What is a better cpu for quad threaded games? (4 core Kaveri vs. 8 core Vishera)

[cbn]

Hypothetically what would be a better cpu for a quad threaded game?

A quad core Kaveri (with two Steamroller modules) or an octocore Vishera (with four Piledriver modules). Assume the same clocks for both processors.

I know Kaveri has modules designed for better throughput, but then the octocore Vishera would be able to use the resources of the entire module for a single integer core. (ie, Since the Vishera has four modules, each module would only have to use one integer core each)

 

[cbn]

Of course, if anyone has gaming benchmarks comparing 8 core Vishera to 4 core Kaveri in games known to use only four threads please post away.

 

[pw257008]

Just a pure guess, but at same clocks I'd guess Kaveri due to increased IPC (including cache improvements, if I'm not mistaken) and lowered module penalty. However, on the topic of cache, Vishera has more of it, so I'm not so sure.

 

[Enigmoid]

Just a pure guess, but at same clocks I'd guess Kaveri due to increased IPC (including cache improvements, if I'm not mistaken) and lowered module penalty. However, on the topic of cache, Vishera has more of it, so I'm not so sure.

I know Kaveri has modules designed for better throughput, but then the octocore Vishera would be able to use the resources of the entire module for a single integer core. (ie, Since the Vishera has four modules, each module would only have to use one integer core each)

This is probably what is going to happen. Kaveri runs a generally 5% higher singlethread performance than trinity/richland and ~10% better MT performance. However scaling is generally around 80% on the second core.

Vishera = 4 x 100% = 400% performance of PD core

Kaveri = 2*105% + 2*80% = 370% performance of PD core.

My numbers may be off by generally the Vishera will be faster when dealing with 4C code.

However, cache must also be analyzed. Kaveri has a shared L2 cache while Vishera shares L2 within an module and L3 is shared between cores. L3 is extremely slow. Therefore code with a lot of cross talk between cores on Vishera may be slower due to the constant fetching of code from L3 (AMD's caches are not inclusive with the exception of jaguar). However the larger 16 MB vs. 4MB total cache can fit more data in it so trips to memory are reduced. Ultimately it will depend on the code as to the effect from cache.

The console chips use dual 4 core jaguar clusters with a 190 cycle penalty to fetch data from the other cluster's cache. Thus upcoming console ports should avoid dependencies on code closely synchronized on more than 4 threads (likely these cores will handle audio and other things that can easily be slapped on a core and left to run). As we are only scaling to 4 cores in this case this will not be a large factor.

 

[TheELF]

Moot point,there are no "quad threaded games"
A games either has a lot of threads or very few of them,but there are no games that are made with a certain amount of cores in mind.

 

[SPBHM]

I think these benchmarks with OC can give you a good idea
http://pclab.pl/art60391-16.html
they've included 4.5GHz Kaveri and Trinity/Richland, also 4.7GHz "half" Vishera

 

[cbn]

Nvm.

 

[TheELF]

Well there is a kind of sweet point and often dividing tasks into more threads has a negative effect but in general and if everything is done correctly than the more the better.
Look at modern games like mordor they run just great even on lowly dual cores. (42 threads and 8% idle CPU)
42 threads is hardly targeted at the 6 cores of the consoles... (well it is but it only works because console cores are soooooo weak)

 

[SlowSpyder]

Look for some FX4300 benches vs. Kaveri. I know that isn't exactly what you want as it is only a two module design in the FX4300, but it has L3 cache and will give you at least a comparison with the worst case scenario for Vishera.

 

[cbn]

I think these benchmarks with OC can give you a good idea
http://pclab.pl/art60391-16.html
they've included 4.5GHz Kaveri and Trinity/Richland, also 4.7GHz "half" Vishera

Thanks for the link. Here are the charts showing the results from OC processors:

 

[cbn]

 

[cbn]

A few comments regarding the gaming charts I just posted:

1. Too bad there wasn't an overclocked Piledriver octocore in the mix. (Eight core Piledriver is becoming a lot more interesting in the value space considering the prices we have seen for FX-8310)

2. The 4.7 Ghz FX-6300 is the clear winner over the 4.5 Ghz Athlon x4 860K.

3. FX-4300 at 4.7 Ghz and the Athlon x4 860K at 4.5 Gh.z are very close together in most titles. The FX-4300 does pull away in Far Cry 3 and Battlefield 4 MP though.

4. The Intel processors (G3258 at 4.7 GHz and Core i3-4160) do extremely well.

 

[frozentundra123456]

A few comments regarding the gaming charts I just posted:

1. Too bad these wasn't an overclocked Piledriver octocore in the mix. (Eight core Piledriver is becoming a lot more interesting in the value space considering the prices we have seen for FX-8310)

2. The 4.7 Ghz FX-6300 is the clear winner over the 4.5 Ghz Athlon x4 860K.

3. FX-4300 at 4.7 Ghz and the Athlon x4 860K at 4.5 Gh.z are very close together in most titles. The FX-4300 does pull away in Far Cry 3 and Battlefield 4 MP though.

4. The Intel processors (G3258 at 4.7 GHz and Core i3-4160) do extremely well.

Yea, I wondered why the FX6300 was doing so well. They overclocked the hell out of it. That is fine, but too bad they did not show stock clock results as well.

But I agree with another poster. Your thread title is difficult to answer, because I dont know if any games are exactly "quad threaded".

 

[cbn]

Yea, I wondered why the FX6300 was doing so well. They overclocked the hell out of it. That is fine, but too bad they did not show stock clock results as well.

They do have stock clock results:

http://pclab.pl/art60391-3.html
http://pclab.pl/art60391-4.html
http://pclab.pl/art60391-5.html
http://pclab.pl/art60391-6.html
http://pclab.pl/art60391-7.html
http://pclab.pl/art60391-8.html

....but I like the overclocked results because it gets rid of the difference in turbos and makes clock for clock comparisons easier.

 

[DrMrLordX]

I have my doubts about some of the benchmark settings. What were the NB settings? NB speeds allegedly don't affect Vishera all that much, but they do a heck of a lot for Kaveri. Too bad it's so hard to OC Kaveri's NB.

 

[Flapdrol1337]

Good find cbn.

 

[Shehriazad]

I have my doubts about some of the benchmark settings. What were the NB settings? NB speeds allegedly don't affect Vishera all that much, but they do a heck of a lot for Kaveri. Too bad it's so hard to OC Kaveri's NB.

Meh...you can easily bump it up 200 MHZ and in most cases you might only have to add 0.04V (or whatever the lowest extra is on your bios) to make that stable.


That said...due to the lack of L3 cache I'd say fast ram > all on Kaveri.

Been running CPU at 4GHZ (basically just turned Tubo into a perma clock since I hate how the Turbo works) and 2000 NB with 2400 MHZ ram at CL10-12-12-32. Managed to do that while actually undervolting the CPU and the NB is only upped by a single bit. Kaveris NB is only hard to OC on the APU...the 860K takes NB overclocks like a charm...especially since it has way less going on in there.

I owned both 7850K and 860K -> The 860K base clocks its' NB at 1.075 vs 1.2 on my 7850K to give you an idea.

I could probably push it further..but I felt like 2000 is a sweet spot without having to go too deep into overclocking since my board is only so-so (Asrock fm2a88x-itx -> I picked it for the m.Sata slot)

 

[cbn]

Here are the stock clock results from the links I posted earlier:

 

[cbn]

 

[cbn]

Based on the stock clocks comparison, it looks like the FX-6300 clearly wins again.

With that mentioned, I have to wonder how far each chip could be overclocked on the stock cooler?

Does the Athlon x4 860K have significantly more OC headroom than the FX-6300 when both are using their stock coolers?

 

[Shehriazad]

Is that a typo on the stock clock comparison or did they actually DOWNCLOCK the 860K? 3400 Mhz is NOT the stock speed.


Also I wouldnt say it's a clear victory for the FX63(5)0. It pulls ahead in optimized multithreaded games by a bit...but in single/dual only games the 860K wins. For gaming single/dual is still a very important thing to think about at least for now. The FX also costs more than it delivers extra performance.

Overe here the 860K costs 60€ and the FX6350 costs 105€ minimum. That's a price premium of 75% for a performance bonus between MINUS 5% to plus 20ish%? Not sure about you guys...but that seems like a bad deal to me. Might aswell splurge and go for the 8core instead since that one is 140€ but delivers the additional performance in threaded games to make it worth it.


EDIT: Not sure how well the FX6350 overclocks with a stock cooler, but the 860K can actually reach 4.4/4.5 GHZ with the stock cooler without going up in flames (still a little too warm for my taste, gonna be high 50's during full load)...which is about the maximum most people reach with it.(cherry picks excluded)

EDIT 2: After some googling I found some people saying they did OCing with stock cooler on FX6350 and reached like 4.3/4.4 but generally reached at least the mid 60s...that's not something you want to run this CPU with. Not to mention the stock FX coolers are super loud. I'd say in a stock cooler only overclocking scenario the 860K wins simply because of the super low power draw.
The FX has a 125W power draw and actually reaches it...the 860K is rated 95W but generally is FAR below that..AMD was just too lazy to give it a proper rating.

 

[cbn]

Also I wouldnt say it's a clear victory for the FX63(5)0. It pulls ahead in optimized multithreaded games by a bit...but in single/dual only games the 860K wins.

I think it is definitely a clear victory for FX 6300 over Athlon x4 860K because when the FX-6300 wins, the margin can be very large. When the Athlon x4 860K wins it is only by ~3 FPS or less.

In fact @ stock clocks the Athlon x4 860K only wins 3 out of the 15 games and this by the slimmest of margins.

 

[cbn]

Looking at the overclocking results of FX-6300 vs Athlon x4 860K, I see the Athlon x 4 860K only winning 1 title and out 15 and this only by a mere 1.3 FPS.

 

[cbn]

Overe here the 860K costs 60€ and the FX6350 costs 105€ minimum. That's a price premium of 75% for a performance bonus between MINUS 5% to plus 20ish%? Not sure about you guys...but that seems like a bad deal to me.

I know you mentioned FX-6350, but the comparison (in this thread) to Athlon x4 860K has been FX-6300 so I'll give a price comparison for that...



In the United States, the FX-6300 currently goes for $99.99 free shipping at Newegg and Amazon. Tiger Direct is also $99.99 but adds a shipping charge.

The Athlon x4 860K currently goes for $89.99 plus .99 shipping at Newegg. Tiger Direct sells for $89.99 plus shipping. I couldn't find an Amazon price.

So $10 difference (at current prices) going by those major retailers.


Now sometimes both processors will go on sale. I've seen the FX-6300 drop as low as $79.99 free shipping at Amazon and $79.99 plus shipping at other places.

When I see the Athlon x4 860K go on sale it is normally in the high $70 range, but NCIXUS currently has it for $69.99 plus shipping.



So going by these trends, I'd say there is only a $10 average difference in price (in the US) between FX-6300 and Athlon x4 860K.

 

[Shehriazad]

I think it is definitely a clear victory for FX 6300 over Athlon x4 860K because when the FX-6300 wins, the margin can be very large. When the Athlon x4 860K wins it is only by ~3 FPS or less.

In fact @ stock clocks the Athlon x4 860K only wins 3 out of the 15 games and this by the slimmest of margins.

The maximum amount it wins by is 20%.

I was simply looking at it from a pricepoint.

The FX6300 is 35%-40% more expensive while the FX6350 is a whopping 75% more expensive. (again, my country)

It just seems like its losing out there. Furthermore the 860K is stronger in single threading (that's where it wins, really).

If the price really only differs by 10$ where you are at like like was claimed in the post above me...then yes, the FX6300 is definitely a nice choice...but keep in mind that you will most definitely need an aftermarket cooler for the FX6300.

Not only is the stock cooler loud as hell...it's also too weak to handle overclocking unless you want to run the CPU at mid 60°C values. If you live in a warm area and your room is constantly 30°C+ during Summer you will need a better cooler even for stock clocks, 860K won't give a damn...even overclocked.

So you're looking at at least 20-30$ for a cooler, no?.


But yes, looking at just performance, in games that multithread well and use 6 threads...the FX will win.