ATI HD 5870 Scaling Performance: X58 vs P55 Showdown
by Gary Key on September 29, 2009 12:00 PM EST- Posted in
- Motherboards
We will start off our P55 coverage this week by answering a question that has been raised numerous times as of late, “Exactly how will the new DX11 cards perform on the Lynnfield/P55 platform compared to the Bloomfield/X58?”. The answer to that question depends on the game engine, settings, and processor choice for the most part. There has been much speculation that the Lynnfield/P55 platform would fail miserably with the next generation cards. That said, the difference in platform performance between the first DX11 capable cards available (ATI HD 5870/5850) is about the same as previous generation cards we tested in the Lynnfield launch article.
This means for single card performance both platforms trade blows for supremacy. However, for those running CF/SLI setups, the X58 continues to be the platform of choice for users wanting the best possible benchmark results. Does that mean the integrated dual x8 PCIe 2.0 logic on Lynnfield is a poor choice compared to the dual x16 PCIe 2.0 sporting X58? Absolutely not based on our initial tests. In fact, it should satisfy most users.
Now for those making an investment into an ultra high-end HD 5870 CrossFireX setup, the Core i7/X58 configuration will simply offer the best possible performance. Of course that performance comes at a cost, particularly power consumption. However, we have a feeling most owners sporting two HD 5870s are not that concerned about Al Gore knocking on their door in the middle of the night.
In the meantime, we have our first performance results comparing the Core i7/860 and P55 against the like priced Core i7/920 and X58 in a variety of games. We are going to state this upfront, this is not a GPU review of the HD 5870. Instead, we decided to pull this information out of the upcoming high-end P55 roundup so it did not get lost in the mix. Our resolution is limited to 1920x1080 that we use in the motherboard test suite. As such, the numbers speak for themselves. So let’s get right to the results today, but first, the test setup.
Test Setup-
For our test results we setup each board as closely as possible in regards to memory timings and sub-timings. The P55 motherboard utilized 8GB of DDR3, while the X58 platform contained 6GB. The P55 and X58 DDR3 timings were set to 7-7-7-20 1T at DDR3-1600 for the i7/920 and i7/860 processors at both stock and overclocked CPU settings. All power management features were enabled on each board and voltages were set at the lowest possible values when overclocking while still retaining 24/7 stability.
The image gallery below contains our Everest memory results with each processor overclocked at similar memory settings along with voltage/uncore/subtiming options. The 860/P55 offers slightly better throughput and overall latency numbers than the 920/X58 when overclocked. At stock, the latency numbers favor Lynnfield with assistance from the turbo mode.
Our game selection today is varied but is missing a couple of titles we wanted to show. We pulled Crysis Warhead as the CrossFireX scaling numbers were under 8% and Need For Speed: Shift does not have a CrossFire profile yet. AMD is currently working on updated CF profiles for the latest games. We also had some corruption problems when running Empire:Total War and Anno 1404 that is under investigation.
All of our gaming performance results are reported in average frame rates per second in the main bar graph with minimum frame rates reported in the text section. The results are sorted by the average frame rates. We are also reporting single card results with the HD 5870 running at PCIe 2.0 x8 speeds on the P55 platform to compare performance to the x16 single card setup. We installed an Intel CT Gigabit network card in the second physical x16 slot in order to force x8 operation.
Facebook
Twitter
RSS
85 Comments
View All Comments
vshin - Wednesday, September 30, 2009 - link
Gulftowns are supposedly -starting- at $1000 with upper-end parts going as high as $1500. Even at $500, that is still too high for the mainstream gamer. So is a $500 video card. If you have this much to spend, then you may as well spend extra for an X58 system.I'm referring to the budget-enthusiast who will want to limit their purchase to <$300 for CPU or video card, not going to use SLI, but plans on overclocking to maximize value. This segment is more interested in running games fast, and less interested in folding projects or encoding video.
The difference between me and TA is that I don't hate the platform I am supposedly "against." If Intel replaced their entire Lynnfield lineup with Gulftown at the same prices, I would be very happy.
silverblue - Wednesday, September 30, 2009 - link
This may play into AMD's hands when they (eventually) launch the Thuban core. It's a drop-in replacement for AM2+/3 thanks to (I presume) a mere BIOS update. If people can get a 6-core CPU for relatively cheap, they won't bother with the far more expensive X58 path, unless Intel decides that it's worth it dropping the price quite a lot.Thuban is due after i9, which is a bit of a worry.
A budget enthusiast setup is a mouthwatering prospect for those of us with bottoms in our pockets (and little jangling around in them).
Gary Key - Wednesday, September 30, 2009 - link
I have added the CPUZ/Everest Screenshots to the gallery for the 920 overclocked at 4.2GHz. I also ensured that B2B settings were the same on both platforms with auto disabled and a setting of 4 enabled. This is it for additional testing. When it comes right down to it, both platforms have their advantages and disadvantages. So choose which one best suits your needs.The short story to the new uncore testing is that it really does not make a real difference in general applications at these speeds. Of course, I have discussed this for more than a year but hey, why not run through it again. :)
I did gain 2/10ths of a second in SuperPi 8M with the higher uncore, the game scores were basically a wash as was a couple of applications although minimum frame rates suffered in FC2 and HAWX.
Uncore at 3407-
Everest Memory
Read - 18006
Write - 15237
Copy - 21306
Latency - 38.6ns
L3 Latency - 3.2ns
Uncore at 3607-
Everest Memory
Read - 17807
Write - 15892
Copy - 20059
Latency - 37.8ns
L3 Latency - 3.1ns
What does that mean to our top three favorite Core i7/X58 game benchmarks?
FC2-
1920x1080 2xAA HQ DX10 Ranch Small
uncore 3407 - 4.2GHz - HD 5870
CF- 137 minimum fps 173.3 average fps
SC- 77 minimum fps 95.3 average fps
uncore 3607 - 4.2GHz - HD 5870
CF- 134 minimum fps 173.9 average fps
SC- 75 minimum fps 95.1 average fps
World in Conflict-
1920x1080 2xAA/16xAF HQ Bench
uncore 3407 - 4.2GHz - HD 5870
CF- 42 minimum fps 103 average fps
SC- 33 minimum fps 58 average fps
uncore 3607 - 4.2GHz - HD 5870
CF- 42 minimum fps 103 average fps
SC- 33 minimum fps 58 average fps
HAWX-
1920x1080 2xAA HQ DX10.1 Bench
uncore 3407 - 4.2GHz - HD 5870
CF- 128 minimum fps 144.5 average fps
SC- 71 minimum fps 82.5 average fps
uncore 3607 - 4.2GHz - HD 5870
CF- 126 minimum fps 145.4 average fps
SC- 71 minimum fps 81.9 average fps
1. One other item that I answered earlier. The NV GTX275/285 cards perform better on Lynnfield than Bloomfield as does the HD 4890 in most cases. We are still investigating the differences with the HD 5870 on Lynnfield and AMD is trying to have an answer for us tomorrow after three days of marathon testing. It could be drivers, it could just be the architectural changes on the card or a conflict with the new PCIe setup on Lynnfield.
ilnot1 - Thursday, October 1, 2009 - link
Gary, your really are saint to run uncore tests (again), but...don't give in too much otherwise the terrorists win!
TimboG - Wednesday, September 30, 2009 - link
Great!
Thanks for clearing that up for me Gary. You have to remember that most readers do read more reviews than what is posted here at AnandTech and in doing do it becomes very difficult to remember so many obscure setting changes that have become available with the new chipsets and each of their overall affect on performance. That was the reason I spoke out against the changes in the default settings. With that in mind, we, (the readers) also noticed the strange behavior of the HD5890 on these platforms compared to previous benchmarking of other graphics cards on these platforms. I, for one, was concerned that the settings changes had produced these strange results. Thank you again for taking the time to remove that from the equation.
GeorgeH - Wednesday, September 30, 2009 - link
Shenanigans!!!! Thought you could pull the wool over our eyes, did you? Maybe you thought we wouldn't notice BOTH uncore speeds you selected for Bloomfield, 3407 and 3607, are PRIME NUMBERS!?!?But oh, look at Lynnfield at 3602 – a speed that is clearly divisible by 2, and most definitely NOT PRIME. This is CLEARLY yet another example of your pro-Lynnfield bias!!!! Just how dumb do you think we are?!?!?
Also, you have the patience of a saint.
Voo - Wednesday, September 30, 2009 - link
@Lifted: You're just kidding right? Oh god please say you're just kidding.@GeorgeH: YMMD - let's await TAs response, I'm really curious with what he'll come up this time.
@Gary: You've really got the patience of a saint.
Lifted - Wednesday, September 30, 2009 - link
I'm starting to think that these trolls work here at AT. I've never seen AT staff respond to anyone so often, especially obvious trolls. Not only are the staff responding to the same troll(s), but they run their tests again and even publish new articles around the trolls comments. The whole situation just doesn't smell right to me. Seems a bit of rattling the cage is going on for obvious reasons.Gary Key - Wednesday, September 30, 2009 - link
"I'm starting to think that these trolls work here at AT. "If you only knew how much I wished that were true. :)
1. TA152H (almost ready for a block based on personal attacks)is actually Rich A. who "was" a freelance writer at Toms Hardware. You can contact our good friends (seriously) Tuan or Chris over at Toms to verify.
2. the Zorro (almost there again) was thezorro (blocked for spamming) was SnakeOil (blocked for useless spamming). SnakeOil has been a fixture over at Tech Report and somehow found his way over here. You can view his comments at TR in the Intel/AMD articles to verify.
3. Our followups are not for the amusement of these commentators (being real nice here). We did receive a lot of requests for the followups and three (Clock for Clock OC / 860 review / P55 scaling) were already in the works. We are just trying to be accommodating here for follow up information when deemed necessary.
iamezza - Friday, October 2, 2009 - link
TA152H wrote for Toms Hardware, wow! This explains a lot.I still don't understand why these guys weren't banned many moons ago though.