AMD will launch AM4 platform in March 2016 says industry source

[AtenRa]

Battlefield 4 64 player had a minimum FPS of 1.

So if you made a you tube video of that gameplay I'm sure we would have noticed that.

Well i wouldnt count this one, but you cannot notice the stuttering in the rest of the game play from a video, unless you play the game or see the graphs above.

 

[DrMrLordX]

How is the G3258's stuttering well documented? Under what conditions?

G3258 stuttering is legendary. The only way to prevent it is to pair it with a weak dGPU and turn settings down low. Which is why nobody recommends the things for gaming PCs despite their overclocking capability.

Here's one example:

https://www.youtube.com/watch?v=TflMlYNgKec

 

[mikk]

Nvidia suffers from only two threads, it should run better with AMD, but some games really need at least 4 threads to run properly. For a gaming system 2/2 CPUs are poor, and he paired it with a GTX 970 lol.

 

[AtenRa]

Nvidia suffers from only two threads, it should run better with AMD, but some games really need at least 4 threads to run properly. For a gaming system 2/2 CPUs are poor, and he pairs it with a GTX 970 lol.

You can pair it with a HD7950 and Mantle and it will still stutter. Dual Cores are not for 2015 gaming onward, simple as that.

 

[DrMrLordX]

Nvidia suffers from only two threads, it should run better with AMD, but some games really need at least 4 threads to run properly. For a gaming system 2/2 CPUs are poor, and he paired it with a GTX 970 lol.

Exactly. But many expected that they could turn the G3258 into a gaming CPU by just running up the clockspeeds, which doesn't work. Even FX CPUs are smoother than that regardless of what dGPU is in use.

i3s are better candidates for gaming CPUs than the 2c/2t chips, generally speaking. Though AtenRa's charts give one food for thought.

 

[cbn]

G3258 stuttering is legendary. The only way to prevent it is to pair it with a weak dGPU and turn settings down low. Which is why nobody recommends the things for gaming PCs despite their overclocking capability.

Here's one example:

https://www.youtube.com/watch?v=TflMlYNgKec

Notice the GPU is GTX 970.

And pairing OC G3258 to that GPU is not a realistic use case scenario.

Oh, and gaming PCs can use smaller GPUs than GTX 970.

 

[cbn]

You can pair it with a HD7950 and Mantle and it will still stutter.

My R7 250X works great with my OC 3258 in BF4 64 player (using Mantle).

Haven't tried anything larger yet on the AMD side. With my 660 GTX (using DX11) it does stutter though.

 

[cbn]

More info from Digital Foundry on the causes of stutter:

https://www.youtube.com/watch?v=Rutk9ErhKG4

So, is this CPU stutter found in the Novigrad stress test a cause for concern? Well, not really. Most in-game stutter is caused when the CPU - not the GPU - is the bottleneck (and that's what we are testing here). As long as you pair your more budget-orientated CPU with an appropriate GPU, you'll hit the GPU limit first - and typically that doesn't cause stutter. And in the case of The Witcher 3, most of the game is GPU-limited (as seen in the first couple of cut-scenes tested here). With that final stress test scene, you'll note that the 4790K is well-matched with the Titan X, the CPU isn't really the bottleneck and thus the latencies are much more consistent.

 

[cbn]

Dual Cores are not for 2015 gaming onward, simple as that.

It is not as simple as that.

Certainly for the games that current A10 APUs play (and most likely A10 Bristol Ridges too) they are valid.

In fact, one argument has been that dual core Intel + dGPU is a better value than A10 APU for that level of gaming.

 

[The Stilt]

It is not as simple as that.

Certainly for the games that current A10 APUs play (and most likely A10 Bristol Ridges too) they are valid.

In fact, one argument has been that dual core Intel + dGPU is a better value than A10 APU for that level of gaming.

With the slight difference being that AMD A10/A8 series chips have four cores, whereas anything below i5 Intels have two :sneaky:

SMT and CMT should never be compared either in energy efficiency or performance wise. The most recent AMD CMT implementation produces > 91% average yield while the most recent Intel SMT implementation peaks at ~26.5%. AMD has a huge performance benefit here when comparing the number of threads the CPU can execute. The CMT efficiency obviously comes at cost thou.

Effectively Intel 2C/4T is worth 2.53 cores while AMDs 4C/4T solution is worth 3.83 cores (Excavator).

 

[cbn]

With the slight difference being that AMD A10/A8 series chips have four cores, whereas anything below i5 Intels have two :sneaky:

I was thinking of the iGPU on the A10 APUs being the bottleneck for gaming comparison of A10 vs. Pentium (stock clocks) and dGPU (like GT 730 GDDR5).

And for the CPU part, I was thinking of how two stock clocked Pentium cores (that is 2C/2T) would compare to a throttled 4C/4T. (This based on gaming charts of 4.5 GHz G3258 vs. 4.5 Ghz Athlon x 4 860K).

 

[cbn]

And regarding those gaming charts of 4.5 Ghz G3258 vs. 4.5 Ghz Athlon x 4 860K, we have to remember big dGPUs were used for those tests.

With small GPUs, I doubt the situation would be the same with the minimum FPS in certain cases.

 

[cbn]

Thinking about Bristol Ridge more, I would hope AMD could put a greater focus on 35W mobile for that processor......then delay AM4 till Zen 8C/16T arrives.

The 35W mobile could even be used for AIO desktop (if necessary), but I really think we need some good laptops. 35W laptops >>>> 35W AIO desktops.

 

[frozentundra123456]

With the slight difference being that AMD A10/A8 series chips have four cores, whereas anything below i5 Intels have two :sneaky:

SMT and CMT should never be compared either in energy efficiency or performance wise. The most recent AMD CMT implementation produces > 91% average yield while the most recent Intel SMT implementation peaks at ~26.5%. AMD has a huge performance benefit here when comparing the number of threads the CPU can execute. The CMT efficiency obviously comes at cost thou.

Effectively Intel 2C/4T is worth 2.53 cores while AMDs 4C/4T solution is worth 3.83 cores (Excavator).

That is only part of the story though. Accepting your analysis for the sake of argument, those 2.5 intel cores (I think hyperthreading could be more efficient than that, and 91% efficiency for 4 modules seems awfully optimistic as well, but whatever) are a lot faster per core than the Bulldozer cores. I would not want a Pentium for gaming unless I was interested only in older games, but for dgpu gaming I would take a Haswell or Skylake i3 any day of the week over any FM2 cpu. In fact an i3 is very competitive with 8 core FX in a lot of games.

 

[DrMrLordX]

Notice the GPU is GTX 970.

Duh. That's the kind of card people seem to want to use with an overclockable chip. They don't want to go low with a 250X. You know, cheap out on the CPU and spend more on the GPU? That's a common mantra around here.

Even if it isn't a 970, you can go out to fleabay and get an HD 7950 for ~$140 shipped, and that stutters on an OCed G3258 too.

If you upgrade your GPU and the games start to stutter, there's a problem.

 

[cbn]

They don't want to go low with a 250X.

If you think a R7 250X (640sp @1000 Mhz with 128 bit GDDR5) is a low end card (for G3258), then how do you feel about the Kaveri/Godaveri and Bristol Ridge 512sp iGPU for desktop?

 

[cbn]

Just going by GPU core alone, 640sp @1000 Mhz (R7 250X) is ~74% more than 512sp @ 720 Mhz (A10-7850K) and 44% more core than 512sp @ 866 Mhz (A10-7870K).

 

[AtenRa]

What the APUs have to do with dGPUs and dual core stuttering ??

And I havent seen anyone recommending the G3258 for low budget dGPUs only. Everyone was looking the review graphs of only Average fps and they recommending the OC G3258 for gaming even with faster dGPU.

Do I have to remind you all the moto of Pentium being faster than FX etc ??? All those recommending the dual Core Pentium over the Quad Core Athlon were fell for the high average fps "promotion" review sites were tooting last year.

Unfortunately those who bought the Pentium G3258 over the Athlon 860K now need to upgrade both their low end dGPUs AND their CPU. So at the end they spend way more than they were originally planing to do.

Well, that is what marketing does and most of the review sites have become to much connected to the Hardware companies that they never transfer the real story to the public. I dont remember any review of the G3258 talking about stuttering last year. And im sure they knew about it but they didnt even mentioned it because they didnt want to "upset" their hardware supplier.

For example, i was looking for some Core i3 6100 reviews the other day and started to read techspot Core i3 6100 review.
The FX8320E currently at $140 on Amazon was faster in almost every CPU benchmark they included in their review on stock clocks when Core i3 6100 is at $125. And using only a GTX960 the gaming results where within a few fps.

And here is their conclusion,

Intel's Core i3 range still trumps AMD's most affordable FX processor and the Skylake Core i3-6100 just cements that fact.

Really ???

They also included the Hasswell Core i5 4430 currently at $190 on Amazon. The Core i5 was faster even than the Core i3 6100 with DDR-4 3000MHz on every CPU and gaming benchmarks. And yet they havent said anything to recommend the Core i5 + H81 + 1600MHz memory at close the same price as the Core i3 with H110 motherboard and DDR-4 3000 Memory.

 

[cbn]

And I havent seen anyone recommending the G3258 for low budget dGPUs only.

I have been since seeing the Eurogamer.net/Digital Foundry article (written in July 2014) mentioned later on in this post. And others have been as well.

Everyone was looking the review graphs of only Average fps and they recommending the OC G3258 for gaming even with faster dGPU.

What happened was the Core i3 was much faster in pretty much all reviews. So anyone wanting to use a relatively strong GPU was pointed away from the G3258 for at least that reason. (ie, too many FPS left on the table with G3258 vs Core i3. This was the most obvious reason. Why waste extra money on dGPU when for $50 more a person could get a Core i3 and gain more that way.)

I dont remember any review of the G3258 talking about stuttering last year. And im sure they knew about it but they didnt even mentioned it because they didnt want to "upset" their hardware supplier.

Tom's hardware did frame time analysis of the G3258 vs. Athlon x 4 750K in June 2014, the month it was released--> http://www.tomshardware.com/reviews/pentium-g3258-overclocking-performance,3849.html.

(OC G3258 won 4 out of 7 games compared to OC Athlon x 4 750K):

http://forums.anandtech.com/showpost.php?p=37175841&postcount=65

No one brings terrible frame times of pentium? Very strange to see people give any weight to average fps numbers these days.

Pentium OC G3258 actually won 4 out of 7 games (in terms of frame time variance) against OC Athlon x4 750K in the Tom's test:

http://forums.anandtech.com/showpost.php?p=36517046&postcount=17

http://forums.anandtech.com/showpost.php?p=36517050&postcount=18

http://forums.anandtech.com/showpost.php?p=36517071&postcount=19 (<---Here is where I tallied results)

,...but one of the newer games Thief there was a large discrepancy in frame time variance favoring OC Athlon x4 750K.

http://forums.anandtech.com/showpost.php?p=36873247&postcount=30

Unfortunately Thief was very new at the time (it was the newest game in the test I believe). My conclusion (after comparing Thief to the oldest game to scale quad core in the test, Metro Last Light) was that is was probably a driver issue.

http://forums.anandtech.com/showpost.php?p=36875563&postcount=49

And Digital Foundry was investigating stuttering the month after G3258 was released:

http://www.eurogamer.net/articles/digitalfoundry-2014-pentium-g3258-review

(sample quote below)

In the highlighted video, you can see that running the game at the high preset (that's one 'notch' down from the ultra-equivalent, very high) in combination with a GTX 760 results in a night-and-day performance differential between the i7 4790K and the Pentium. The additional fidelity in the game simulation, coupled with the immense increase in GPU set-up costs, sees the Anniversary Edition Pentium struggle horrendously to keep pace. What we're seeing here is a classic case of a lack of hardware balance: the G3258 simply can't feed the GTX 760 quickly enough to sustain a consistent frame-rate.

Now, compare and contrast with the secondary analysis, where we drop the GPU down to a far more modest GTX 750 Ti, and lower the overall quality preset to the medium level. In this case, for the most part it is the graphics card that is the bottleneck, and the overall performance level lowers the i7 advantage significantly.

 

[AtenRa]

750K is a Trinity (Piledriver) 2012 product. At the time of G3258 release the Athlon 860K was available that has 15-20% higher IPC and higher CMT scaling over 750K.

 

[cbn]

750K is a Trinity (Piledriver) 2012 product. At the time of G3258 release the Athlon 860K was available that has 15-20% higher IPC and higher CMT scaling over 750K.

860K didn't launch till August 2014.

http://www.cpu-world.com/CPUs/Bulldozer/AMD-Athlon II X4 860K.html

 

[AtenRa]

yeap you are right, it was released in August.

Anyway the TH review did show the high frame times of G3258 but the vast majority of users ignored it and looked at average fps.

 

[cbn]

Anyway the TH review did show the high frame times of G3258 but the vast majority of users ignored it and looked at average fps.

In that Tom's review, the OC Athlon x 4 750K had higher frame times in more games than the OC G3258.

And that is probably one reason why folks looked the other way. The other two reasons (CPU/GPU balance and Core i3) I have already mentioned.

It wasn't till we got to GTA V that the issue of smooth gameplay and G3258 was brought up again, but this time the focus was on minimum frame rates and the competing processor was now the Athlon x 4 860K rather than Athlon x 4 750K:

http://pclab.pl/art57777-23.html

http://pclab.pl/art57777-24.html

Notice in the GTA V benchmark above with the Nvidia GTX 970 OC, the OC G3258 has a minimum FPS of 15 and an average FPS of 31.7. The OC Athlon x4 860K has a much better minimum FPS of 25, but a slightly lower average FPS of 29.9 FPS.

Obviously 25 minimum FPS is much better than 15 FPS!

So OC Athlon x 4 860K wins when using a big overclocked Nvidia card

However, when looking at the same benchmark using the OC R9 290X, the OC Athlon x4 860K doesn't do that much better than the OC G3258 in minimum FPS:

(Athlon x 4 860K @ 4.6 GHz = 18 min FPS, 21.6 avg FPS)
(Pentium G3258 @ 4.5 Ghz = 15 min FPS, 26.1 avg FPS)

Athlon x4 860K is only 3 FPS better in minimum FPS compared to OC G3258 and is 4.5 FPS slower in average FPS.

So it appears for GTA V, the brand of video card is another variable to consider.

And in this case of a OC R9 290X having a OC Athlon x 4 860K rather than OC G3258 might even be worse if the person insists on having at least 25 avg FPS. (With this mentioned, I do consider both OC GTX 970 and OC R9 290X to be unrealistic video cards to pair with such low end CPUs for a practical PC build-up.....and if using a smaller video card with OC G3258 I believe we would see a different situation with regard to the minimum frame rates).

 

[DrMrLordX]

If you think a R7 250X (640sp @1000 Mhz with 128 bit GDDR5) is a low end card (for G3258), then how do you feel about the Kaveri/Godaveri and Bristol Ridge 512sp iGPU for desktop?

Nice try, pal. Nice try. Many who actually bought and used overclocked G3258s completely passed over the 860k, 7850k, and so forth, thinking that the G3258 would wind up being the better gaming CPU (or thinking that they had a better CPU upgrade path, which is generally correct).

There are many people out there who paired the G3258 with GPUs that are "too big", either older cards they had lying around from previous builds, or cards that are simply popular with everyone else. The G3258 is an Intel chip and it lets you overclock, how could they go wrong????

Well, as it turns out, all the single-threaded IPC in the world isn't going to save you on newer software titles.

You can try to defend the G3258 all you want, but the fact is that it has some absurd limitations when it comes to what GPUs you can and can't use with it. You said yourself that a GTX 660 was causing stuttering. Seriously? A card from 2012 and the G3258 can't handle it?

It's a cool chip for what it does. Some emulators fly on it. If you like high SuperPi scores on a budget, it's still an interesting option (though a 5775c will probably smoak everything right now). If you like StarCraft 2, hey, good budget choice! But the actual expectation that this chip would be good for at least mid-range gaming is completely off. It turns out that a lot of FM2+ chips are better host for mid-to-high end dGPUs than the G3258. That hasn't stopped a slew of people from trying the G3258 + high-end dGPU combo anyway.

Anyway the TH review did show the high frame times of G3258 but the vast majority of users ignored it and looked at average fps.

Bing bing bing, we have winnar.

 

[cbn]

They don't want to go low with a 250X.

If you think a R7 250X (640sp @1000 Mhz with 128 bit GDDR5) is a low end card (for G3258), then how do you feel about the Kaveri/Godaveri and Bristol Ridge 512sp iGPU for desktop?

Nice try, pal. Nice try. Many who actually bought and used overclocked G3258s completely passed over the 860k, 7850k, and so forth, thinking that the G3258 would wind up being the better gaming CPU (or thinking that they had a better CPU upgrade path, which is generally correct).

R7 250X is a good combo with G3258.

And if you are going to say that is too weak for G3258 then we need to compare it to the AMD A10 APUs as well.

Fact is dollar for dollar that G3258 + R7 250X was a much better value than any A10 desktop APU (and one reason was the throttling CPU cores onthe APU for the average user who didn't know how to manipulate AMDMrsTweaker and the other reason was lack of bandwidth on the A10 APU).

A10-7850K with CPU cores throttled down to 3.0 Ghz and 512sp iGPU @ 720 Mhz with cost adding DDR3 2400?

vs.

G3258 @ 4.0+ GHz and 640sp dGPU @ 1000 Mhz with 128 bit GDDR5?

That is no contest. (Even at stock CPU clocks I'll bet G3258 wins)