Should AMD have focused on a Steamroller high TDP product line instead of Vishera?

[mrmt]

Impressive how a game that was the posterchild of the future performance of AMD in games went to a "cripple AMD function" example.

 

[BallaTheFeared]

Impressive how a game that was the posterchild of the future performance of AMD in games went to a "cripple AMD function" example.

Yes, while using an already patched problem with multiplayer as your "example" for a SP result.

 

[Exophase]

Yes, while using an already patched problem with multiplayer as your "example" for a SP result.

And exhibited on both Intel and AMD CPUs and both nVidia and AMD GPUs. Something I'm sure galego noticed once he went looking for more examples of people complaining. Could explain why he didn't link to any more of them.

 

[galego]

Offloading background tasks from the OS to the HT threads, and a clock speed advantage.

Check.

The clock speed only can account for a 2%. If the rest of 10% was due to OS tasks, then we would see a increase in performance in lots of games with HT activated. It is not what we see. Moreover, we would see the 12% increase in crysis 3 always, whereas we see it in Welcome to Jungle in The Lost Island we see the 2% due to the extra 0.1GHz

Your claim that version 1.0 has not HT and the 3.0 patch adds HT is unfounded.

Impressive how a game that was the posterchild of the future performance of AMD in games went to a "cripple AMD function" example.

No. As I said before (http://forums.anandtech.com/showpost.php?p=35141712&postcount=375)

Crysis 3 is optimized for four cores. A game optimized for a half of a 8-core FX chips is not the "posterchild". What was said by several people, including myself, is that the scores of the 1.0 version of the game give an idea of how future games will perform on AMD hardware thanks to better multithreading.

Now that the game has been 'patched' to give more performance in Intel chips whereas dropping the performance in AMD chips, the engine is still more far from the engines of future games (Engines like that in Killzone are already using six cores with efficiency).

 

[BallaTheFeared]

The clock speed only can account for a 2%. If the rest of 10% was due to OS tasks, then we would see a increase in performance in lots of games with HT activated. It is not what we see. Moreover, we would see the 12% increase in crysis 3 always, whereas we see it in Welcome to Jungle in The Lost Island we see the 2% due to the extra 0.1GHz

Your claim that 1.0 has not HT and the 3.0 patch adds HT is unfounded.

Only if a lot of games could highly saturate a quad core, which despite what you would love to believe, very few do.

My claim is more founded than the cripple_amd in a gaming evolved title based on multiplayer issues is though.

We have proof showing it does not use HT in the quads or the hex core i7's prior to this patch in question, what we don't have is usage post patch. Perhaps you could fire up the game and provide us your results using your Core i7 ™ 4th generation processor?

Go look up some AMD x3 reviews and compare them to x2 chips in dual threaded games.

I hope that helps :thumbsup:

 

[frozentundra123456]

The clock speed only can account for a 2%. If the rest of 10% was due to OS tasks, then we would see a increase in performance in lots of games with HT activated. It is not what we see. Moreover, we would see the 12% increase in crysis 3 always, whereas we see it in Welcome to Jungle in The Lost Island we see the 2% due to the extra 0.1GHz

Your claim that version 1.0 has not HT and the 3.0 patch adds HT is unfounded.



No. As I said before (http://forums.anandtech.com/showpost.php?p=35141712&postcount=375)

Crysis 3 is optimized for four cores. A game optimized for a half of a 8-core FX chips is not the "posterchild". What was said by several people, including myself, is that the scores of the 1.0 version of the game give an idea of how future games will perform on AMD hardware thanks to better multithreading.

Now that the game has been 'patched' to give more performance in Intel chips whereas dropping the performance in AMD chips, the engine is still more far from the engines of future games (Engines like that in Killzone are already using six cores with efficiency).

What part of "gaming evolved" do you not understand? I cant believe you would persist in the paranoid assumption that a game that is an AMD title would be patched to decrease the performance of FX, especially when there is hard data that shows to the contrary.

 

[Exophase]

So galego, are you going to keep ignoring my link and keep pretending that the update causes a massive performance only in AMD CPUs and a massive performance improvement in Intel CPUs, or what? Your whole point, the part you bolded, is completely invalidated by it, and I can't believe that you didn't already come to this realization yourself.

 

[galego]

We have proof showing it does not use HT in the quads or the hex core i7's prior to this patch in question, what we don't have is usage post patch.

You have not explained how the facts fit to your theory. The fact are we see a 12% increase in "Welcome to Jungle" but we only see the 2% due to the extra 0.1GHz in "The Lost Island".

If, as you claim, the game had not HT, then from where is becoming the 12% increase between i7 and i5 in "Welcome to Jungle"? Clock explains only a 2%.

If, as you claim, that 12% is because HT is helping OS background tasks then why we only see a 2% increase between i7 and i5 in "The Lost Island"?

 

[BallaTheFeared]

I already attempting to give an explanation, you just ignored it and carried on as you always do.

It's the same reason it's been touted while other results like multiplayer where FX gets destroyed are completely ignored.

 

[galego]

So galego, are you going to keep ignoring my link and keep pretending that the update causes a massive performance only in AMD CPUs and a massive performance improvement in Intel CPUs, or what? Your whole point, the part you bolded, is completely invalidated by it, and I can't believe that you didn't already come to this realization yourself.

Can you explain me where you read the word "massive" in the bolded part? And in the rest of the message?

The review shows that AMD chips, both 6-core and 8-core, lost performance whereas both i5 and i7 increased performance with the new patch. Other people in this thread also noticed the lost of performance for AMD chips. I will not insist on this.

 

[BallaTheFeared]

Performance lost/gain is within margin of error.

 

[galego]

Returning on-topic. Some supposed benchmarks of Centurion and its Steamroller successor

AMD FX 9590 at 5.2 Ghz trounces intel's flagship 3930k at 5.0 Ghz by +15%. The AMD FX 9650 at 5.4 Ghz on the other hand is in a whole new playing field, surpassing intels 3930k at 5.0 ghz by a whopping 27%, and beating their own Piledriver flagship at 5.2 Ghz by +11.5%.

http://amdfx.blogspot.com.es/2013/06/amd-fx-9590-crushes-intel-in-this-brand.html


Cinebench 11.5 Score – Multi-core Test
4C/8T – Intel i7 3770k : 7.52
6C/12T – Intel i7 3970x : 10.84
4M/8T – AMD FX 8350 : 6.93
4M/8T – AMD FX 9450 : 7.55
4M/8T – AMD FX 9650 : 9.15

Cinebench 11.5 Score – Single Core Test
4C/8T – Intel i7 3770k : 1.66
6C/12T – Intel i7 3970x : 1.62
4M/8T – AMD FX 8350 : 1.11
4M/8T – AMD FX 9450 : 1.21
4M/8T – AMD FX 9650 : 1.47

http://www.ptt.cc/bbs/PC_Shopping/M.1369585924.A.364.html

 

[Exophase]

Can you explain me where you read the word "massive" in the bolded part? And in the rest of the message?

The review shows that AMD chips, both 6-core and 8-core, lost performance whereas both i5 and i7 increased performance with the new patch. Other people in this thread also noticed the lost of performance for AMD chips. I will not insist on this.

You kept bringing up that people with AMD processors actually have gotten huge losses. I have shown that those huge losses happened to someone with an i5 too. Therefore they have nothing to do with the huge losses you kept mentioning. The review, which you said was not representative of what people were really experiencing but are now standing behind shows that the AMD processor performed about the same. The difference is too close to actually consider statistically significant. Actually standing behind that to justify your claim that performance got worse for AMD makes you look even sillier than continuing to suggest a link between the big losses you cited earlier and using an AMD processor.

This would be a lot easier if you'd just admit you were wrong to claim that there was a huge performance degradation for people with AMD processors instead of now pretending that didn't happen.

It all comes back down to your underlying claim - that the new results are invalid because they made the Intel score better and the AMD score worse, therefore they're cheating. But the AMD score wasn't really worse. If they used code that Intel happens to get a boost from but AMD does slightly worse with - AVX being the most obvious example - how would that be cheating? How would that not be a legitimate win for AMD? If they gained big wins by using XOP or FMA would you consider that cheating? Would you be using the old version if the Intel score stayed the same but the AMD score got better?

Somehow I don't think so. You'd probably claim that they removed parts that were artificially crippling AMD the first time around.

Also I don't get how you're fine ragging on Cinebench and calling it a synthetic benchmark but when someone uses it to make AMD look good it's totally legitimate.

 

[BallaTheFeared]

Also I don't get how you're fine ragging on Cinebench and calling it a synthetic benchmark but when someone uses it to make AMD look good it's totally legitimate.

All the unconfirmed results link directly back to the same "AMDFX" blog site :thumbsup:

 

[Mallibu]

Returning on-topic. Some supposed benchmarks of Centurion and its Steamroller successor

http://amdfx.blogspot.com.es/2013/06/amd-fx-9590-crushes-intel-in-this-brand.html


Cinebench 11.5 Score – Multi-core Test
4C/8T – Intel i7 3770k : 7.52
6C/12T – Intel i7 3970x : 10.84
4M/8T – AMD FX 8350 : 6.93
4M/8T – AMD FX 9450 : 7.55
4M/8T – AMD FX 9650 : 9.15

Cinebench 11.5 Score – Single Core Test
4C/8T – Intel i7 3770k : 1.66
6C/12T – Intel i7 3970x : 1.62
4M/8T – AMD FX 8350 : 1.11
4M/8T – AMD FX 9450 : 1.21
4M/8T – AMD FX 9650 : 1.47

http://www.ptt.cc/bbs/PC_Shopping/M.1369585924.A.364.html

seems legit

 

[Shift_]

I wonder if galego started that AMDFX blog...every source links to it...but shows no screen shots or any verification.

 

[Ayah]

It's a pretty pointless site.

 

[Hitman928]

I wonder if galego started that AMDFX blog...every source links to it...but shows no screen shots or any verification.

If you track it down to the original blog post on AMDFX, you get this disclaimer:

Assuming: 30% more Ops / Cycle = 20% IPC Benefit over Piledriver . . .


45% increase in performance vs Bulldozer (First Gen) and a 32% increase in performance vs Piledriver (Second Gen). This is taken directly from AMD!

The person gets the 45% and 32% from assuming a 20% IPC increase and also assuming a clock speed bump to 4.5 Ghz. In other words, it's just theoretical calculations based on what could possibly happen if AMD hit these numbers. You could put 10.9% IPC increase and 12.58% overall performance improvement and it would be just as valid. galego, you might want to validate your sources a bit better next time.

 

[zlejedi]

"PCSuperStore.com has the FX-9370 listed for $576.27 and the "5GHz" FX-9590 at a pricey $920.31. Both chips are said to be unlocked, and the part numbers are FD9370FHHKWOF and FD9590FHHKWOF, respectively. Retail-boxed AMD processors usually have "BOX" somewhere in the model code, so I assume these are tray units without heatsinks, manuals, or cool stickers included."

http://techreport.com/news/24968/fx-9000-processors-listed-at-u-s-e-tailer

That pricing is the biggest joke ever since Pentium IV EE cpus.

 

[monstercameron]

That pricing is the biggest joke ever since Pentium IV EE cpus.

the prices aren't verified yet...so stop spreading fud.

 

[galego]

You kept bringing up that people with AMD processors actually have gotten huge losses. I have shown that those huge losses happened to someone with an i5 too. Therefore they have nothing to do with the huge losses you kept mentioning. The review, which you said was not representative of what people were really experiencing but are now standing behind shows that the AMD processor performed about the same. The difference is too close to actually consider statistically significant. Actually standing behind that to justify your claim that performance got worse for AMD makes you look even sillier than continuing to suggest a link between the big losses you cited earlier and using an AMD processor.

This would be a lot easier if you'd just admit you were wrong to claim that there was a huge performance degradation for people with AMD processors instead of now pretending that didn't happen.

It all comes back down to your underlying claim - that the new results are invalid because they made the Intel score better and the AMD score worse, therefore they're cheating. But the AMD score wasn't really worse. If they used code that Intel happens to get a boost from but AMD does slightly worse with - AVX being the most obvious example - how would that be cheating? How would that not be a legitimate win for AMD? If they gained big wins by using XOP or FMA would you consider that cheating? Would you be using the old version if the Intel score stayed the same but the AMD score got better?

Somehow I don't think so. You'd probably claim that they removed parts that were artificially crippling AMD the first time around.

Also I don't get how you're fine ragging on Cinebench and calling it a synthetic benchmark but when someone uses it to make AMD look good it's totally legitimate.

You continue inventing and guessing.

You invented the word "massive", because yjay word is not in my post #129. Now you invent the word "huge", which again is not in my post #129. Why cannot just read that I wrote in #129.

In the past, I cited a link to a FX user that claimed a big loss of performance, you said me that he was inventing that. But when you cited later a user that claims the same in a Intel chip then he is not inventing.

I have not cited the FX user again neither mentioned him, neither used the words "huge" or "massive" in my last posts. Therefore, I fail to see you point.

The performance drop observed in the German review is not statistical because is observed in the six core and the eight core. If you read the comments in the German review, you find people noticing that AMD perform poor now and you can even find some users with conspiracy theories about Intel paying for that. I am not inventing the performance drop in FX chips, albeit you insist.

Cinebench is a biased benchmark, but I can cite it when AMD wins. Simply the real win will be larger. Again fail to see your point.

I wonder if galego started that AMDFX blog...every source links to it...but shows no screen shots or any verification.

No. But I note that when some poster puts here his Intel benchmarks and he shows no screen shots or any verification, nobody worries.

If you track it down to the original blog post on AMDFX, you get this disclaimer:

The person gets the 45% and 32% from assuming a 20% IPC increase and also assuming a clock speed bump to 4.5 Ghz. In other words, it's just theoretical calculations based on what could possibly happen if AMD hit these numbers. You could put 10.9% IPC increase and 12.58% overall performance improvement and it would be just as valid. galego, you might want to validate your sources a bit better next time.

Where is that in the link?

http://amdfx.blogspot.com.es/2013/06/amd-fx-9590-crushes-intel-in-this-brand.html

 

[Phynaz]

No. But I note that when some poster puts here his Intel benchmarks and he shows no screen shots or any verification, nobody worries.

Some poster? Or a reliable trusted poster? Some random blog writer is not the latter.

 

[Exophase]

You continue inventing and guessing.

You invented the word "massive", because yjay word is not in my post #129. Now you invent the word "huge", which again is not in my post #129. Why cannot just read that I wrote in #129.

In the past, I cited a link to a FX user that claimed a big loss of performance, you said me that he was inventing that. But when you cited later a user that claims the same in a Intel chip then he is not inventing.

I have not cited the FX user again neither mentioned him, neither used the words "huge" or "massive" in my last posts. Therefore, I fail to see you point.

The performance drop observed in the German review is not statistical because is observed in the six core and the eight core. If you read the comments in the German review, you find people noticing that AMD perform poor now and you can even find some users with conspiracy theories about Intel paying for that. I am not inventing the performance drop in FX chips, albeit you insist.

Cinebench is a biased benchmark, but I can cite it when AMD wins. Simply the real win will be larger. Again fail to see your point.

You said that there were 50-60 FPS drops for AMD users. It doesn't matter if you used the word massive or not because you were describing massive drops. It doesn't matter if you said it in the particular post you bolded because you said it several times before.

You originally linked to one post where an AMD user said that there was a big performance drop after the patch. What I said is that one vague report is too little and could be wrong, not that he was making it up. You then claimed that other people were reporting similar losses, but since YOU refused to give any links that forced me to go search for it myself. When I found that indeed a lot of people were reporting big performance problems. Including Intel users. At this point it became obvious to me that you knew this and didn't want to say it because it hurt your argument. Instead you had no problem getting other people to waste their time finding it for themselves. You have some serious nerve still talking about "AMD users" reporting that the patch performs poorly now. That has nothing to do with AMD users. Any actual performance degradation specifically on AMD processors, whatever may be present from the TINY amount observed in real measurements, is nothing that someone would give such complaints towards and probably not even something they'd notice outside of benchmarking.

But yes, it could be that there's a tiny degradation; I said that originally it could be because it's using AVX now. But all you say is that performance improved for Intel but got worse for AMD so you should look at the first version and not the newest one. In your world the fact that performance for AMD only went down at most a very small amount apparently means nothing, when it should mean everything.

You said before that Cinebench is SYNTHETIC, not just biased. That should mean it's a bad benchmark and you shouldn't bother with it. But you have no problem bothering with it. You don't give proof that the bench is unfair to AMD. You just assume it's ICC and you assume anything compiled with ICC will penalize AMD, even though that won't necessarily apply to code that has dispatching off. And if it's really biased against AMD in any serious capacity AMD themselves are morons for making it their sole showcase benchmark for the Jaguar SoCs.

 

[Hitman928]

Where is that in the link?

http://amdfx.blogspot.com.es/2013/06/amd-fx-9590-crushes-intel-in-this-brand.html

Like I said, you have to follow the trail,

http://amdfx.blogspot.com.es/2013/05/amd-steamroller-fx-9650-45-ghz-48-ghz.html

Is where it starts. For your link, the blogger makes the assumptions I listed above, finds one of the fastest hwbot/cinebench/whatever benchmarked 8350 submissions he can find, and then uses the assumptions to calculate the theoretical streamroller part. This is not worth posting.

 

[dac7nco]

For your link, the blogger makes the assumptions I listed above, finds one of the fastest hwbot/cinebench/whatever benchmarked 8350 submissions he can find, and then uses the assumptions to calculate the theoretical streamroller part. This is not worth posting.

I can't wait for reviews... This is "FX" 8150 part-II, which was also released at a ridiculous price. Tell me these things aren't actually going for nearly a grand. When Bulldozer was reviewed, this forum was semi-adult again... for a few months. Of course, Intel paid all of the big review sites, and has Jimmy Hoffa in a freezer.

Daimon