AMD Ryzen 4 Core rumoured to have Hyper Threading disabled.

[Drak3]

49 minutes ago, Simon771 said:

Trust me, I also don't trust AMD with those 2 videos about rendering.

We need much more than that to see if Ryzen is realy worth it. It's also kinda weird to me that clock on shown chip was 3,4GHz only ... Is that maximum? If those CPUs are capable of 4,0GHz and beyond, why didn't they show us that in rub it in Intel's face.

So yeah we are all Jon Snow's right now because we know nothing

 

Untill I see full details of CPUs and benchmarks from Cinebench, I won't concider buying it. But I do hope AMD will finally bring something good to the table.

3.4GHz stock for an octocore is better than 3.0GHz (5960X) or 3.2GHz (6900K). Both of those chips are capable of 4GHz under a capable cooler as well, and require more power.

[Sintezza]

13 minutes ago, Drak3 said:

3.4GHz stock for an octocore is better than 3.0GHz (5960X) or 3.2GHz (6900K). Both of those chips are capable of 4GHz under a capable cooler as well, and require more power.

We have a thread going over at L1 techs forum on which several users have replicated those render tests that AMD was showing at the event, with the semples they provided.

One very interesting score that got my interests was a 5960X at 2.9GHz,

finnished the job in 30.70 seconds in Mac OSX10.10.4.

So that is about 5 seconds faster as the Ryzen 8C / 16T at 3.4GHz.

[Drak3]

2 minutes ago, Sintezza said:

We have a thread going over at L1 techs forum on which several users have replicated those render tests that AMD was showing at the event, with the semples they provided.

One very interesting score that got my interests was a 5960X at 2.9GHz,

finnished the job in 30.70 seconds in Mac OSX10.10.4.

So that is about 5 seconds faster as the Ryzen 8C / 16T at 3.4GHz.

Does it take into account any OS optimizations?

 

I've ran benchmarks on the same physical system before, only to have different results due to different operating systems.

[Simon771]

15 minutes ago, Drak3 said:

3.4GHz stock for an octocore is better than 3.0GHz (5960X) or 3.2GHz (6900K). Both of those chips are capable of 4GHz under a capable cooler as well, and require more power.

But is the 3,4GHz realy base clock? WHat if taht's higher clock possible on this chip?

[Sintezza]

1 minute ago, Drak3 said:

Does it take into account any OS optimizations?

 

I've ran benchmarks on the same physical system before, only to have different results due to different operating systems.

Yes OS overhead obviously seems to play a significant role here.

We have also users that have done the same blender test on both Linux and Windows.

And on Linux the results were constantly significantly better arround the board.

[Drak3]

1 minute ago, Simon771 said:

But is the 3,4GHz realy base clock? WHat if taht's higher clock possible on this chip?

It seems that AMD got it up to 3.6GHz according to another thread.

 

But base clocks on Intel i7 Extreme chips aren't based on the highest clocks possible (if they were, we'd be seeing 4GHz 5960X's and 6950X's out of box). They're based on TDP, how much heat they put out and how much power it takes to run them. That's why you see custom water cooling loops often on HEDT systems.

2 minutes ago, Sintezza said:

Yes OS overhead obviously seems to play a significant role here.

We have also users that have done the same blender test on both Linux and Windows.

And on Linux the results were constantly significantly better arround the board.

Then I wouldn't really use those results. Too many variables.

Still good to know, I might dual boot my main machine at some point.

[Sintezza]

20 minutes ago, Drak3 said:

It seems that AMD got it up to 3.6GHz according to another thread.

 

But base clocks on Intel i7 Extreme chips aren't based on the highest clocks possible (if they were, we'd be seeing 4GHz 5960X's and 6950X's out of box). They're based on TDP, how much heat they put out and how much power it takes to run them. That's why you see custom water cooling loops often on HEDT systems.

Then I wouldn't really use those results. Too many variables.

Still good to know, I might dual boot my main machine at some point.

 

I also have a score from a 5930K at 4.5GHz Linux mint18.

30.82 seconds.

On Windows7 38.38 seconds.

 

As you can see a significant overhead for Windows based operating systems.

On Windows10 similar story for what its worth.

 

[Dabombinable]

4 hours ago, Senzelian said:

What dual core are you talking about?

Cause I'm sure a Skylake M3 destroys anything AMD currently hast to offer on the notebook market. Especially when you consider, that with a M3 battery lifes of up to 10 hours of useage are no problem. 

This dual core:

https://ark.intel.com/products/88198/Intel-Core-m3-6Y30-Processor-4M-Cache-up-to-2_20-GHz

[RagnarokDel]

On 12/01/2017 at 5:27 AM, Simon771 said:

Depends how you look at it.

4 Ryzen cores for 50$ doesn't seem like a bad deal to me.

riiiiiiiiiiiiiiiiiiiite.

[Trik'Stari]

A question I feel is worth asking:

 

Are there any disadvantages to having HyperThreading enabled? I would hope/assume it only makes a difference in programs that can use it, and that with programs it cannot, it just runs like a normal 4c 4t CPU.

 

Or is there some kind of performance loss on a HyperThreading enabled CPU, when running programs that do not take advantage of HyperThreading.

 

@LinusTech Perhaps a TechQuickie is in order.....

[Curufinwe_wins]

38 minutes ago, Trik'Stari said:

A question I feel is worth asking:

 

Are there any disadvantages to having HyperThreading enabled? I would hope/assume it only makes a difference in programs that can use it, and that with programs it cannot, it just runs like a normal 4c 4t CPU.

 

Or is there some kind of performance loss on a HyperThreading enabled CPU, when running programs that do not take advantage of HyperThreading.

 

@LinusTech Perhaps a TechQuickie is in order.....

In theory no, but in practice a poorly threaded code trying to run on 8 logical threads can preform worse a similar code running on 4 physical threads. In a lot of older codes it can actually make a pretty big difference (and when I was working at ANL the default recommendation was always one task per x86_64 physical core.)

[Trik'Stari]

5 minutes ago, Curufinwe_wins said:

In theory no, but in practice a poorly threaded code trying to run on 8 logical threads can preform worse a similar code running on 4 physical threads. In a lot of older codes it can actually make a pretty big difference (and when I was working at ANL the default recommendation was always one task per x86_64 physical core.)

Interesting.

 

Is there now way for the CPU or program to say "This program is not properly optimized for HT" and sort of.... I don't know, dynamically "turn HT off"? So that the 4 individual physical cores act as such?

 

I know nothing about CPU architecture and coding. But I would think, as advanced as tech has gotten, someone would have thought of this kind of situation. Or at least, would have set things up so that when said program runs default as "one task per physical X84_64 physical core", that the HT wouldn't come into play? AFAIK HyperThreading is sort of a software/firmware thing, and the physical makeup of the CPU is still essentially the same, or is this incorrect? (I.E, isn't the difference between a 4690k and a 4790k just HT being "turned on" for the 4790k? Or is there actual physical hardware missing from the 4690k that is specifically related to HyperThreading?)

[Curufinwe_wins]

8 minutes ago, Trik'Stari said:

Interesting.

 

Is there now way for the CPU or program to say "This program is not properly optimized for HT" and sort of.... I don't know, dynamically "turn HT off"? So that the 4 individual physical cores act as such?

 

I know nothing about CPU architecture and coding. But I would think, as advanced as tech has gotten, someone would have thought of this kind of situation. Or at least, would have set things up so that when said program runs default as "one task per physical X84_64 physical core", that the HT wouldn't come into play? AFAIK HyperThreading is sort of a software/firmware thing, and the physical makeup of the CPU is still essentially the same, or is this incorrect? (I.E, isn't the difference between a 4690k and a 4790k just HT being "turned on" for the 4790k? Or is there actual physical hardware missing from the 4690k that is specifically related to HyperThreading?)

Yea you can definately add flags explicitly for HT enabled chips and to allow specific workloads to use those logical cores where applicable, but most of the time the program just sees 8 threads in a 4c/8t and the cpu does the dynamic logic to try to process everything explicitly out of order and function efficiently. 

 

Thankfully outside of a number of legacy operations, most poorly threaded codes are perfectly happy to pound 4 threads or less and (and order from 0 up) and thus avoid hitting the same core for on two threads (on anything other than a dual core where the extra umpf might be wanted anyways). 

 

Additionally, even if the individual program doesn't gain much or anything from HT, HT does allow for easier handling of low priority background tasks while the main program is pounding the crap out of a cpu (which is part of the reason why an i5 and i7 might have very similar average fps, but the i7 often delivers a smoother experience with fewer  high frame time event).

 

 

[leadeater]

4 minutes ago, Trik'Stari said:

Interesting.

 

Is there now way for the CPU or program to say "This program is not properly optimized for HT" and sort of.... I don't know, dynamically "turn HT off"? So that the 4 individual physical cores act as such?

 

I know nothing about CPU architecture and coding. But I would think, as advanced as tech has gotten, someone would have thought of this kind of situation. Or at least, would have set things up so that when said program runs default as "one task per physical X84_64 physical core", that the HT wouldn't come into play? AFAIK HyperThreading is sort of a software/firmware thing, and the physical makeup of the CPU is still essentially the same, or is this incorrect? (I.E, isn't the difference between a 4690k and a 4790k just HT being "turned on" for the 4790k? Or is there actual physical hardware missing from the 4690k that is specifically related to HyperThreading?)

Hyper-threading is a physical thing in the CPU die, CPUs with it disabled still actually have it present but locked off. Basically this is why I think it is BS to have it disabled at all.

 

Operating systems and compilers have been HT aware for a long time so it's going to be a rare case to find anything that is going to have a negative impact. Much older complied programs (pre 2000's) or programs where they have gone very low level in the coding, not using advanced threading helpers in the language to give compiler hints, is the place where you will see issues.

 

Old versions of Microsoft SQL Server 2005 and all previous versions recommend to disable HT.

[Trik'Stari]

1 minute ago, Curufinwe_wins said:

Yea you can definately add flags explicitly for HT enabled chips and to allow specific workloads to use those logical cores where applicable, but most of the time the program just sees 8 cores in a 4c/8t and the cpu does the dynamic logic to try to process everything explicitly out of order and function efficiently. 

 

Thankfully outside of a number of legacy operations, most poorly threaded codes are perfectly happy to pound 4 threads or less and (and order from 0 up) and thus avoid hitting the same core for on two threads (on anything other than a dual core where the extra umpf might be wanted anyways). 

 

Additionally, even if the individual program doesn't gain much or anything from HT, HT does allow for easier handling of low priority background tasks while the main program is pounding the crap out of a cpu.

 

 

I find this interesting, as I've only had and used CPU's without HyperThreading except for the I3 I put in the office PC I built for my mom before she passed when she was still working. When it still had the SSD installed as the OS drive (instead of a WD Blue HDD) it ran great for everything she did.

 

I'm wondering if there is any situation where a game, or commonly used office program would actually perform better on a non HT CPU than on a HT enabled CPU. As you say, legacy programs are an issue (aren't they always?), but I can't help but wondering about more modern stuff. I would assume most programs that aren't HT optimized would simply see 4 cores and run as such, and the CPU would just accommodate the program as such (can a HT enabled core function as a "single" core?, or does it simply assign 4 logical threads as though they were physical cores?).

 

CPU's are so complex it's not even funny. I find it difficult to understand how we have created such complex machines.

[Trik'Stari]

3 minutes ago, leadeater said:

Hyper-threading is a physical thing in the CPU die, CPUs with it disabled still actually have it present but locked off. Basically this is why I think it is BS to have it disabled at all.

 

Operating systems and compilers have been HT aware for a long time so it's going to be a rare case to find anything that is going to have a negative impact. Much older complied programs (pre 2000's) or programs where they have gone very low level in the coding, not using advanced threading helpers in the language to give compiler hints, is the place where you will see issues.

 

Old versions of Microsoft SQL Server 2005 and all previous versions recommend to disable HT.

How is it locked off? Is it locked by physically removing something? Or is there something in the "firmware" (not sure what the proper term would be here, machine language?) that simply says "do not allow hyperthreading"? (something in code telling it to be disable, as opposed to something being physically removed that prevents it from being enabled)

[Curufinwe_wins]

8 minutes ago, Trik'Stari said:

How is it locked off? Is it locked by physically removing something? Or is there something in the "firmware" (not sure what the proper term would be here, machine language?) that simply says "do not allow hyperthreading"? (something in code telling it to be disable, as opposed to something being physically removed that prevents it from being enabled)

Intel laser disables cores and cache and memory that cutdown sku's don't use, so there is probably at very least a hardware flag saying to not allow HT.

[leadeater]

13 minutes ago, Trik'Stari said:

How is it locked off? Is it locked by physically removing something? Or is there something in the "firmware" (not sure what the proper term would be here, machine language?) that simply says "do not allow hyperthreading"? (something in code telling it to be disable, as opposed to something being physically removed that prevents it from being enabled)

It's disabled in the CPU firmware/microcode so when you install it in a motherboard it will not present HT as a feature of the CPU.

 

Edit:

3 minutes ago, Curufinwe_wins said:

Intel laser disables cores and cache and memory that cutdown sku's don't use, so there is probably at very least a hardware flag saying to not allow HT.

Can be this too, not sure which way they do it. HT specifically.

 

As for your question around how programs handle threads vs cores that it mainly the responsibility of the OS, everything is a thread to a program. Programs ask for resource and the OS schedules that accordingly then it is passed down to the CPU which does it's own scheduling which is fed back to the OS so both can play nicely with each other.

 

Super basic explanation of course, been a while since I've covered this topic so don't want to say anything too wrong or out of date. Plus this whole topic is very complex and there are geniuses out there much smart than me who live and breath it for years.

[Curufinwe_wins]

12 minutes ago, Trik'Stari said:

I find this interesting, as I've only had and used CPU's without HyperThreading except for the I3 I put in the office PC I built for my mom before she passed when she was still working. When it still had the SSD installed as the OS drive (instead of a WD Blue HDD) it ran great for everything she did.

 

I'm wondering if there is any situation where a game, or commonly used office program would actually perform better on a non HT CPU than on a HT enabled CPU. As you say, legacy programs are an issue (aren't they always?), but I can't help but wondering about more modern stuff. I would assume most programs that aren't HT optimized would simply see 4 cores and run as such, and the CPU would just accommodate the program as such (can a HT enabled core function as a "single" core?, or does it simply assign 4 logical threads as though they were physical cores?).

 

CPU's are so complex it's not even funny. I find it difficult to understand how we have created such complex machines.

Almost all of those type of programs are strictly single threaded, so it isn't really an issue, but yea an HT core can operate at basically full capacity (in the same way as a non-ht one) if only one thread is being generated.

 

In the current age where a light general load is 30+ background tasks, I would be very surprised to see much improvement ever by disabling HT on non-legacy workloads. 

[Trik'Stari]

6 minutes ago, Curufinwe_wins said:

Intel laser disables cores and cache and memory that cutdown sku's don't use, so there is probably at very least a hardware flag saying to not allow HT.

I for some reason have a vague memory of @LinusTech Mentioning this in a video, a long time ago.

3 minutes ago, leadeater said:

It's disabled in the CPU firmware/microcode so when you install it in a motherboard it will not present HT as a feature of the CPU.

 

Edit:

Can be this too, not sure which way they do it. HT specifically.

 

As for your question around how programs handle threads vs cores that it mainly the responsibility of the OS, everything is a thread to a program. Programs ask for resource and the OS schedules that accordingly then it is passed down to the CPU which does it's own scheduling which is fed back to the OS so both can play nicely with each other.

 

Super basic explanation of course, been a while since I've covered this topic so don't want to say anything too wrong or out of date. Plus this whole topic is very complex and there are geniuses out there much smart than me who live and breath it for years.

I would assume they are almost beyond genius level. I don't really understand how you go from ones and zeros, essentially on and off switches, to....well, all of this. Even with my own "above average" level of knowledge (compared to the general public mind you) I still feel like it's magic lol

[Curufinwe_wins]

11 minutes ago, Trik'Stari said:

I for some reason have a vague memory of @LinusTech Mentioning this in a video, a long time ago.

I would assume they are almost beyond genius level. I don't really understand how you go from ones and zeros, essentially on and off switches, to....well, all of this. Even with my own "above average" level of knowledge (compared to the general public mind you) I still feel like it's magic lol

If you go back into Anandtech's history, they have some really detailed and good articles in the past about CPUs and the inner workings about how tech like HT works.

 

Like if you read their very detailed very through Pascal article, in the days before Intel released its 7th generation of iterative improvements to the Core uArch, they had similarly detailed cpu articles (PPro-Netburst and k6 Era is a good place to take a look)

 

EDIT: HERE is the article for Nehalem, which is insanely detailed and good. The HT subarticle on its own is a great read.

http://www.anandtech.com/show/2594

[AresKrieger]

Eh whatever the price will determine whether that is a good or bad thing overall, but frankly why the hell do we still have so many (ry)zen rumor threads, people need to just wait for the damn thing to launch before jumping to conclusions, hyping it or belittling does nothing the end product is all that matters

[Trik'Stari]

2 minutes ago, Curufinwe_wins said:

If you go back into Anandtech's history, they have some really detailed and good articles in the past about CPUs and the inner workings about how tech like HT works.

 

Like if you read their very detailed very through Pascal article, in the days before Intel released its 7th generation of iterative improvements to the Core uArch, they had similarly detailed cpu articles (PPro-Netburst and k6 Era is a good place to take a look)

I'll give it a look. Although I feel at this point I'd be so far behind it would take me years to catch up, even with my rather fast reading skills (Not trying to brag, but I've always been a fast reader. A gift I got from my mom, who taught me to read before I started school. I will be forever thankful for that).

 

It's amazing to me that anyone can study this stuff fast enough to actually get a job creating it. I suppose they never stop reading or learning. I wonder if we will hit a point where people cannot train fast enough to actually keep up with the demand for hardware engineers.

[Curufinwe_wins]

12 minutes ago, Trik'Stari said:

I'll give it a look. Although I feel at this point I'd be so far behind it would take me years to catch up, even with my rather fast reading skills (Not trying to brag, but I've always been a fast reader. A gift I got from my mom, who taught me to read before I started school. I will be forever thankful for that).

 

It's amazing to me that anyone can study this stuff fast enough to actually get a job creating it. I suppose they never stop reading or learning. I wonder if we will hit a point where people cannot train fast enough to actually keep up with the demand for hardware engineers.

See the edit. It's pretty easy to catch up. I'm a fast reader myself, but a few months ago just for shits and gigs, I read through every anandtech uArch article on the site in a few hours. Pretty good and informative time, and interesting to see perspectives shift and not-shift.

[Trik'Stari]

2 minutes ago, Curufinwe_wins said:

See the edit. It's pretty easy to catch up. I'm a fast reader myself, but a few months ago just for shits and gigs, I read through every anandtech uArch article on the site in a few hours. Pretty good and informative time, and interesting to see perspectives shift and not-shift.

Will do.

 

Lets let this thread get back to wild speculation and whatnot.