AMD Polaris 10 Has 390/390X Performance

Well that and it will give us a good insight into what Vega will bring to the table.
 
razor1 said:
well they had voltage increases, don't expect Finfet do be able to handle voltage increases the same way as just a frequency increase.

Also you were pushing your card out of validation limits, it might have worked well for you. It is not something a company likes to do when they have to warranty their cards for certain amount of time.
Click to expand...

Fair call. Was more meaning that such clockspeeds were possible with early 28nm GCN sillicon. With FF by now I would expect higher clocks than that as base.
 
jamesgalb said:
When these chips come out, I'd like to see them downclocked to the speed of their 300 series counterparts, to see how much performance is gained from the new architecture improvements... Would be an interesting benchmark for sure.
Click to expand...

well 1080 if downclocked to 980ti, it wont be much faster, so architecture wise there are minor enhancements.
 
N4CR said:
Fair call. Was more meaning that such clockspeeds were possible with early 28nm GCN sillicon. With FF by now I would expect higher clocks than that as base.
Click to expand...
Is this 1300mhz 7970 still alive? lol what voltage was it running? I had to flash mine just to get 1150mhz...that card must have been on high end water?:)
 
primetime said:
Is this 1300mhz 7970 still alive? lol what voltage was it running? I had to flash mine just to get 1150mhz...that card must have been on high end water?:)
Click to expand...

Good question - it died dec last year while I was out of the country. But someone else was fucking around with my PC and could've possibly ESD'd it or PSU going took it out... dual PSU rig and it would've got most of the issues being mostly on one PSU. Volts from memory were maxed on slider (MSI AB) for 1.3GHz but about 1.25VDDC from memory, which is 100mV over stock I think? Actual voltage was well under 'set voltage' with droop. Temps 80-90 maxed out.

Was a ref card, stock. Did 1.2 on stock volts with the bios flash! Before that it wasn't so easy to OC. Used the ASUS bios, still have it if you want to try.

That said my current 290x doesn't clock for shit and frankly it's a bit faster than the old 7970 without all the noise and power whoring, more than enough for now.

edit: [H] review card was practically the same as my experience and there were more than a few of them too.
 
Last edited:
schizo said:
Seems about right. Was that a 390x or a 390?
Click to expand...
It's unknown if it's 390 or 390X, but I would say it's likely 390X.

67DF is the cutdown Polaris 10. The 67C0 which we haven't seen is the full one. If we get 390X performance from the cutdown one, we could get 980Ti or more from the full one.
 
If it's being compared to a 390x then yes, that's definitely promising. But there's a huge gap between a 390x and a 980ti!
 
razor1 said:
390X gets higher than that more around 50 fps at those settings.

Just check the ladder 1440 high

Ashes of the Singularity

Fury at those settings

Ashes of the Singularity

Something isn't right about those benchmarks unless its a patch that is older then the latest one out there.
Click to expand...


You have to compare at least the same CPU which is why the 390 series and Fury series benchmarks are useful for comparison, it's all done by the same person and on the same system. The game version is listed as well, the 390 series and Polaris 10 1440p benchmarks are on the same game version, while the Fury is done on a slightly older version.
 
Ones with i5 4 cores have higher frames than the one you showed, that should not happen especially at the higher batch amounts........

This is like the same thing with the gtx 1080 AOTS benchmarks, numbers aren't aligning like they should.
 
50fps is the average for the R9-390 series on a 4790K (a much faster CPU than the 3970X that guy used). But it varies from 28 to 75 in only 12 samples, and we can't distinguish between 390 and 390x-- I just don't think there's enough data to come to any sort of meaningful conclusion there.

Ashes of the Singularity

All we KNOW is that on this guy's specific machine, the cut-down Polaris 10 performed similarly to either a 390 or 390x. So... take it for what it is, a performance indicator for sure, but not an answer.
 
  • Like
Reactions: N4CR
like this
There are other factors as well such as system RAM speed. If it's a test system, there may be development software running in the background eating resources too. These aren't official benchmarks or anything.
 
Last edited:
I would take any benchmark leak-posted result with a very large pinch of salt for NVIDIA and AMD when it comes to their 14/16nm GPUs.
Cheers
 
variant said:
There are other factors as well such as system RAM speed. If it's a test system, there may be development software running in the background eating resources too. These aren't official benchmarks or anything.
Click to expand...
Yep, we're talking about the difference between 28 and a whopping 75 FPS with the exact same CPU and one of two GPUs. That variance is ginormous. The data simply isn't useful.
 
schizo said:
I admire your optimism sir, but I do not share it.
Click to expand...
Eh not much time left anyway. we will all know. i want it to slot in just under the 980ti hopefully and for 250ish. Then I would be all over it.
 
  • Like
Reactions: N4CR
like this
That would be fabulous for AMD and consumers. And maybe they'll pull it out of their hat. Just seems doubtful given all we've seen so far.
 
{NG}Fidel said:
Eh not much time left anyway. we will all know. i want it to slot in just under the 980ti hopefully and for 250ish. Then I would be all over it.
Click to expand...

That would be quite a coup, given that the 1070 is likely around a 980ti (give or take) for $380/$450.
 
Well it also depends on Polaris 10 memory DDR5 or DDR5X, I just don't think AMD will be able to make up the bandwidth disparity between previous generations using GCN wider bus/HBM and using a 256bit bus and DDR5. I also have a hard time believing AMD highest end Polaris will be a 130w part - 150w would be rather low and a 175w part would still be good. Unless AMD wants to make the whole card as cheap as possible for maximum amount of sells. AMD will have a winner if the top end Polaris is > 390x performance, less than $300, less than 180w. there will be different versions of Polaris and yes a 130w and less will probably be available but a 175w version is not out of the question either. Memory bandwidth maybe the restriction on how fast Polaris will be.

As for clock speeds, it is a new GCN arch, new node, who knows what was the goals - clock speed could be high as well or not. We just have to wait and see unless someone already has a card that represents what is going to be released. For the most part everything is more speculation then not here.
 
I think you've touched the point with a needle on the bandwidth. I think we can assume there's no way P10 at $300 will have DDR5X. Not happening. Regardless of how fast the chip is, you are going to be limited with regular old DDR5 and a 256-bit bus. We may be looking at a card that performs admirably at 1080p and adequately at 1440p, but is a sufferance at higher resolutions -- in other words, a chip aimed at the mid-range.
 
Memory interface on the chip itself will be the limitation. The AIBs will make it with 20gb of XDDR10 ram and all the rest of the marketing fluff in due course.
 
releasing a really good mid-range card is probably more important to amd than release the very highest end card. I know id rather sell a million $300 cards than sell a hundred thousand cards at $600. low-mid range cards are the highest volume, im sure they are aiming for that.
 
  • Like
Reactions: NKD
like this
razor1 said:
They will not drop margins if they don't have to, new nodes are more expensive now then ever, and actually they are already dropping margins by going lower than $300 bucks going by rumored transistor counts.....

Do you realize nV had dropped margins with the 1080 and relatively the same with the 1070 at the pricing they mentioned at their announcements?
Click to expand...


Did they drop margin? the 970 launched with a price target of 330, the 1070 380, that extra 50 dollars yields less margin for nvidia? Is it THAT much more expensive to build the chips on the smaller nodes (but also smaller die size)?
 
per transistor the 16nm node is ~x2 the cost, die size that means of the gtx 1080, gtx 1070 are half the die size of the gtx 980 or gtx 970 which it isn't its larger than half the size of gm104 and is priced the same as both then you end up with similar margins to the 980ti.

Going by the transistor amounts that we now have, the gtx 1080 kept margins the same since it has 10% less transistors, 10% less cost, the gtx 1070 margins seemed to have gone up by 10%, but thats not the case, since the transistor counts of the gm104 is 20% lower than what is on the gp 104, so the increased cost offsets this, so the margins will be around the same. That is also if we don't count into the cost of the extra ram and the use of GDDR5x on the 1080.
 
Last edited:
tybert7 said:
Did they drop margin? the 970 launched with a price target of 330, the 1070 380, that extra 50 dollars yields less margin for nvidia? Is it THAT much more expensive to build the chips on the smaller nodes (but also smaller die size)?
Click to expand...

Actually, I really do think it is more expensive to make. Also, we were a bit spoiled because Maxwell had a lower price than the prior gen. So there isn't as much of a price hike (not counting FE cards) as some think. From my post in another thread:

GTX 570 = $349
GTX 670 = $399
GTX 770 = $399
GTX 970 = $329

GTX 580 = $499
GTX 680 = $499
GTX 780 = $649
GTX 980 = $549

So, the 1070 at $379 is right where it's typically been. The 1080 at $649 is a bit of a hike. FE cards are just over priced, IMO.
 
Rvenger said:
[SiSoft] Polaris at almost 1,3GHz - Frequenzy of test samples increasing
Click to expand...

That is ...disappointing
tumblr_inline_mrl679BBpq1qz4rgp.gif
 
  • Like
Reactions: Zuul
like this
Serious? Saying it jokingly? Who knows, but with all the rumored design wins for Polaris, we can be pretty sure nvidia took notice.
 
Is it me or they have technologically deficient people doing this stream? LOL

yeah taking these peoples word for it when they can't even seem to do a proper streaming video.....hmmm
 
If this is true then I will definitely buy Polaris 10. I've been looking for a card that has a pretty low TDP with solid performance for 1080p gaming. Just wish it was HMB though but the price tag is very pleasing for me.
 
if it is indeed running at 1.3 with samples I bet its gonna be one hell of a performer for the price. We shall see..
 
You must log in or register to reply here.
Back
Top