AMD TR 2 32 core? Too spendy? Updated with strong potential price leak!

juanrga said:
You cannot find one still because Intel has not still announced CascadeLake-X chips.

Sorry, but comparing the expected pricing of an unreleased 32 core Threadripper to the pricing of a 8P 28 core Xeon continues being biased and unfair.
Click to expand...

The 28c Xeon is also one of the most complex dies in the history of dies, and any 28c consumer processor from Intel is going to be balls expensive unless Intel wants to take a loss just to top the benchmark charts. The big Xeons have been getting more and more expensive each generation, with lower and lower yields - the 18c Haswell part was never supposed to exist and the full Broadwell-EP (24c) die only existed in the form of a $7K 8P flagship.
The 2990X is a huge win for AMD's MCM + Infinity Fabric strategy, it doesn't cost much more than four Ryzen 2700X's to make. It's also easier to find 4 8c dies that clock well than one 28c die were all the cores can achieve high clock speeds.
 
KazeoHin said:
So, because AMD's 32 core is ~$1800, best to get intel? What does Intel's competing product cost? how much does Intel's product that matches this CPU's performance cost?

That's like saying "Welp, the new corvette is going to cost $120K, Guess I'm going Lamborghini now..." That makes no sense...
Click to expand...

That mindset is why I bought a ch6 instead of an el cheapo b350 mobo that went up in price.
 
bwang said:
The 28c Xeon is also one of the most complex dies in the history of dies, and any 28c consumer processor from Intel is going to be balls expensive unless Intel wants to take a loss just to top the benchmark charts. The big Xeons have been getting more and more expensive each generation, with lower and lower yields - the 18c Haswell part was never supposed to exist and the full Broadwell-EP (24c) die only existed in the form of a $7K 8P flagship.
The 2990X is a huge win for AMD's MCM + Infinity Fabric strategy, it doesn't cost much more than four Ryzen 2700X's to make. It's also easier to find 4 8c dies that clock well than one 28c die were all the cores can achieve high clock speeds.
Click to expand...

Multdie approach is clearly cheaper. The counterpart of the MCM approach are performance and power penalties.
 
notarat said:
"Pro-Intel / Anti-AMD Just FYI"

Can I get that post with a grain of salt?
Click to expand...

Dude I normally disagree with Jaun but damn man .... how is jaun not right. It's like you comparing a family car to a construction machine. It's no different. Until intel is afforded a chance to combat AMDs 32 core the AMD really has nothing in the hedt segment to be compared too ..... yet.

You shouldn't, though your free too, whip out a top tier Xeon server chip and be like ... this is the competition because we all know good and damn well Intel is building or maybe it has alreasy built the direct HEDT competition and we must sit back and wait a little bit and wait for official announcement and price tag.
 
Last edited:
bwang said:
The 28c Xeon is also one of the most complex dies in the history of dies, and any 28c consumer processor from Intel is going to be balls expensive unless Intel wants to take a loss just to top the benchmark charts. The big Xeons have been getting more and more expensive each generation, with lower and lower yields - the 18c Haswell part was never supposed to exist and the full Broadwell-EP (24c) die only existed in the form of a $7K 8P flagship.
The 2990X is a huge win for AMD's MCM + Infinity Fabric strategy, it doesn't cost much more than four Ryzen 2700X's to make. It's also easier to find 4 8c dies that clock well than one 28c die were all the cores can achieve high clock speeds.
Click to expand...
They could just sell a chilled water unit with each cpu?
 
tangoseal said:
Dude I normally disagree with Jaun but damn man .... how is jaun not right. It's like you comparing a family car to a construction machine. It's no different. Until intel is afforded a chance to combat AMDs 32 core the AMD really has nothing in the hedt segment to be compared too ..... yet.

You shouldn't, though your free too, whip out a top tier Xeon server chip and be like ... this is the competition because we all know good and damn well Intel is building or maybe it has alreasy built the direct HEDT competition and we must sit back and wait a little bit and wait for official announcement and price tag.
Click to expand...

My main question is this:

What does Intel have to act as competition to the 2990X? Now, the average person would say "Nothing", but Let us see how the Intel professional astroturfers shills answer that question. They are quick to say "CPU A is not meant to compete with the Ryzen chip" But they offer NO alternative.
 
KazeoHin said:
My main question is this:

What does Intel have to act as competition to the 2990X? Now, the average person would say "Nothing", but Let us see how the Intel professional astroturfers shills answer that question. They are quick to say "CPU A is not meant to compete with the Ryzen chip" But they offer NO alternative.
Click to expand...

We know the answer for months. Intel is updating the HEDT line with new 22-core chips and releasing a new ultra-HEDT line with 28-core chips.

intel-consumer-roadmap-2018-2019-leak-1030x588-png.png
 
Last edited:
juanrga said:
We have known the answer to this question for months. Intel is updating the HEDT line with new 22-core chips and releasing a new ultra-HEDT line with 28-core chips.

View attachment 93334
Click to expand...
Now these will run at a high ghz without chilled water? What about speed on air? Are they making the die big like AMD's to spread the heat?
 
Hagrid said:
Now these will run at a high ghz without chilled water? What about speed on air? Are they making the die big like AMD's to spread the heat?
Click to expand...

(i) There was two 28-core systems during Intel conference. One was pushed to 5GHz with a chiller. The other was pushed above 4GHz (est.) with liquid cooler.

(ii) Both demos used a modified Xeon to simulate the forthcoming HEDT chip. Recall that Derhauer also had to use a chiller on an EPYC chip to simulate the performance of the forthcoming Threadripper chip.

(iii) The dies are same size than Xeon dies so far as I know. Multidie and higher core counts aren't coming until Cascade Lake.

(iv) I don't know the stock clocks. The 28-core seems to have a 2.7GHz base clock plus all-core 3.5GHz plus 5GHz max boost.
 
Last edited:
juanrga said:
Recall that Derhauer also had to use a chiller on an EPYC chip to simulate the performance of the forthcoming Threadripper chip.
Click to expand...
That was due to der8auer's overclocking setup. The Epyc chips were not binned for such high frequencies, so drastic measures had to be taken.

AMD's Computex demo cooled their 24 core and 32 core TR2000 on air. And at least the 24 core performed exactly as you would expect a 24 core TR to perform. I wrote about it earlier:
chithanh said:
This means overall, in this test setup, the 24 core TR2 performs roughly 50-55% faster than a 1950X which would be very much in line with the increase in core count, if we assume identical clocks.
Click to expand...
 
chithanh said:
That was due to der8auer's overclocking setup. The Epyc chips were not binned for such high frequencies, so drastic measures had to be taken.
Click to expand...

Neither the Xeon chips used for the HEDT demo were binned for such high frequencies. The Xeon used for the ~4GHz and 5GHz demos has max boost of 3.8GHz. Chillers were used on both EPYC and Xeon for the same reason.
 
Hagrid said:
Now these will run at a high ghz without chilled water? What about speed on air? Are they making the die big like AMD's to spread the heat?
Click to expand...

Not sure about Intel but AMD appears to clock at 4.0ghz all 32 cores which is absolutely blazing awesome. However Intel has been known to always have the clock speed science down pat for the most part. I used to own a 16 core TR but now x299. At the time I couldn't justify the cores but now my needs have changed and I could totally use the cores. I was very impressed with the aggregate performance of 16 physical cores plus 16 SMT threads as well. But I felt like it was weak in gaming. However, looking back I had a corrupted OS which I didn't realize and blamed AMD for which was the 2nd reason I sold the AMD main one being I didn't need so many cores. Thus I went with a delidded 7820x which is posting 205 single thread cinebench.

So I want to wait and see what Intel charges and how the 22core benches before deciding if Inwant to stay x299s questionable ass platform or go back to team red.

Not to mention I really want to use Store MI tech in a way. Tired of having so many damn drive letters.
 
juanrga said:
(i) There was two 28-core systems during Intel conference. One was pushed to 5GHz with a chiller. The other was pushed above 4GHz (est.) with liquid cooler.

(ii) Both demos used a modified Xeon to simulate the forthcoming HEDT chip. Recall that Derhauer also had to use a chiller on an EPYC chip to simulate the performance of the forthcoming Threadripper chip.

(iii) The dies are same size than Xeon dies so far as I know. Multidie and higher core counts aren't coming until Cascade Lake.

(iv) I don't know the stock clocks. The 28-core seems to have a 2.7GHz base clock plus all-core 3.5GHz plus 5GHz max boost.
Click to expand...

(i) first I've ever heard of this 4GHz on water chip. Source?

(ii) DerBauer is an end-user with access to retail binned-chips. Intel is both designer and manufacturer of the chip with access to literally every tier of bin for their silicon. Are you saying they would be so irresponsible as to randomly grab a retail piece of silicon binned for 3.5GHz max boost and strap a chiller to it and just HOPE it works, or would you think that (if they could) they would try to bin for high frequency to find the best example of their upcoming chip?
 
tangoseal said:
You have to get that comparison to server shit out of here man. Those days are over. We are now in the age of high core desktop processors where Intel is about to drop 8 core as their highest tier regular desktop chip. And Amd is going to 10 and 16 for standard desktop chips. Dude we would still be on quad core Intel chips as the PREMO market had AMD never been successful with Ryzen and the number of people that were awakened to just how productive those 8 core chips were. Now Intel is struggling to get their 8 core I9's out as fast as possible.

This is a core war now.

Alas, as a disclaimer I am entitled to my opinion just as much as anyone reading this is entitled to ignore the shit out of me.
Click to expand...

Well, you certainly are an entitled one, eh? :D Opinions are like you know what, everyone has one and they all smell.
 
KazeoHin said:
And you are the shining beacon of hope against the biased and unfair...
Click to expand...

LOL! Thanks for the face splitting laugh. :) I am happy with my R7 1700 running at 3.8 Ghz and will be for a few years. However, 32C/64T is a monster.
 
pgaster said:
Do not expect great performance by Threadripper in SolidWorks. Of course you said you have other uses for it as well, so that's good.
SolidWorks is multi-threaded, but because of how it has to recalculate things it only uses maybe 2 cores for model rotations and part rebuilds. If you go to the model tree and change something that was done very early in the part design, then SolidWorks has to rebuild the part one feature at a time. It can't rebuild all steps in parallel. More cores do not matter. IPC is king and AMD is behind on that for now.
It pains me to say that since I like AMD and have a Ryzen, but it's the truth.
Click to expand...

I have been thinking about this a lot lately and tbh I think I am just going in circles.

I had to re-purpose my desktop workstation I was using to a side business my brother and I have been building doing aerial 3d mapping and ortho stuff. Mostly for construction companies. Many of the jobs can take 28-30+ hours to render a single job. It has been getting us behind, so are contemplating paying for server time, but that would be expensive and really not ideal for a variety of reasons. The least of which being we can buy the servers for what it would cost for a few months of server usage.

We are running mostly on my 6800k 6c/12t system with 64gb memory. But I have noticed the loads are mostly 100% on the 6c with the other 6 threads sitting idle, and only getting work in a few stages of the process. I bumped the cpu up from 3.4 to 4.0 since it's on a decent corsair aio, but it only made a marginal difference. The memory usage bounces anywhere from 10gb to 16gb on the jobs.

My uncertainty comes from my somewhat limited understanding of the difference between Intel and AMD's multi threading. It's my understanding that "basically" if you thought about it from an assembly line perspective. Intel has cores that will process up to 100 units (just as an example) and if 70 units were it's work load, and the next load was < 30 units that cores HT would pick it up. But if the next load was 40 units, the HT would stay still and the load would wait for the next cpu cycle.

On AMD side, it's cores can only do 50 units (again just as an example) but it can process up to 100 units by using it's "second" core, but if the load was say 110 units, it would only process 50 and the on the next cycle both cores would work to process the next 60.

At least that is my understanding.

So that is where I am at. Would a 32 core / 64 thread $1500 AMD cpu be better for this purpose than a 16 core $1400 Intel i9 cpu? I just don't know. I am waiting for some answers from the creators of the render engine itself, but responses are basically 1 reply every 10-12 days so... I did get out of them that the software will scale "efficiently" to 24 cores. That is something I guess. If I do end up with TR2, I could do two 12 core nodes and an 8 core node.

Also, why can you get a 16c/32t 3.4ghz threadripper for $770, but the 16core/32t 2.4ghz epyc is $840?
 
Is there anything stopping them from using a single channel interface per core? They all have a controller. The pin to core isn't far. It's just an interposer.
 
KarsusTG said:
Also, why can you get a 16c/32t 3.4ghz threadripper for $770, but the 16core/32t 2.4ghz epyc is $840?
Click to expand...
It's a server chip, may be higher binned (though not clocked as high, for power/thermal reasons probably), and most importantly has many more pcie lanes and more memory channels.
 
schmide said:
Is there anything stopping them from using a single channel interface per core? They all have a controller. The pin to core isn't far. It's just an interposer.
Click to expand...

I think that would break backwards compatibility with current mobos.

You'd also get a performance penality with lightly threaded applications. If you really need good bandwidth/latency on all the cores, well, that's what EPYC is for.
 
I thought about it and you can share address lines as long as you gate the strobe, but that would require atomic access and that basically blows any routing advantage it would give. It was dumb.
 
KazeoHin said:
(i) first I've ever heard of this 4GHz on water chip. Source?

(ii) DerBauer is an end-user with access to retail binned-chips. Intel is both designer and manufacturer of the chip with access to literally every tier of bin for their silicon. Are you saying they would be so irresponsible as to randomly grab a retail piece of silicon binned for 3.5GHz max boost and strap a chiller to it and just HOPE it works, or would you think that (if they could) they would try to bin for high frequency to find the best example of their upcoming chip?
Click to expand...

(i) https://videocardz.com/newz/asus-dominus-spotted-with-28-core-cascade-lake-x-cpu
https://techgage.com/news/a-closer-look-at-the-intel-28-core-cpu-the-gigabyte-system-running-it/

(ii) All EPYC and Xeons Platinum chips are binned for efficiency. So both der8auer and Intel used a chiller to overclock the server chips and simulate the performance of forthcoming HEDT chips. Not a mystery here. The HEDT chips don't require chillers. Also both der8auer and Intel did run the chips on modified server boards, but the forthcoming HEDT chips will work in corresponding HEDT boards. Again there is no mystery.
 
juanrga said:
Neither the Xeon chips used for the HEDT demo were binned for such high frequencies. The Xeon used for the ~4GHz and 5GHz demos has max boost of 3.8GHz. Chillers were used on both EPYC and Xeon for the same reason.
Click to expand...
KazeoHin said:
(ii) DerBauer is an end-user with access to retail binned-chips. Intel is both designer and manufacturer of the chip with access to literally every tier of bin for their silicon. Are you saying they would be so irresponsible as to randomly grab a retail piece of silicon binned for 3.5GHz max boost and strap a chiller to it and just HOPE it works, or would you think that (if they could) they would try to bin for high frequency to find the best example of their upcoming chip?
Click to expand...
AMD has shown at Computex that the chiller is unnecessary. The TR2 were cooled with Wraithripper air coolers. In contrast, der8auer's Epyc OC test is only meant to represent the performance of TR2, not the temps / cooling / power draw.

AMD had access to chips that were capable of running at their advertised frequency on air, while Intel had not. This means only one company already had a viable retail product to show.
 
chithanh said:
AMD has shown at Computex that the chiller is unnecessary. The TR2 were cooled with Wraithripper air coolers. In contrast, der8auer's Epyc OC test is only meant to represent the performance of TR2, not the temps / cooling / power draw.
Click to expand...

Exactly! And the same happened with the Intel demo at Computex. Intel used an OC Xeon only to represent the performance of forthcoming 28-core HEDT chip, not the temps / cooling / power draw of the HEDT chip.

chithanh said:
AMD had access to chips that were capable of running at their advertised frequency on air, while Intel had not. This means only one company already had a viable retail product to show.
Click to expand...

AMD HEDT chip launches in Q3. Intel HEDT chip launches in Q4. It is evident what company has first access to the final product.
 
Nope, the Intel demo was something which had nothing to do with the final chip (5 GHz all core boost on air? Keep dreaming)
juanrga said:
AMD HEDT chip launches in Q3. Intel HEDT chip launches in Q4. It is evident what company has first access to the final product.
Click to expand...
What? Intel SKL-XSP XCC was launched in mid-2017. That is much longer than Pinnacle Ridge was around (early 2018).

So Intel takes much longer to turn the XCC die into a usable HEDT SKU which is quite telling. In addition, despite Intel having more than half a year head start, they were still unable to bin dies that can reach advertised speeds on air. Not good.
 
Last edited:
  • Like
Reactions: N4CR
like this
Hagrid said:
Oh, I guess all the news is wrong that Intel showcased a 5ghz 28 core product. Dam news!
Click to expand...


Where it says "5 GHz all core boost on air"? If you have some other misunderstanding about SKL-A check the several threads opened about it, instead derailing this thread about 2nd gen Threadripper with silly remarks about SKL-A.
 
juanrga said:

Where it says "5 GHz all core boost on air"? If you have some other misunderstanding about SKL-A check the several threads opened about it, instead derailing this thread about 2nd gen Threadripper with silly remarks about SKL-A.
Click to expand...
Oh, so they plan on including a water chiller with the cpu. Good to know!
 
Awww c'mon! I was being nice! :p

But a 32-core TR at $1500-$1700 seems to be just about right for that level of HEDT chip. That would beat the pricing crap outta Intel for sure. I can't see that 28-core chip being anywhere near that price - prolly 1.5x - 1.8x the price.
 
Skillz said:
But how well will it run BOINC?
Click to expand...
Depends on how many times you can launch Boinc I guess :). All you need to do is have numbered directories from X to XX and assign appropriate number of cpu towards each client .
 
juanrga said:
Who said 5GHz all-core on air? Intel? Me? Someone else?
Click to expand...
It was you who wrote that the Computex demo was representing the performance of the 28 core chip:

juanrga said:
Intel used an OC Xeon only to represent the performance of forthcoming 28-core HEDT chip
Click to expand...
And Intel during Computex presentation emphasized that it was running 28 cores on 5 GHz during the Cinebench run (at 0:58:08).

In contrast, AMD demonstrated 24 and 32 core TR2 on air. Which means the AMD demo system is representative for the final retail product regarding both the performance and the initial/operational cost (assuming that the Wraithripper cooler isn't vastly better and more expensive than other TR4 coolers, I'm sure Kyle will tell us soon).
 
You must log in or register to reply here.
Back
Top