2990WX review thread Its live !

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

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,432
136
the most interesting thing would be to see a comparison with a 32 core Epyc,
because the results are mostly bad outside of rendering, and it's likely the memory configuration, but how much is just the infinity fabric a problem for those tests is not clear without having Epyc to compare... hopefully someone will do that.

in any case, it looks to heavily compromised to be a viable high end desktop CPU,
Anandtechs review has the 32 core EPYC 7601, and yes memory is great on that thing.
 
  • Like
Reactions: Drazick

ub4ty

Senior member
Jun 21, 2017
749
898
96
Crikey.. those interconnect power numbers!
Ryzen cores themselves are very efficient, but the fabric penalty is staggering.

They're not staggering. They're inline with Intel's mesh uncore #'s and understood from an industry wide perspective to be what you'd expect. Intel's mesh has just about the same consumption. You want core scaling? This is what it costs. Cores don't connect via magic pixie dust. They require pretty active high speed interconnects that burn power.
Uncore%207980XE_575px.png


AMD has a major highway. Intel has tons of through streets. Both consume almost the same amount of power.
IF%20Power%202950X_575px.png

IF%20Power%202990WX_575px.png


Compare the various thread counts TOTAL Power Utilization between AMD and Intel. They're the same. You expect the highway to consume more power. It alleviates per junction (core<->interconnect) power utilization. So, It is what you can expect.. More consumption on AMD's highway (Infinity Fabric) and less power consumption at the cores. This is basic multi-core computer architecture taught in college. Both Intel and AMD's approach have limits. It's a matter of tradeoffs which unironically sum to the same power envelopes albeit one cost way less to manufacture.

From a high level, should the U.S have been connected with a bunch of through streets w/ no major highways? Of course not, traffic flow would be a nightmare. The same goes for cores. You want a highway. Everybody needs to eventually make highways (this includes Intel). Highways take a lot to construct and are pretty wasteful when they're not in use. Those interconnects have to be hot at all times. You can't play power games w/ it.

Btw, anandtech needs to clarify how they got their #'s :
In this case, the IF Power we measure includes the intra-silicon interconnect as well as the inter-silicon interconnect.
This seems a bit suspect and its not official #'s from AMD. I don't AMD or Intel would publish these #'s and I doubt there is a way to actually probe them. Also, is this uncore or IF? They say uncore for intel then IF for AMD? I'm calling shenanigans...

And here is the confirmation of shenanigans :
While based on the older first generation Zen cores, EPYC has additional memory controllers and IO to worry about, all of which fall under the uncore power category.

And even more confusion .. Is it IF or uncore?
This raises an interesting point – if we are purely considering the academic merits of one core compared to another, does the uncore power count to that contribution? For a real-world analysis, yes, but for a purely academic one? It also means I can claim the following prophecy:

After core counts, the next battle will be on the interconnect. Low power, scalable, and high performance: process node scaling will mean nothing if the interconnect becomes 90% of the total chip power.

Come on guys.. what is this? So, essentially my guess is that they're blindly doing some backhand calculation of subtracting core power utilization from total power utilization and THE FIGURE THEYRE REFERRING TO IS UNCORE POWER UTILIZATION for both AMD/INTEL. In this case, the numbers
are just fine for AMD because it has far more I/O in its UNCORE.

Reviewers need to start detailing the methodology behind arriving at these unofficial #'s.
I'm growing tired of the IPC (Instruction per clock) like memes.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
I was coming back here to say the same thing.
I knew it was going to be bad, but, holy smokes!
Direct link for those that miss it: https://www.anandtech.com/show/13124/the-amd-threadripper-2990wx-and-2950x-review/4

Those #'s are bogus and the writer needs to clarify what uncore is and how in the world they claim to have gotten IF power utilization #'s. They're doing back-hand napkin math as those #'s aren't accessible from Intel or AMD.
Pro-tip : AMD has almost double the I/O capacity in their uncore than Intel.
I highly doubt IF power utilization was measured which is why the article writer keeps contradicting his own commentary : IF in one sentence then uncore in another.

The numbers are going to largely sum to the same values for intel and AMD. You can take a highway and split it up into a bunch of through streets. You're going to end up w/ the same amount of road. Highways are more efficient not less.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
AT is refering the power draw of SoC domain as "IF Power", correct?
If that is the case then it is pretty inaccurate, as the SoC power plane powers the memory controllers and the IO as well (among other misc blocks).
Exactly, this is beyond a big gaff. The Article writer needs to fix this pronto. No way he has Intel or AMD's Interconnect power utilization #'s. This is just plain guess work and misinformation. The same bogus mess reviewers pull in regards to IPC (Instructions Per clock). No one is calculating that number properly.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
Some really impressive numbers in some benchmarks, but kinda underwhelming in others. It looks like AMD has done a great job of minimising NUMA overhead in two-die situations, but bump it up to four and performance plummets again.

EDIT: The 2990WX's gaming numbers on Tech Report make for ugly reading - performance is all the way down to Bulldozer levels in several cases. I was worried that the original Threadripper would be a repeat of Quad FX, which I'm happy to say turned out not to be the case, and isn't really the case for this one either since it at least does post some very solid multi-thread numbers (Quad FX couldn't even manage that), but still, clearly some major pitfalls that need to be worked around in AMD's future designs.
These processors aren't for gaming.
There's nothing they need to do.
Take a high core count server grade Xeon and see how it performs in gaming vs. an 8700k.
I just got done reading a review filled with workstation grade benchmarks and that's what people are looking at when considering a $1800 processor. If you're thinking of gaming on this... I don't know what to say.
 

french toast

Senior member
Feb 22, 2017
988
825
136
They're not staggering. They're inline with Intel's mesh uncore #'s and understood from an industry wide perspective to be what you'd expect. Intel's mesh has just about the same consumption. You want core scaling? This is what it costs. Cores don't connect via magic pixie dust. They require pretty active high speed interconnects that burn power.
Uncore%207980XE_575px.png


AMD has a major highway. Intel has tons of through streets. Both consume almost the same amount of power.
IF%20Power%202950X_575px.png

IF%20Power%202990WX_575px.png


Compare the various thread counts TOTAL Power Utilization between AMD and Intel. They're the same. You expect the highway to consume more power. It alleviates per junction (core<->interconnect) power utilization. So, It is what you can expect.. More consumption on AMD's highway (Infinity Fabric) and less power consumption at the cores. This is basic multi-core computer architecture taught in college. Both Intel and AMD's approach have limits. It's a matter of tradeoffs which unironically sum to the same power envelopes albeit one cost way less to manufacture.

From a high level, should the U.S have been connected with a bunch of through streets w/ no major highways? Of course not, traffic flow would be a nightmare. The same goes for cores. You want a highway. Everybody needs to eventually make highways (this includes Intel). Highways take a lot to construct and are pretty wasteful when they're not in use. Those interconnects have to be hot at all times. You can't play power games w/ it.

Btw, anandtech needs to clarify how they got their #'s :


And here is the confirmation of shenanigans :


And even more confusion .. Is it IF or uncore?


Come on guys.. what is this? So, essentially my guess is that they're blindly doing some backhand calculation of subtracting core power utilization from total power utilization and THE FIGURE THEYRE REFERRING TO IS UNCORE POWER UTILIZATION for both AMD/INTEL. In this case, the numbers
are just fine for AMD because it has far more I/O in its UNCORE.

Reviewers need to start detailing the methodology behind arriving at these unofficial #'s.
I'm growing tired of the IPC (Instruction per clock) like memes.
Ok, assuming AT did proper testing of the connections and not using whole uncore...
Now IF is worse than intel mesh...16 core consume slighly more power than intel 18 core at max...but the real story is MINIMUM connection power consumption...it is considerably more on AMD IF than on intel mesh..and nearly an order of magnitude worse when comparing intel ring bus to pinnacle ridge IF...(baseline).
Power consumption looks reasonably equivalent between mesh and IF when using AMD 2 dice Vs 18 monolithic intel die...but connecting 4 dice is a massive amount of power for either Epyc or 2990wx...using that much percentage of power is not ideal.

I'm saying using IF to connect multiple small 4 core CCX for desktop is a bad situation long term, even worse when using multiple IF links to connect 4 of those dice together for threadripper.
They would be better off moving to a new more efficient topology for a larger CCX, better connections between CCX, finally some kind of active interposer with advanced topology for connecting dice.
Most people seem to think ..large ring bus CCX + active interposer+butter donut is a better long term solution.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
This would suggest that maybe an active interposer is the way forward? Mount the CCX’s to the silicon interposer with the minimum routing capabilities required for inter chip communications and the butter donut topology and maybe both power and latency decrease.

They're converge multiple CCXs onto a local lighter weight and more power efficient interconnect and then retain the IF for connections. So, 16 cores on a localized interconnect and then IF for broader connections. So, 8-8-8-8 [32] .. 16-16-16-16 [64]. You can keep doing this up to 128.

Towns/cities/states/countries.
The same way we connect such topologies.

A series of through roads and highways.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
Ok, assuming AT did proper testing of the connections and not using whole uncore...
Now IF is worse than intel mesh...16 core consume slighly more power than intel 18 core at max...but the real story is MINIMUM connection power consumption...it is considerably more on AMD IF than on intel mesh..and nearly an order of magnitude worse when comparing intel ring bus to pinnacle ridge IF...(baseline).
Power consumption looks reasonably equivalent between mesh and IF when using AMD 2 dice Vs 18 monolithic intel die...but connecting 4 dice is a massive amount of power for either Epyc or 2990wx...using that much percentage of power is not ideal.

I'm saying using IF to connect multiple small 4 core CCX for desktop is a bad situation long term, even worse when using multiple IF links to connect 4 of those dice together for threadripper.
They would be better off moving to a new more efficient topology for a larger CCX, better connections between CCX, finally some kind of active interposer with advanced topology for connecting dice.
Most people seem to think ..large ring bus CCX + active interposer+butter donut is a better long term solution.
That's not a valid assumption because there is no access given to third parties by either Intel or AMD to make such measurements. So, more importantly, in an age of misinformation (see the bogus IPC meme floating around every review). I'd like to know how exactly the article writer came to arrive at IF power utilization both within a CCX and outside of it. Just using common sense and the article's own confusing language, clearly this is a rough number arrived at by back hand calculations from core power utilization vs. total envelope whereby you arrive at uncore (total chip power consumption - per core power = uncore). It's not hard to just state this. Everyone likes being edgy and having that detail someone else doesn't and then you start forming patently false memes like Instruction per clock measures.

As far as the other commentary, this is the problem when misinformation occurs. Multi-core chip architecture 101 : Make highways and step down traffic or meshify w/ routing. You're going to arrive at the same problems that will result in cross-over. The mesh will eventually need highways and the highway approach will eventually need mesh. Both Intel and AMD have it and use it in different ways. To start, AMD went with highways and mesh-lite. Intel went full-mesh. Neither approach is a final product. Both run at just about the same efficiency. How so? Simple, each core in Intel's architecture needs a beefy junction interconnect to the uncore mesh network. So, its cores will consume more power. AMD's uses highways and stepped down interconnects so its cores will consume less power.

So, how do you compare Intel/AMD ((properly))?
You compare total power consumption at the core count level.
K.I.S.S

What do you see?
Just about the same power consumption from both Intel/AMD

Where the memes/misinformation begin is when people try to dig into details they don't understand nor have official data for (a lot of micro-architectural details). Don't do this. It leads to confusion and incorrect conclusions.
 
Last edited:

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Edit: Its I was right on a couple of points in my 2990WX builders thread. First, that MSI motherboard I ordered appears to be the one everyone is using, and second, overclocked to 4 ghz is common, and draws 500-600 watts, so my custom water loop is required.

You definitely want that 1KW PSU if you are going to overclock:
https://www.techspot.com/review/1678-amd-ryzen-threadripper-2990wx-2950x/page9.html

Using the Asus PBO method the idle draw is roughly halved down to 95 watts but even so the load consumption still hits an eye watering 694 watts, in this test. I say in this test because I obscured total power draw hitting 780 watts in Cinebench and 848 watts for the fixed voltage overclock.

Yikes.
 

french toast

Senior member
Feb 22, 2017
988
825
136
That's not a valid assumption because there is no access given to third parties by either Intel or AMD to make such measurements. So, more importantly, in an age of misinformation (see the bogus IPC meme floating around every review). I'd like to know how exactly the article writer came to arrive at IF power utilization both within a CCX and outside of it. Just using common sense and the article's own confusing language, clearly this is a rough number arrived at by back hand calculations from core power utilization & total envelope more whereby you arrive at uncore (total chip power consumption - per core power = uncore). It's not hard to just state this. Everyone likes being edgy and having that detail someone else doesn't and then you start forming patently false memes like Instruction per clock measures.
I don't know what you mean by bogus IPC measurements...perhaps you mean trying to derive IPC by multicore benchmarks or something.
Anyway, you are right, we need some confirmation on how anandtech came to those numbers...even if it is uncore and not the connections themselves, it is still significantly more than intel... primarily at low core utilisation, which affects desktop more than enterprise chips.
AMD uncore consumes vastly more power than intel's.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
I don't know what you mean by bogus IPC measurements...perhaps you mean trying to derive IPC by multicore benchmarks or something.
Anyway, you are right, we need some confirmation on how anandtech came to those numbers...even if it is uncore and not the connections themselves, it is still significantly more than intel... primarily at low core utilisation, which affects desktop more than enterprise chips.
AMD uncore consumes vastly more power than intel's.

To get similar numbers for Intel, the power consumption of at least VCCIO and VCCSA planes should be included.
 

french toast

Senior member
Feb 22, 2017
988
825
136
To get similar numbers for Intel, the power consumption of at least VCCIO and VCCSA planes should be included.
Spot on, Yea we need some clarification on the testing methodology, what are those power planes used for? How is that different to AMD?
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
the moar coarz war...

the reviews told me as I understand them what I thought- we cannot use except special cases (distributed computing, or the benchmark runners e stuff guys) more than 16C seriously

for me the table still stands

1. 10-12C ~5GHz
2. 8C 5GHz (i9-9900K)
3. 16C ~4,5GHz
4. everything else

and that power of the oced 2990X- wow, I need a personal mini nuclear plant on the balcony...just when I thought the 7980XE is a power hog...

Lets wait for the winner builds with 2990X and 2x1180Ti with 2kW PSU
 
Last edited:

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,432
136
Last edited:
  • Like
Reactions: Drazick and ub4ty

ub4ty

Senior member
Jun 21, 2017
749
898
96
You definitely want that 1KW PSU if you are going to overclock:
https://www.techspot.com/review/1678-amd-ryzen-threadripper-2990wx-2950x/page9.html



Yikes.
This is why enterprise grade processors are clocked so low. If they want to enter into the higher realms of professional computing, the enthusiast is going to learn to calm down and not play with their processors. They're optimized out of the factory like modern day cars. No more shad tree mechanic foolery under the hood. You can still do it, you're just going to create an inferno of inefficiency.
 

dnavas

Senior member
Feb 25, 2017
355
190
116
Spot on, Yea we need some clarification on the testing methodology, what are those power planes used for? How is that different to AMD?

Would changing the memory speeds change the IF speed? Would making a chart from 2400 - 3600 yield anything interesting?
 

BigDaveX

Senior member
Jun 12, 2014
440
216
116
These processors aren't for gaming.
There's nothing they need to do.
Take a high core count server grade Xeon and see how it performs in gaming vs. an 8700k.
I just got done reading a review filled with workstation grade benchmarks and that's what people are looking at when considering a $1800 processor. If you're thinking of gaming on this... I don't know what to say.
Yes, this is a workstation CPU. Doesn't mean people don't occasionally like to game on their workstations - I speak as someone who bought a Haswell-E at a time when Skylake was the best pure gaming chip on the market - and from that standpoint, the 7980XE and 2950X both hold up reasonably well compared to their desktop counterparts, but the 2990WX collapses to as little as half the performance of either a Ryzen 7 or a 2-die Threadripper.

Besides, I wasn't commenting on the 2990WX's usefulness as a gaming CPU per se, I was pointing out that the overheads from NUMA are still very much there (albeit less than they were a decade ago), and can still hurt performance in single/low-thread situations.
 
  • Like
Reactions: pcp7

ub4ty

Senior member
Jun 21, 2017
749
898
96
I don't know what you mean by bogus IPC measurements...perhaps you mean trying to derive IPC by multicore benchmarks or something.
Anyway, you are right, we need some confirmation on how anandtech came to those numbers...even if it is uncore and not the connections themselves, it is still significantly more than intel... primarily at low core utilisation, which affects desktop more than enterprise chips.
AMD uncore consumes vastly more power than intel's.
I mean what anyone with a degree in computer engineering means when they refer to it which is why you never hear it referred to outside of very strict official circles. Instruction Per Clock/cycle or whatever bogus reference is made in reviews reflects a fundamental misunderstanding of computer architecture.

https://en.wikipedia.org/wiki/Instructions_per_cycle
The number of instructions executed per clock is not a constant for a given processor; it depends on how the particular software being run interacts with the processor, and indeed the entire machine, particularly the memory hierarchy

When Real IPC #'s are referenced, they're official #'s ... and are statically referenced to varying strict tests. They refer to micro-architectural details that your average person without a pretty thorough degree has no use for.

IPC is used to analyze and evaluate instruction pipeline details. So, referring to this measure incorrectly has absolutely zero validity in any mainstream reviews. IPC as referred to in most if not all mainstream reviews is bogus. They're taking something as simple and readable as how long a particular task takes on one processor vs another and reverse engineering some bogus and flawed reference to IPC.

I
...even if it is uncore and not the connections themselves, it is still significantly more than intel... primarily at low core utilisation, which affects desktop more than enterprise chips.
I explained this in one of my posts. The problem and I'll be frank is that people are speaking beyond their depth of knowledge and getting into micro-architectural details/analysis they don't understand. This leads to misinformation and ill-formed opinions. For this reason, even possessing a graduate degree in this area, I like to K.I.S.S. If you want to compare intel to AMD, compare total package power utilization per consumption per core and power efficiency for a particular workload.

Mainstream reviewers aren't equipped to go into micro-architectural performance analysis and Intel nor AMD publishes significant details about it because its where all of their IP centers. So, by and large, what you see in mainstream outlets, regarding micro-architecture analysis, beyond official information, is quite bogus.
 
  • Like
Reactions: SirDinadan

ub4ty

Senior member
Jun 21, 2017
749
898
96
the moar coarz war...

the reviews told me as I understand them what I thought- we cannot use except special cases (distributed computing, or the benchmark runners e stuff guys) more than 16C seriously

for me the table still stands

1. 10-12C ~5GHz
2. 8C 5GHz (i9-9900K)
3. 16C ~4,5GHz
4. everything else

and that power of the oced 2990X- wow, I need a personal mini nuclear plant on the balcony...just when I thought the 7980XE is a power hog...

Lets wait for the winner builds with 2990X and 2x1180Ti :) 2kW PSU :)
Mark's got this covered. Also, if people aren't aware, a large number of residential breakers would be tripped by such loads. Big boy league indeed requires big boy power. Mark has this covered from what I've inquired.

15 AMPS x 120 volts = 1800 watts
20 AMPS x 120 volts = 2400 watts.

Next is ambient air cooling because you can't dump that kind of heat into a room without turning it into an inferno. I have temp-controlled dedicated in-room A/Cs for my various computer installations. Got a beast of a computer or computers already drawing 1kw+? You're going to need another breaker for the A/C. Power/cooling isn't a joke. It's one of the biggest expenditures for data centers. If people want big boy performance, they're going to have to pay for big boy power/cooling. This is why I only power on machines when I'm actually using them and have dedicated machines for various tasks. I don't game or do anything but work related tasks on my high core count machines. My normal use machines are dual and four core including my gaming rig.
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,248
136
I'm not in the market....But the benchmarks are all over the place.

What's with low resolution gaming results on workstation class cpus? Seems pointless.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
Yes, this is a workstation CPU. Doesn't mean people don't occasionally like to game on their workstations
If you have $1800 to blow on a 32 core processor, I'd assume this isn't your only machine and are a professional intending on professional highly multi-threaded workflows.

As such, I can't imagine many people who would legitimately game on such a machine. It's not a processor suited for it nor are Intel's processors and will suck down huge amounts of power if not fully utilized.

Most game engines aren't NUMA aware and don't scale to such core counts. So, it's a waste and you're going to get reduced performance. AMD has a toggle for 16 core for NUMA/Gaming mode. I read they didn't put one on 2990wx .. The clear message is : Hey dude.. it's not a gaming processor.. Don't game on it. We sell other processors for that.

Enthusiast gamers and others w/ no use case for such processors like to pretend they care about all around performance but at these levels, it's way outside their affordability. It is also quite foolish given the performance numbers. So, were finally at a distinguishing point between people who have a legitimate case for such high performance vs those that just want to do [my computer is bigger than yours]. The only thing that matters are serious workload performance on this processor. Need a gaming rig? Build one. A complete build would cost 1/2 as much as this processor alone. I'd assume a person with $1800 for a processor could afford or already has a gaming PC.

- I speak as someone who bought a Haswell-E at a time when Skylake was the best pure gaming chip on the market - and from that standpoint, the 7980XE and 2950X both hold up reasonably well compared to their desktop counterparts, but the 2990WX collapses to as little as half the performance of either a Ryzen 7 or a 2-die Threadripper.
Yep, the 2990WX's details are what I expected... Big Diminishing returns :
- 2 Dies have no direct I/O connection
- Still need Infinity fabric blazing to connect them in full mesh config.
Benefits are :
- Higher clocks than EPYC

So, the consumers of this product will have to think quite carefully and be very knowing of their workload's particular performance. From 8 core to 16 core, there was almost perfect scaling, 100% increase across the board. From 16-32 on Threadripper, you're getting about 50/60% increase. For some, this will be worth it .. but its : niche/niche

Besides, I wasn't commenting on the 2990WX's usefulness as a gaming CPU per se, I was pointing out that the overheads from NUMA are still very much there (albeit less than they were a decade ago), and can still hurt performance in single/low-thread situations.
That's a given and expected in this realm of computing. It seems like this is a new paradigm for enthusiast but it isn't for people with workloads intended for these processors. My only comment was : You don't game on these processors. Build another more efficient/suited rig. If you have the money and valid use case for such a processor, you should have enough or already have a gaming rig for entertainment.
 
  • Like
Reactions: lightmanek

french toast

Senior member
Feb 22, 2017
988
825
136
I mean what anyone with a degree in computer engineering means when they refer to it which is why you never hear it referred to outside of very strict official circles. Instruction Per Clock/cycle or whatever bogus reference is made in reviews reflects a fundamental misunderstanding of computer architecture.

https://en.wikipedia.org/wiki/Instructions_per_cycle
The number of instructions executed per clock is not a constant for a given processor; it depends on how the particular software being run interacts with the processor, and indeed the entire machine, particularly the memory hierarchy

When Real IPC #'s are referenced, they're official #'s ... and are statically referenced to varying strict tests. They refer to micro-architectural details that your average person without a pretty thorough degree has no use for.

IPC is used to analyze and evaluate instruction pipeline details. So, referring to this measure incorrectly has absolutely zero validity in any mainstream reviews. IPC as referred to in most if not all mainstream reviews is bogus. They're taking something as simple and readable as how long a particular task takes on one processor vs another and reverse engineering some bogus and flawed reference to IPC.


I explained this in one of my posts. The problem and I'll be frank is that people are speaking beyond their depth of knowledge and getting into micro-architectural details/analysis they don't understand. This leads to misinformation and ill-formed opinions. For this reason, even possessing a graduate degree in this area, I like to K.I.S.S. If you want to compare intel to AMD, compare total package power utilization per consumption per core and power efficiency for a particular workload.

Mainstream reviewers aren't equipped to go into micro-architectural performance analysis and Intel nor AMD publishes significant details about it because its where all of their IP centers. So, by and large, what you see in mainstream outlets, regarding micro-architecture analysis, beyond official information, is quite bogus.
Look, if you want to be exact then yes everyone with a brain (on forums like these) understands IPC is dependant on the software and instructions used, different architecture has its strengths and weakness in different software, even the software has different optimisations for different manufacturers, dependant on favouritism and/or funding.

But for getting a general idea of a processors IPC in desktop workloads...to compare against previous or different processor's..then you certainly can go read a decent review and get a solid idea.
You can read an anandtech review for instance and know skylake is between 6-12% faster per clock than Zen.
To say all of it is complete bogus and worthless is in itself quite bogus.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,432
136
If you have $1800 to blow on a 32 core processor, I'd assume this isn't your only machine and are a professional intending on professional highly multi-threaded workflows.

As such, I can't imagine many people who would legitimately game on such a machine. It's not a processor suited for it nor are Intel's processors and will suck down huge amounts of power if not fully utilized.

Most game engines aren't NUMA aware and don't scale to such core counts. So, it's a waste and you're going to get reduced performance. AMD has a toggle for 16 core for NUMA/Gaming mode. I read they didn't put one on 2990wx .. The clear message is : Hey dude.. it's not a gaming processor.. Don't game on it. We sell other processors for that.

Enthusiast gamers and others w/ no use case for such processors like to pretend they care about all around performance but at these levels, it's way outside their affordability. It is also quite foolish given the performance numbers. So, were finally at a distinguishing point between people who have a legitimate case for such high performance vs those that just want to do [my computer is bigger than yours]. The only thing that matters are serious workload performance on this processor. Need a gaming rig? Build one. A complete build would cost 1/2 as much as this processor alone. I'd assume a person with $1800 for a processor could afford or already has a gaming PC.


Yep, the 2990WX's details are what I expected... Big Diminishing returns :
- 2 Dies have no direct I/O connection
- Still need Infinity fabric blazing to connect them in full mesh config.
Benefits are :
- Higher clocks than EPYC

So, the consumers of this product will have to think quite carefully and be very knowing of their workload's particular performance. From 8 core to 16 core, there was almost perfect scaling, 100% increase across the board. From 16-32 on Threadripper, you're getting about 50/60% increase. For some, this will be worth it .. but its : niche/niche


That's a given and expected in this realm of computing. It seems like this is a new paradigm for enthusiast but it isn't for people with workloads intended for these processors. My only comment was : You don't game on these processors. Build another more efficient/suited rig. If you have the money and valid use case for such a processor, you should have enough or already have a gaming rig for entertainment.
You forgot one of the biggest reasons to get this instead of EPYC 7601. PRICE ! $4600 vs $1800, almost one third the price.
But yes, the other reason is clocks and OC ability. I may not do 4 ghz due to power draw and heat, but 2.2 ghz ? vs at least 3.4 ?
 

french toast

Senior member
Feb 22, 2017
988
825
136
Would changing the memory speeds change the IF speed? Would making a chart from 2400 - 3600 yield anything interesting?
At least on desktop yes, the fabric is linked to the IMC in a 2:1 ratio if i remember, not sure about Epyc.
Would making a chart be interesting? For power consumption? It would be interesting but I'm not sure you can derive anything conclusive from it, without proper testing how would isolate the core+ uncore from the IF power consumption?
 
  • Like
Reactions: ub4ty