Review Cascade-X Review and Availability Thread

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Diamond Member
Apr 15, 2012
6,696
12,373
136
Review embargo I believe lifts at 9 am ET so I will update this post as soon as reviews are published. As always, if there is a review you don't see in this post that you would like to have added, please PM me.

Video reviews
Linux Tech Tips

Print reviews
Hothardware
Tweaktown
Anandtech

For Sale
10980xe and 10920x listed on Newegg but both out of stock.
 
Last edited:

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
Why shouldn't the 3950X be compared to the 10980XE? They perform about the same with the former even beating the latter in many common use cases for these CPUs.
Because HEDT? Are we looking strictly at CPU performance only now? The 3950x and 9900K(S) beats AMD's own TR3s in many light workloads. What distinguishes HEDT from MAINSTREAM is platform features and more cores. The 3950x only meets one of those criteria.

In all of these charts, we have another line with PCIe Gen but we decided to go with the number of lanes, then the effective Gen3 bandwidth multiplying lanes by 2 for Gen4 lanes. We are still early in the Gen4 product cycle which means there are still more Gen3 devices out there. In the case of the Ryzen 3950X, despite having similar PCIe bandwidth as the Core i9-10980XE, using a PCIe Gen3 x16 GPU and x16 NIC (e.g. a 100GbE adapter) is not possible. With more lanes, the HEDT parts can handle that scenario.
As some background here, I personally have been on the journey from the dual Intel Xeon workstations (from Nahelem to Broadwell.) A reason I cannot use a mainstream part is the relatively small maximum memory size (64GB with full speed memory) as well as the lack of PCIe lanes. 1GbE and 10GbE are great, but even my home network has been on 40GbE/ 25GbE since 2015 just to get to network storage. That means I need more PCIe lanes for networking, local NVMe, and a GPU. The mainstream platforms simply fall short of servicing this use case. That is where HEDT parts, like the Core i9-10980XE and Threadripper parts, show an enormous benefit.

.............................

When we again truncate results, one can see the appeal of the Intel Xeon W-3275, albeit at a higher price. If the features like AVX-512 and large ECC memory capacities are going to be used, then Intel has a strong value proposition. Conversely, if one does not need those features, then the new Threadripper parts offer a strong value. At only $979, when we look in this view the Core i9-10980XE looks like a part appropriately priced between the Ryzen 3950X and the Threadripper 3960X, so long as the platform features are important.

 
  • Like
Reactions: CHADBOGA

beginner99

Diamond Member
Jun 2, 2009
5,318
1,763
136
This is really sad how uncompetitive they become, it reminds me of AMD from 4 years ago.

That is a pretty big exaggeration. Intel still is faster single-threaded and compared price by price competitive even multi-threaded. AMD before Zen might have been cheap but their CPU was just so much slower you couldn't really recommend it because the 8-core was worse than a cheap dual-core Intel one for gaming. If you are >50% slower in ST, MT simply can't save you.

In fact given the process node advanatge AMD has vs Intel (and NV) their lead is not that big outside of power use in CPU.

EDIT: About availability I contacted a big workstation builder here and it seems both Intel and AMD HEDT will not be really available in Europe in the next couple weeks, probably have to wait for 2020.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,730
136
Because HEDT? Are we looking strictly at CPU performance only now? The 3950x and 9900K(S) beats AMD's own TR3s in many light workloads. What distinguishes HEDT from MAINSTREAM is platform features and more cores. The 3950x only meets one of those criteria.
If platform features and AVX512 are the sole meaningful plus points of Cascade Lake-X, then it must be said that it exists only to satisfy a small niche. That niche includes use cases which depend on Intel MKL. For the vast majority of users who are in the market for >=16 core CPUs the 3950X is the better choice, followed by TR 3000.
 
  • Like
Reactions: lightmanek

TheGiant

Senior member
Jun 12, 2017
748
353
106
Not so fast! Go look at the Rome benchmarks vs Xeon Platinum, and watch 64c Rome win AVX512 benchmarks. That's against 28c Xeons. 64c TR3 against 18x 10980XE in an AVX512 benchmark = AMD victory. Sorry to burst your bubble.
Sorry to burst your bubble.
I am talking about 4ghz AVX512 overclock of 10980xe with proper optimisation
Its definitely not a lot of apps but there are some. Look at phoronix and if there is linear scaling it might be possible.
Or tied
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,730
136
Sorry to burst your bubble.
I am talking about 4ghz AVX512 overclock of 10980xe with proper optimisation
Its definitely not a lot of apps but there are some. Look at phoronix and if there is linear scaling it might be possible.
Or tied
Y-cruncher is about as optimized as it gets(apart from LINPACK perhaps) and even there these high core count CPUs are memory bandwidth starved. You can forget about linear scaling except in very specific scenarios.
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
21,067
3,574
126
The x299 board is too antiquated.

Only a handfull offers Bifurication.
None of them are PCI-E gen.4
Intel ROC chip is utter nonsense as well as BS.

These prices are 1 generation too late as well.
The 9980X should of been 1000 dollars, and not the price its at.
And the 10k series should have a board upgrade like possibly a X299 Rev.2 with all the things i mention up top.
Otherwise it seems like intel is fooling itself in people thinking this is a good buy.
 
  • Like
Reactions: spursindonesia

TheGiant

Senior member
Jun 12, 2017
748
353
106
Y-cruncher is about as optimized as it gets(apart from LINPACK perhaps) and even there these high core count CPUs are memory bandwidth starved. You can forget about linear scaling except in very specific scenarios.
tell me that, CFD is as bandwitch starved as it gets
but there are benches around which dont look bandwitch constrained
anyway, you must really know what you are buying to find a use for this cascade lake x when the new ryzens are here
 

Topweasel

Diamond Member
Oct 19, 2000
5,437
1,659
136
Because HEDT? Are we looking strictly at CPU performance only now? The 3950x and 9900K(S) beats AMD's own TR3s in many light workloads. What distinguishes HEDT from MAINSTREAM is platform features and more cores. The 3950x only meets one of those criteria.



But as I brought up. Where are these Benchmarks that would distinguish a 3950x from a 10980xe. I get what it is missing. But where is the practical application of it in reviews. You mention the light workloads but the 3950x beats it in a lot of heavy workloads as well. It's certainly a hybrid CPU. HEDT class in compute but consumer in platform. But even then with the 570x you start to see a lot of erroding functionality difference. It has twice the perf from the chipset, offers nearly the same PCIe throughput. You get NVME performance of striped setup from a single drive but you can also stripe that PCIe 4.0 NVME setup on this consumer platform. So realistically we are just down to memory capacity and memory bandwidth. But with 32GB sticks starting to appear capacity is starting to be less of an issue (though it also means these HEDT setups double up on capacity). So where are these benchmarks that would require the extra memory bandwidth? What do tools that require this extra bandwidth cost? Do they have ones that would run better on a slower HEDT CPU better than a faster consumer chip? If so how do things like the TR2 series compare with their slower and funky CPU performance but similar bandwidth?

When all the benchmarks of general prosumer and even fully pro tools says the "consumer" CPU is faster basically say the consumer CPU is faster then what are you actually getting out of the HEDT lineup?
 

coercitiv

Diamond Member
Jan 24, 2014
7,374
17,480
136
Because HEDT? Are we looking strictly at CPU performance only now? The 3950x and 9900K(S) beats AMD's own TR3s in many light workloads. What distinguishes HEDT from MAINSTREAM is platform features and more cores. The 3950x only meets one of those criteria.
Where were you last year when some posters tried to explain on these forums that TR2 2920X @ $649 would not be in direct competition with the cheaper mainstream 9900K?

AtenRa could have used your expert opinion while he was attempting to show to everybody that "the i9 9900K is not suited for workstation/professional workloads. It is better to invest in TR2 or Intel 2066 socket." Such a pity you weren't there to help. Oh wait, but you were there, defending the 9900K by not only avoiding to speak about "platform features and more cores", but also outright deflecting the discussion back to PERFORMANCE.

This is in contrast to speculation by some well known pro AMD posters on this forum who believe the 12 core AMD destroys the 9900k in multithreaded in the vast majority of cases.
But not performance per watt, which Is his original argument. Plus, the 9900k held it's own. It even won more than a couple of the benchmarks in the HEDT section. Well, not only Cinema 4D is missing. I may be wrong but I didn't see any DAW/Audio tests either...…..
That's as real world as it gets. Plus latency is critical in DAW and we know how Threadripper does in that category. Other factors are hard pagefaults and cpu stalls etc, which all deal with cache and mem subsystem.
Edit: Plus, from the AnandTech review, Intel handles legacy code better than AMD

Those were the times, when someone tried to explain that workstations were suited for i9 7920X while someone else kept saying 9900K beats 2920X in select workloads. Times never change, do they?
 

Nereus77

Member
Dec 30, 2016
142
251
136
Because HEDT? Are we looking strictly at CPU performance only now? The 3950x and 9900K(S) beats AMD's own TR3s in many light workloads. What distinguishes HEDT from MAINSTREAM is platform features and more cores. The 3950x only meets one of those criteria.




Dude, if you like single threaded performance so much, get a quad-core Intel CPU, turn off SMT and overclock it to 5.1GHz. Then you can keep telling yourself that its faster than a 9900K and 10980XE.

The whole idea of 18, 24, 32 and 64 core CPUs is entirely for multithreaded workloads and a lot of them. You're supposed to hook up 5 monitors and do as much as possible at the same time. This is the entire point of the exercise. This is the sort of thing that they exist for. These machines are basically servers that can do normal every day stuff as well big bad enterprise-level workloads. Different beasts entirely to the gaming Ryzens and Core i7s. If a 9900K can beat the thing by a few frames in gaming then whoop-de-doo. However, it will be annihilated in multi-threaded workloads because that is the entire purpose of these beasts.

Analogy: Your speedboat might be faster than the aircraft carrier, but how many jets can land on it?
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
What I don't like about these HEDT platforms is their power consumption. Both load and idle, desktop platforms are way more efficient in this regard. ArsTechnica has 3970x IDLE system power usage @167W. That is outrageous if there is nothing wrong with their setup. Sure it is fast when workload makes use of 32C, but all those e-peen and "have an itch" users are gonna be burning ridiculous amounts of electricity.
Intel is somewhat better at 69 watts, but that is also nasty compared to desktop platform efficiency, one pays hefty power tax if said system is not making use of those extra cores most of the time.

Developing and compiling something once in a while? Pay power tax 24/7 to cut compile times. Same with other workloads that are intermittent and not say rendering/encoding 24/7 at full load.
 

DrMrLordX

Lifer
Apr 27, 2000
22,937
13,023
136
Sorry to burst your bubble.
I am talking about 4ghz AVX512 overclock of 10980xe with proper optimisation

64c TR3 is going to beat a 4 GHz 10980XE running AVX512.


It's hard to make any estimates based on the EPYC 7742 (x2) results since the clocks aren't static, but the 3700x @ 4.3 GHz is telling: with DDR4-3600, it turns in a time of 41.545s. A 3990X @ 3 GHz all-core static OC should turn in a time of around 7s. A 10980XE @ AVX512 clocks of 4 GHz should complete the same benchmark in around 22s.

@JoeRambo

I could see that being a problem if you're looking at HEDT for workstation deployments or something like that. But one appliance burning ~170W in the home is not that big of a deal except in the most extreme of circumstances. Years ago I had lightbulbs that burned 60-100W on their own.
 
Last edited:
  • Like
Reactions: lightmanek

Abwx

Lifer
Apr 2, 2011
11,885
4,873
136
64c TR3 is going to beat a 4 GHz 10980XE running AVX512.


It's hard to make any estimates based on the EPYC 7742 (x2) results since the clocks aren't static, but the 3700x @ 4.3 GHz is telling: with DDR4-3600, it turns in a time of 41.545s. A 3990X @ 3 GHz all-core static OC should turn in a time of around 7s. A 10980XE @ AVX512 clocks of 4 GHz should complete the same benchmark in around 22s.
.

No need of 64C, the 3960X already outperform the 10980XE in y-Cruncher, the latter is only 45% faster than a 3700X, dunno how were performed the tests you re linking but Anand got the same results as Computerbase.de.




y-Cruncher is also AVX-512 accelerated, however when all the 32 threads come together in MT mode on AVX2, having that many pushes through 18 cores with AVX-512.


113594.png
 
  • Like
Reactions: lightmanek and IEC

DrMrLordX

Lifer
Apr 27, 2000
22,937
13,023
136
No need of 64C, the 3960X already outperform the 10980XE in y-Cruncher, the latter is only 45% faster than a 3700X, dunno how were performed the tests you re linking but Anand got the same results as Computerbase.de.

Bear in mind that the 9980XE in that test was probably running @ stock with an AVX512 clockspeed of 2.8 GHz. With an AVX512 clock of 4 GHz, the same chip should turn in a score of around 5.1s in the 250m bench. You would have to OC the 3960x or 3970x to beat that score. Which is appropriate, since a 4 GHz AVX512 clockspeed for a 10980XE (or 9980XE) is a 42% overclock.
 
  • Like
Reactions: lightmanek

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,730
136
Bear in mind that the 9980XE in that test was probably running @ stock with an AVX512 clockspeed of 2.8 GHz. With an AVX512 clock of 4 GHz, the same chip should turn in a score of around 5.1s in the 250m bench. You would have to OC the 3960x or 3970x to beat that score. Which is appropriate, since a 4 GHz AVX512 clockspeed for a 10980XE (or 9980XE) is a 42% overclock.
Does y-cruncher scale linearly with frequency, especially on high core count CPUs?
 

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
I see you're intent on turning this into an HEDT vs Mainstream debate. I wonder why. Blender 2.79 incorporates AVX optimizations, and the highly overclockable chip on the older and less power efficient process node is hit with the far greater impact. Better cooling, and even a 100MHz less overclock to 4.8GHz could make a world of difference in power consumption while still maintaining high clocks.
Let's return to an HEDT vs HEDT debate.

In the video you kindly provided a link to, we can easily compare the overclocked 10980XE (521W) to the stock 3960X (3970X is 236W so I imagine 3960X is lower).

Adobe: 10980XE OC wins 2, and they tie on 1.
Blender: 3960X stock wins 2 of 2
7-zip: 3960X stock wins 2 of 2
Chaos Group V-ray: 3960X stock wins 1 of 1

Lowering the power consumption of the 10980XE won't bring it any more wins in those benchmarks.
 

KentState

Diamond Member
Oct 19, 2001
8,397
393
126
I've looked high and low for these online and they aren't available. Has anyone found the 10980 in the wild? I wonder if the 14nm shortage will make this impossible to find.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,260
16,117
136
I've looked high and low for these online and they aren't available. Has anyone found the 10980 in the wild? I wonder if the 14nm shortage will make this impossible to find.
Based on their performance, why would you even want one vs the new threadripper ?
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
Based on their performance, why would you even want one vs the new threadripper ?
Because it's cheaper and performs on par, clock for clock with TR3. Not everyone desires, or is ready to cough up the extra $400 - $1,000 if they don't have a use case for 24 or 32 cores.
 

heymrdj

Diamond Member
May 28, 2007
3,999
63
91
Based on their performance, why would you even want one vs the new threadripper ?

For me it’s because Microsoft and AMD don’t support Windows 10/Server 2016/Server 2019 Hyper-V Nested Virtualization on AMD.
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
What I don't like about these HEDT platforms is their power consumption. Both load and idle, desktop platforms are way more efficient in this regard. ArsTechnica has 3970x IDLE system power usage @167W. That is outrageous if there is nothing wrong with their setup. Sure it is fast when workload makes use of 32C, but all those e-peen and "have an itch" users are gonna be burning ridiculous amounts of electricity.
Intel is somewhat better at 69 watts, but that is also nasty compared to desktop platform efficiency, one pays hefty power tax if said system is not making use of those extra cores most of the time.

Developing and compiling something once in a while? Pay power tax 24/7 to cut compile times. Same with other workloads that are intermittent and not say rendering/encoding 24/7 at full load.
this, I said it many times we are heading wrong way
at this number of cores the best parameter to benchmark is OEE
but that is for the real use as you said, not for the olololol forum warriors
I wont buy a bus, even if its more affordable than ever before
but now enthusiast is called the bigger bus owner...ofc until he has to pay it
 

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
10980XE vs 3960x in 334 benchmarks on Linux

The 10980XE gives up 12 threads to the 3960x and is $400 cheaper but manages to win a quarter of the tests.

View attachment 13709
Well, of course. We don't expect performance to scale exactly with price. That's why a $190 Ryzen 3600 is only 10% behind a $525 9900KS at 1080p gaming.

Interesting results nonetheless:

Narrow margin tests
67 of the tests were within 10% of each other.
Intel won 43% of those tests. Of Intel's 86 wins, 33% were among this narrow victory group.
AMD won 57% of those tests. Of AMD's 248 wins, 15% were among this narrow victory group.

Blowouts
Of all the test >50% margin (86 total), AMD won 80 and lost 6.

Non-blowouts
If we completely exclude the bloodbath results (>50% one way or the other) there are 248 tests.
AMD won 168 (almost exactly 2/3) and Intel won 80 (almost exactly 1/3).

The question is whether paying 43% more in order to have the fastest chip for 75% of the time is worth it, and as with everything, it's workload dependent.