• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Ryzen SMT vs Intel HT

I

Irquk

Junior Member
Given the issues with the Windows 10 scheduler and Ryzen's SMT and the gaming benchmarks showing that Ryzen 1800X gaming benchmarks can improve when SMT is disabled, are there any head to head gaming comparisons of the Ryzen 1800X with SMT disabled and an Intel i7 (maybe a 7700k?) with Intel's hyper threading disabled? If so, could you direct me to them as I'm curious to see the raw core comparison without the HT boost on the i7. Thanks in advance.
 
Intel Core i7-7700K without Hyper-threading (minus 0.4 GHz all-core frequency, minus 2 MB L3 cache) = i5-7600K
 
Im hoping to see some articles on Ryzen's SMT and how/what makes it operate more efficiently with better scaling than HT. Crazy for first attempt
 
piesquared said:
Im hoping to see some articles on Ryzen's SMT and how/what makes it operate more efficiently with better scaling than HT. Crazy for first attempt
Click to expand...
While Intel's HT is likely the best known use of SMT, it isn't the best. Good? Yes, even very good. The IBM Power8 and Power9 CPU's take SMT to an entirely new level. Intel's HT has been around for quite a while now. More than enough time for people doing a clean sheet design to do a better job.
 
I think some design goals for SMT came at the derogation of ST. Turning it off helps ST but all the resources can not be used as effectively as it could have. It's a trade off that hurt gaming performance. Intel's HT did not have the same goal. Their goal was for HT to use unused resources. The trade offs played out in different ways.

Skylake/Kabylake HT is as good as AMD's SMT but with superior single core IPC. Zen with less cores will show even more deficit in performance.
 
Conroe said:
I think some design goals for SMT came at the derogation of ST. Turning it off helps ST but all the resources can not be used as effectively as it could have. It's a trade off that hurt gaming performance. Intel's HT did not have the same goal. Their goal was for HT to use unused resources. The trade offs played out in different ways.

Skylake/Kabylake HT is as good as AMD's SMT but with superior single core IPC. Zen with less cores will show even more deficit in performance.
Click to expand...
I think it's a bit early to jump to a conclusion there. MS still needs to get it's act together for example.

While I do think Ryzen is a great chip, and it will be the centerpiece of my next build this summer, the launch blew chunks. Motherboard manufacturers dropped the ball badly. Tons of CPU's available, but no motherboards to put them in for example. The Bios issues are part AMD, and part motherboard manufacturers. Since some work a LOT better than others.

Overall, I'd give the Ryzen design an A-. The launch a C-. And there is enough in the design, that I expect things in the CPU market overall to start picking up some heat and excitement. Well, tech nerd excitement anyway.

It's a great part, and one I will be buying. I just wish it had launched a little better.
 
Some truth to that, but MS has worked with Intel's HT. Why did AMD not ride in on their coat tales?
 
Conroe said:
I think some design goals for SMT came at the derogation of ST. Turning it off helps ST but all the resources can not be used as effectively as it could have. It's a trade off that hurt gaming performance. Intel's HT did not have the same goal. Their goal was for HT to use unused resources. The trade offs played out in different ways.


Skylake/Kabylake HT is as good as AMD's SMT but with superior single core IPC. Zen with less cores will show even more deficit in performance.
Click to expand...

I wouldnt say HT is as good as Ryzen's SMT but its close. Zen 2 has a number of areas available for improvement, so i expect expect the IPC in Zen 2 to go up more relative to Coffelake while Ryzen's SMT efficiency remains.
Games need optimizations, thats just how it is. AMD's architecture will need optimizations just as intel's architecture does.
4 core Ryzen is very competitive with 4 core Kabylake, those tests have already been done.
 
Conroe said:
Some truth to that, but MS has worked with Intel's HT. Why did AMD not ride in on their coat tales?
Click to expand...
Because AMD & Intel's SMT (or HT as it's called) implementations might be vastly different? You can already see how SMT works for POWER, it works greats so long as there's a good level of MT performance (or parallelism) that can be exploited.
 
Conroe said:
Some truth to that, but MS has worked with Intel's HT. Why did AMD not ride in on their coat tales?
Click to expand...
Ryzen's SMT is excellent. The MS problem is in their Windows 8 to 10 scheduler. Both Linux and Windows 7 don't seem to have this issue. Linux in particular is NUMA aware.
 
Okay, this is just AMD's beta release phase. It'll be ready for Naples. The design goal never was for gamers.

AMD went for SMT over ST. Get used to it. It's not going to change in this iteration.
 
Last edited:
Conroe said:
Okay, this is just AMD's beta release phase. It'll be ready for Naples. The design goal never was for gamers.

AMD went for SMT over ST. Get used to it. It's not going to change in this iteration.
Click to expand...
They designed an architecture that was competitive with Intel wrt IPC, the GF node they're struck on though (14nm) might decide if they can clock the heck out of Zen as Intel's been able to do with their third iteration (KL) on the same node. The gamer thing is way overblown anyway, if you can do 5GHz (or more) on Devil's Canyon you don't need KL unless of course it's an OCD thing.

SMT vs ST isn't even a thing, they got ST performance as high as they possibly could on 14nm LPP(?) the SMT bonus is on top of that.
 
If people consider the financial side of things most of the money comes from selling servers.
So I think they are targeting Broadwell-D and hence they made their choices by comparing Broadwell class chips like 6900K.
There is also EX series and the Naples platform is probably designed to target it as well.

Servers don't run at very high frequency (@ 4 Ghz) but more sedate at 2.0-3.0 Ghz. So performance must be fine-tuned to hit that target at as low power as possible. This is important as server cooling plays a large part in operational costs. Many server workloads are bursty with long periods of low IPC and SMT helps in filling up the execution units. There is also the software licensing side of things where software is licensed on number of cores,sockets etc but I'm not sure how important role it plays.

It may be noted that this is also true for mobile platforms as they live in thermally constrained environments.

This also proves the following conclusion:
1) The ability to overclock the cpu and pushing it beyond its design specifications is the probably the least important aspect of the chip. ( I know people here are not going to like hearing this)
2) Also low in the pecking order is the performance of the chip in games. AMD may not have tested their chip in gaming related scenarios as much as they tested server workloads. (It seems that this is most like a scheduler bug or some problem related to PCIe lanes on the chipset. )
 
scannall said:
Ryzen's SMT is excellent. The MS problem is in their Windows 8 to 10 scheduler. Both Linux and Windows 7 don't seem to have this issue. Linux in particular is NUMA aware.
Click to expand...

Windows 10 currenty must be set with BCDedit manually, but can be set as NUMA aware...
 
As well as the tests in CB review already linked, Hardware.fr covered performance of SMT&HT on an off for Ryzen 7 1800X and Core i7-6900K
http://www.hardware.fr/articles/956-7/impact-smt-ht.html
For apps, both gained from SMT.
For games, both mostly lost from SMT but Intel lost less than AMD. On average, for gaming Intel had 98.1% of their non-SMT performance; AMD had 91.1% of their non-SMT performance although with 'Core parking' disabled, they too had 98.1%.
getgraphimg.php
[/IMG]
getgraphimg.php
 
And I have been saying for like ever that HT does not help games. If a game needs more threads you need more actual cores.
 
As I recall, the first Pentium 4 with hyperthreading had problems running a little slower with everything when hyperthreading was enabled. How did they get around that back then?
 
zephyrprime said:
As I recall, the first Pentium 4 with hyperthreading had problems running a little slower with everything when hyperthreading was enabled. How did they get around that back then?
Click to expand...
I still have a 3.06Ghz Northwood with HT, and it still works fine. It was the first desktop chip with HT. IIRC the performance losses with it was simply because most software was unaware of how to utilize it properly.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top