, there's a lot more under the hood.eek2121 - Monday, November 25, 2019 - link
Well that and the obvious point that AMD CPUs do not support AVX-512.DanNeely - Monday, November 25, 2019 - link
AVX-2 is 256 bits wide, and thus only does have as much/instruction as AVX-512.JayNor - Monday, November 25, 2019 - link
I believe for 10 cores and up there are dual avx512 units per core. You can see the dual avx512 units in the Execution Engine diagram at this link.https://en.wikichip.org/wiki/intel/microarchitectu...
Also, cascade lake added dlboost 8 bit operations in avx512 to support ai inference convolutions.
Dolda2000 - Monday, November 25, 2019 - link
But Zen 1/2 also has two 256-bit FMAs per core. And Intel also has two SSE units per core as well, so I don't see how that would explain the ratios.Thanny - Wednesday, November 27, 2019 - link
Zen does not support AVX-512 instructions. At all.AVX-512 is not simply AVX-256 (AKA AVX2) scaled up.
Something to consider is that AVX-512 forces Intel chips to run at much slower clock speeds, so if you're mixing workloads, using AVX-512 instructions could easily cause overall performance to drop. It's only in an artificial benchmark situation where it has such a huge advantage.
Everett F Sargent - Monday, November 25, 2019 - link
Obviously, AMD just caught up with Intel's 256-bit AVX2, prior to Ryzen 3 AMD only had 128-bit AVX2 AFAIK. It was the only reason I bought into a cheap Ryzen 3700X Desktop (under $600US complete and prebuilt). To get the same level of AVX support, bitwise.I've been using Intel's Fortran compiler since 1983 (back then it was on a DEC VAX).
So I only do math modeling at 64-bits like forever (going back to 1975), So I am very excited that AVX-512 is now under $1KUS. An immediate 2X speed boost over AVX2 (at least for the stuff I'm doing now).
rahvin - Monday, November 25, 2019 - link
I'd be curious how much the AVX512 is used by people. It seems to be a highly tailored for only big math operations which kinda limits it's practical usage to science/engineering. In addition the power use of the module was massive in the last article I read, to the point that the main CPU throttled when the AVX512 was engaged for more than a few seconds.I'd be really curious what percentage of people buying HEDT are using it, or if it's just a niche feature for science/engineering.
TEAMSWITCHER - Tuesday, November 26, 2019 - link
If you don't need AVX512 you probably don't need or even want a desktop computer. Not when you can get an 8-core/16-thread MacBook Pro. Desktops are mostly built for show and playing games. Most real work is getting done on laptops.Everett F Sargent - Tuesday, November 26, 2019 - link
LOL, that's so 2019.Where I am from it's smartwatches all the way down.
Queue Four Yorkshiremen.
AIV - Tuesday, November 26, 2019 - link
Video processing and image processing can also benefit from AVX512. Many AI algorithms can benefit from AVX512. Problem for Intel is that in many cases where AVX512 gives good speedup, GPU would be even better choice. Also software support for AVX512 is lacking.Everett F Sargent - Tuesday, November 26, 2019 - link
Not so!https://software.intel.com/en-us/parallel-studio-x...
It compiles and runs on both Intel and AMD. Full AVX-512 support on AVX-512 hardware.
You have to go full Volta to get true FP64, otherwise desktop GPU's are real FP64 dogs!
AIV - Wednesday, November 27, 2019 - link
There are tools and compilers for software developers, but not so much end user software actually use them. FP64 is mostly required only in science/engineering category. Image/video/ai processing is usually just fine with lower precision. I'd add that also GPUs only have small (<=32GB) RAM while intel/amd CPUs can have hundreds of GB or more. Some datasets do not fit into a GPU. AVX512 still has its niche, but it's getting smaller.thetrashcanisfull - Monday, November 25, 2019 - link
I asked about this a couple of months ago. Apparently the 3DPM2 code uses a lot of 64b integer multiplies; the AVX2 instruction set doesn't include packed 64b integer mul instructions - those were added with AVX512, along with some other integer and bit manipulation stuff. This means that any CPU without AVX512 is stuck using scalar 64b muls, which on modern microarchitectures only have a throughput of 1/clock. IIRC the Skylake-X core and derivatives have two pipes capable of packed 64b muls, for a total throughput of 16/clock.I do wish AnandTech would make this a little more clear in their articles though; it is not at all obvious that the 3DPM2 is more of a mixed FP/Integer workload, which is not something I would normally expect from a scientific simulation.
I also think that the testing methodology on this benchmark is a little odd - each algorithm is run for 20 seconds, with a 10 second pause in between? I would expect simulations to run quite a bit longer than that, and the nature of turbo on CPUs means that steady-state and burst performance might diverge significantly.
Dolda2000 - Monday, November 25, 2019 - link
Thanks a lot, that does explain much.rolfaalto - Monday, November 25, 2019 - link
For me the most important CPU features are AVX512, very high Ghz, and lots of fast memory. Plenty of PCI lanes are a significant plus, because they feed all the Volta GPUs ... which of course are PCIe-3. I don't care much about high core counts because that's the point of the GPUs. Mainly I need a few very fast cores to handle all the stuff that can't be massively parallelized. So, this chip checks all the boxes, especially at half the price! :-)blobcat - Monday, November 25, 2019 - link
Looking at the retailers today, the price landed at about $60 higher than prices listed here (and everywhere else leading up to release). 10900x is $649, 10920x is $749, etcIrata - Monday, November 25, 2019 - link
That's because the price given was a price when buying 1,000 units. I think the article even stated that.Holliday75 - Monday, November 25, 2019 - link
It says the following at the bottom of page 1."*Intel quotes OEM/tray pricing. Retail pricing will sometimes be $20-$50 higher."
blobcat - Monday, November 25, 2019 - link
Yes I saw that, but thought it might be helpful to some to know actual retail pricing. Also worth noting that the markup landed at the high end of the range given and then some.Dorkaman - Monday, November 25, 2019 - link
How about some overclocking prime95 non avx stable. I'd love to see it against an overclocked 9900KS in games and rendering tests.Dionysos1234 - Monday, November 25, 2019 - link
Why no discussion of (lack of) ecc memory support?Jorgp2 - Monday, November 25, 2019 - link
That's probably baked into the chipset, and couldn't be added in if they wanted to.That's just Intel's idiotic product segmentation from the original launch biting them in the ass.
SBKch - Monday, November 25, 2019 - link
Could you add Ryzen 3950X to web benchmarks as I've noticed that it's missing?Sychonut - Monday, November 25, 2019 - link
Awesome! Looking forward to next generation on 14+++++++.wow&wow - Monday, November 25, 2019 - link
‘consumers don’t care about process nodes, so you shouldn’t either’The ex-CEO said, Intel processors function as intended, no bug, so using the design shortcut of partial addresses, causing many more security holes than AMD's, is intended, and enterprise and consumers don't care, so you shouldn’t either!
The more the security holes, the more the demand; what an amazing company and business!
nt300 - Tuesday, November 26, 2019 - link
Intel Processors even dating back to 2008+ all have massive amounts of security and malware vulnerabilities nightmare. Not only is ZEN superior technologically, its faster, securer and much more cost effective.HideOut - Monday, November 25, 2019 - link
You state that the AMD loses on PCI lanes but those PCI lanes are 4.0 vs 3.0. They are twice as fast per lane. With the right hardware the total bandwidth is the exact same.Thanny - Wednesday, November 27, 2019 - link
Which means nothing if you can't use your RAID controller or 10g network card because there aren't enough lanes to create the required expansion slots.I don't think there's a single X570 board capable of running a computer with a RAID controller or 10g network card, both of which require x8 slots. You could, in principle, bifurcate lanes to create more slots, but no one does that. So the fact that the lanes are PCIe 4.0 is utterly irrelevant.
If you want a non-toy computer, you need either Intel's -E/-X series processors, or, since 2017, AMD's Threadripper processors.
kc77 - Tuesday, November 26, 2019 - link
That summary makes no sense. The desktop AMD chip is beating the HEDT one with a $500 savings when you factor in motherboard cost. It's even beating it in gaming.If you picked up an older threadripper part for productivity it will walk all over the HEDT part and still be cheaper.
It doesn't matter what Intel does there's an AMD part available for cheaper.
Thanny - Wednesday, November 27, 2019 - link
TR 2000 is now cheaper, but its per-core performance lags behind Cascade Lake-X. It also has a higher-latency topology. If you want a real computer capable of running a RAID controller and 10g network card, for example, that also games reasonably well, and don't need really high core counts, then you'll get better results with Cascade Lake-X at the lower end.AMD has blundered. They should have released a 16-core Zen 2 chip, made it compatible with X399, and made it no more expensive than the MSRP of the 2950X.
kc77 - Wednesday, November 27, 2019 - link
If you're going HEDT you need the cores that's the whole point. Further more if you need ECC you won't get that with these HEDT parts while you will on all Ryzen CPUs from the bottom to the top.For HEDT ECC can be mandatory. If you want that with Intel you'll spend an extra $1000. Nope not joking.
Jimbo Jones - Friday, November 29, 2019 - link
"If you need HEDT but don't need cores ..."i7-7740x anyone? That CPU was laughed out of existence. Even the 8 core AMD TR died a quick death. That's how many people need a low core count HEDT processor.
peevee - Wednesday, November 27, 2019 - link
AMD NEEDS 20-core TR right under $1000 to fill the gap between $750 3950X and $1400 3650X and take the ground from under Intel's 18-core part. Unfortunately, they are almost out of numbers. 3955X? ;)They would benefit from a $800 16-core part too, perfect for those who are limited by PCI lanes or memory on 3950X.
Thanny - Wednesday, November 27, 2019 - link
As I see it, AMD is being daft by not releasing an X399-compatible 16-core Threadripper 3955X based on Zen 2.Korguz - Wednesday, November 27, 2019 - link
ever think that maybe they cant ??sarafino - Friday, January 10, 2020 - link
My guess is that the vast majority of people willing to pay for a Threadripper motherboard, especially a TR4 board, went for the higher core count TR CPU's. I wouldn't be surprised if it was a case of AMD deciding not to waste money on making such low volume SKU.umano - Wednesday, November 27, 2019 - link
I don't think that 10nm intel in 2021 will match or beat amd ryzen 5000. If they will match ryzen 4000 I'd say it's a miracleballsystemlord - Wednesday, November 27, 2019 - link
What happened to the 10980XE in the y-Cruncher MT test?hoohoo - Thursday, November 28, 2019 - link
@Ian - your price chart are misleading: you are using retail price for AMD and wholesale (ie per 1000 units) price for Intel.The 10980XE will retail with a markup, it will cost 25 or 30 percent higher than the $979 you are using here.
ballsystemlord - Thursday, November 28, 2019 - link
He can't be certain what the markup will be, so he's using the tray prices. Not that I recommend that, but it's about as fair as he can be for the time being.NetMage - Thursday, November 28, 2019 - link
"and so we expect our results here to be consummate with most users’ performance" - I certainly hope not. Perhaps you meant "commensurate"?Gonemad - Friday, November 29, 2019 - link
Tiny typo... Sub $1k sheet.:DRAM Capacity 256GB vs. 128MB.
Everybody know memories are in the 128GB order of magnitude... but... yeah... GB or MB?
'Cause we are never sure these days.
- Nitpicking.
sharath.naik - Thursday, December 19, 2019 - link
Message for intel, the only people who are willing to upgrade are those who already invested in the x299 lga2066 socket motherboards. So when you release a 10nm make sure you release at least one on LGA2066 socket. I am done with Intel if they go after the money and do one more socket change to annoy those who stuck with intel despite their stale processes. For it will make more sense to just move to AMD if you need to buy a new motherboard anyway, Atleast they will support the socket for a few generations of CPUs